Side Finds

XPeng's operating loss shrank to RMB2.77 billion in 2025 — the best figure in company history. Open the income statement in the annual report (20-F), though, and you find a line that did not come from selling cars: "other income, net" of RMB1,761.4 million, which management attributes "primarily due to the increase in government subsidies." Without that line, the operating loss would have exceeded RMB4.5 billion.

Subsidies are everyday business in China's EV industry, and XPeng discloses them cleanly. But they are also the opposite of earned margin: the 20-F itself warns that expiry or reduction of such support can hurt demand and results — and the most important support of all, the NEV purchase-tax exemption of up to RMB30,000 per vehicle, has already been cut in half since January 1, 2026. Remember: before you celebrate the narrowing losses, check how much of the narrowing the state paid for.

Inside XPeng's record year 2025 sits a shift that barely registers in revenue but changes everything in profit: the "services and others" segment — mostly technical R&D services for the E/E architecture collaboration with Volkswagen, plus parts and carbon credits — contributed just RMB8.34 of RMB76.72 billion in revenue (10.9 percent). But at a segment gross margin of 68.2 percent, it delivered roughly RMB5.69 of the RMB14.47 billion in consolidated gross profit — about 39 percent. For comparison: the vehicle business itself managed a 12.8 percent gross margin in 2025.

The annual report (20-F) names the concentration risk unusually directly: XPeng has "a limited track record" in such technology services and relies "primarily on the Volkswagen Group" for these revenues. Translated: a substantial part of the record year's earnings quality hangs on a single partner — one that also owns 4.9 percent of the company and sits deep inside XPeng's technology through the joint architecture development.

The spectacular "XPeng flying cars" from the headlines are, strictly by the annual report (20-F), not an XPeng product at all: they are developed by HT Flying Car Inc. ("Huitian"/AeroHT) — listed in the 20-F's related-party table as "significantly influenced by the Principal Shareholder," that is, by founder and CEO Xiaopeng He. XPeng itself holds only financial investments in Huitian (preferred shares since January 2021, plus debt instruments) and earns from the flying-car venture as a service provider: a cooperation framework agreement — most recently renewed to run from January 1, 2026 through December 31, 2028 — has XPeng supplying R&D services, technology consulting and sales-agent services to Huitian.

For the equity story that is a subtle but important distinction: if the flying car takes off, much of the upside belongs to the founder's company — XPeng shareholders participate only through the investment positions and the service business. The fantasy flies under someone else's registration.

XPeng's humanoid-robot story has a footnote few investors know: the robotics work sits inside Dogotix Inc., a British Virgin Islands entity that, per the annual report (20-F), has been "primarily engaged in research and development of robots with human-robot interaction functions since 2021." On September 29, 2023, XPeng agreed to buy 74.82 percent of Dogotix for about $98.96 million (roughly RMB710 million) — from sellers that explicitly included a wholly-owned company of CEO Xiaopeng He. Since closing in October 2023, Dogotix has been carried as a subsidiary of the group.

The balance sheet has shown a "robotics platform technology" intangible ever since: RMB777.7 million at cost, RMB602.7 million net book value at the end of 2025, amortized over ten years. All of it disclosed, none of it illegal — but anyone buying the stock for the robot story should know: the foundation of that business was sold to the company by its own founder, priced not by a market but by a contract between related parties.

The annual report of jewelry and gold dealer Envela contains an unexpected AI passage: industrial demand for silver is described as driven by battery storage, electric vehicles — and "AI-driven data centers" (10-K 2025, Item 1). The recommerce group, which our AI dossier rates as "neutral", thus benefits from the AI boom through the very metal it recovers from old chains and circuit boards: data centers need silver, and Envela collects it — 3.3 metric tons of precious-metal refining scrap in 2025 alone.

According to its 2025 annual report, Envela prefers to expand into former bank branches: the most recent new stores were opened in purchased or leased ex-bank buildings, because vaults and security infrastructure are already in place and the locations are prime. Buildings once used to store money thus become places where gold changes hands across the counter again — six of the 18 locations, including the headquarters, are owned by the company itself (10-K 2025, Item 1 and Item 2).

MSG Networks is currently the only Sphere Entertainment segment that reliably delivers operating profit ($38.6 million in 2025, $32.1 million in the first quarter of 2026 alone). But as part of the June 27, 2025 debt restructuring, the media rights agreements with the New York Knicks and New York Rangers did not just get cheaper — they were also shortened: they expire after the 2028-29 NBA and NHL seasons, with MSG Networks keeping only a right of first refusal. The agreements with the other teams run off at varying dates over the next six NHL seasons, per the 10-K.

On top sits a double concentration risk on the revenue side: "Substantially all of our affiliation fee revenue comes from our top four Distributors." — and the subscriber count most recently fell about 14.5 percent year over year (Q4 2025). Whoever buys Sphere stock also buys a TV network with a shrinking audience, four dominant customers and core content that has to be renegotiated in 2029 — with teams whose owner family simultaneously controls Sphere Entertainment.

There is a number in Sphere's balance sheet that looks wrong at first glance: the restructured MSG Networks loan had an outstanding principal of $158.9 million as of December 31, 2025 — but per Note 14 of the annual report (10-K) it is carried at $303.7 million. The reason is the U.S. accounting rule for troubled debt restructurings: because the lenders forgave a large part of their claim, expected future interest and potential contingent payments (from so-called Contingent Interest Units) must be baked into the carrying amount — so that the borrower does not book more gain at the time of the restructuring than economically remains.

The side effect: the reported one-time gain of $346.1 million recorded on June 27, 2025 is smaller than the raw haircut ($829.1 million down to $210 million) would suggest — and in return, hardly any interest expense for this loan will run through the income statement going forward, because interest payments reduce the carrying amount instead of hitting earnings. If you read Sphere's interest burden off the P&L, you will systematically underestimate it.

The company's most valuable piece of land costs — no fixed rent. Per the annual report (10-K) for 2025, the Sphere in Las Vegas sits on ground next to The Venetian Resort under a 50-year ground lease. The deal: "The ground lease has no fixed rent; however, if certain return objectives are achieved, The Venetian will receive 25% of the after-tax cash flow in excess of such objectives."

For shareholders that means: as long as the orb struggles, the land is effectively free — an unusually merciful arrangement. But the better the business performs, the larger the neighbor's share of the success becomes. Just as "The Wizard of Oz at Sphere" is making the venue profitable, the question starts ticking when the 25 percent participation kicks in — the filing does not say where those return objectives sit.

The annual report (10-K) for 2025 contains a footnote with involuntary drama: in 2021, ServiceNow granted the CEO and certain executives a performance option package with a grant-date fair value of $232 million (the "2021 Performance Awards"). The eight tranches vest only if both conditions are met: a subscription revenue target and a stock price target, measured on the 180-day and 30-day volume weighted-average price (VWAP). The deadline for everything is September 30, 2026 — whatever is not achieved by then expires.

As of December 31, 2025, per the filing, only four of the eight tranches had vested. The punchline: with revenue growing 21 percent, the revenue conditions are likely the smaller problem — but the price conditions measure 180-day average prices, and the stock sits roughly 60 percent below its high. Management thus has a very personal, very dated interest in a fast recovery — a detail worth keeping in mind when reading the next quarterly releases.

ServiceNow financed the largest acquisition in its history in an unusually sporty way: for the cybersecurity provider Armis, the company paid approximately $7.8 billion in cash on April 20, 2026, per the quarterly report (10-Q) — against liquidity of $7.9 billion beforehand (March 31, 2026). What made it possible was a $4.0 billion term loan signed on April 17, 2026, that matures as soon as October 16, 2026 (extension option: six months), plus a new $3.0 billion revolving credit facility dated April 1, 2026.

A multi-billion six-month loan is bridge financing — it has to be replaced within months by bonds, cash flow or an extension. For a company that was practically debt-free until then ($1.5 billion in notes due 2030), that is a cultural break: ServiceNow bought aggressively in the middle of a crash instead of protecting its cash — and wrote itself a refinancing deadline that still falls in 2026.

If you picture ServiceNow as a company with thousands of corporate customers, one line in the fine print deserves a second look: a single customer — per the quarterly report (10-Q) a "U.S. federal channel partner and systems integrator", the reseller through which U.S. federal agencies buy their ServiceNow contracts — accounted for 12 percent of total revenues in the first quarter of 2026 and 19 percent of the entire accounts receivable balance as of March 31, 2026 (December 31, 2025: 11 percent). In fiscal years 2025 and 2024, 11 percent of company revenue ran through this one channel each year; in 2023, no customer crossed the 10 percent threshold.

The concentration cuts two ways: it hangs on the U.S. federal budget (shutdowns, savings programs, procurement policy) and on a single counterparty whose payment behavior hits the balance sheet directly. ServiceNow itself notes "no historical collection concerns" with this customer — but one address holding a fifth of the receivables is an unusual concentration for a company with more than 8,000 contracted customers.

AI stocks profit from the data-center boom — 10x Genomics is among the companies it hinders. The Risk Factors of the 2025 annual report (10-K) spell out a rarely so bluntly stated side effect: "the rapid expansion of global artificial intelligence infrastructure has precipitated shortages and extended lead times for high-performance computing components, including GPUs and memory."

The filing goes on to explain that suppliers preferentially allocate capacity to the large technology companies with vast order volumes — so 10x could come away empty-handed or pay steep premiums. For a company whose instruments, like the Xenium analyzer, depend on powerful image processing, that is a concrete manufacturing risk. The irony: while the market celebrates AI beneficiaries, 10x sits on the other side of the ledger — as a customer competing for the very same chips.

Buy TXG and you buy Class A stock with one vote per share. Alongside it sits a second class: Class B with ten votes per share. As of January 31, 2026, 117.7 million Class A shares and only 10.1 million Class B shares were outstanding — the Class B stock is a small slice of the capital but carries ten times the voting weight. The annual report (10-K) draws the conclusion: "the holders of our Class B common stock collectively control a majority of the combined voting power of our common stock" — the Class B holders together control a majority of the voting power and thus effectively every shareholder vote.

Behind the Class B stock stand, in essence, the two co-founders — CEO Serge Saxonov and Chief Scientific Officer Ben Hindson. For investors that cuts both ways: the scientific minds stay at the wheel, but a hostile takeover or an externally forced change of course is effectively ruled out. The share price can fall as it likes — control does not move with it.

10x Genomics runs an unusually aggressive patent strategy. The annual report (10-K) states it plainly: "It is our general policy not to out-license our patents but to protect our sole right to own and practice them." In 2025, that policy paid off in hard cash. In February 2025 the company ended its worldwide patent litigation with Vizgen ($26.0 million upfront: $9.2 million as a gain on settlement, $16.8 million as license revenue); in May 2025 the settlement with Bruker followed, totaling $68.0 million in four quarterly installments ($40.7 million gain on settlement, $27.3 million license revenue).

Together the two settlements poured roughly $94 million into the 2025 result — $44.1 million as one-off revenue, $49.9 million as gains on settlement — and, on top of that, cut legal expenses inside selling, general and administrative costs by $25.6 million. Against a 2025 net loss of $43.5 million, that means: without the trip through the courtrooms, 2025 would have stayed deep in the red. At 10x, litigation is not a sideshow — it is an earnings driver.

Applied Digital's cloud unit (GPU computing power for AI, most recently with exactly one customer generating 14 percent of company revenue) had been looking for a buyer since 2025. The buyer it found: Ekso Bionics — a Nasdaq-listed maker of medical exoskeletons. Via a Contribution and Exchange Agreement, the cloud subsidiary was contributed into Ekso; the combined company operates as ChronoScale Corporation. Closing came on May 5, 2026; Applied Digital additionally put about $15.75 million into an Ekso private placement (1,311,407 shares at $12.01 each).

The detour was expensive: because the deal meant the unit no longer qualified as "held for sale," Applied Digital had to move it back into continuing operations and, in the quarter ended February 28, 2026, take a $59.7 million write-down in the process. That a company whose own stock market shell has already housed temp staffing, flight safety and blockchain now places its AI cloud into the shell of an exoskeleton builder is a punchline no screenwriter would dare to pitch.

The quarterly report (10-Q) as of February 28, 2026, contains a related-party construction you have to read twice: Applied Digital guarantees, in favor of The Babcock & Wilcox Company, the performance of a design-build agreement for a gas power plant with roughly 1.2 gigawatts of nameplate capacity — but the counterparty is not Applied Digital itself. It is Base Electron, Inc., and the filing says about it, verbatim: "Base Electron is an independent power producer owned and managed by a combination of third parties, as well as certain officers and directors of the Company acting in their individual capacities" — an independent power producer privately co-owned by certain officers and directors of Applied Digital.

Exiting the guarantee costs money: the termination fee is $50 million (if paid by August 1, 2026) and $100 million thereafter — alternatively the guarantee ends if Base Electron lists on an exchange or raises at least $50 million. At the same time, Applied Digital holds warrants on plant builder Babcock & Wilcox (fair value up $19.2 million in nine months) and a stake in Base Electron itself (up $2.0 million). The conflicts of interest are disclosed, but they are built in: the company carries the guarantee risk — while some of the beneficiaries on the other side sit in its own executive offices.

Pull up the SEC registration history of Applied Digital (CIK 1144879) and you find a corporate biography straight out of a textbook on market fashions: the entity was named Reel Staff Inc. (a staffing firm) from 2001, Flight Safety Technologies Inc. from 2002, Applied Science Products, Inc. from 2011, Applied Blockchain, Inc. from April 2021 — and only since February 2023 Applied Digital Corporation. From crypto mining host to self-styled "AI factory" took barely two more years.

This is not an accusation — today's business of real data centers in North Dakota has nothing to do with the earlier shells, and the name change from "Blockchain" to "Digital" is cleanly documented with the SEC. But it is a useful reminder: the same stock market shell has already carried two hype cycles before AI came along. If you buy the stock for the label, you should know how often the label has been swapped.

If you scanned the ticker list for stock market newcomers, you would file RF Industries in the wrong drawer: per the annual report (10-K), the company was incorporated in Nevada on November 1, 1979, completed its IPO in March 1984 under the name Celltronics, Inc., and has been called RF Industries, Ltd. only since November 1990. As a listed company, the stock has lived through the early cell phone era, the dotcom crash and the 3G, 4G and 5G build-outs — as a permanent resident of the micro-cap segment.

Today the group consists of five acquired subsidiaries (Cables Unlimited, Rel-Tech Electronics, C Enterprises, Schroff Technologies, Microlab/FXR) with a combined 289 full-time employees at five U.S. sites (as of October 31, 2025). That long look back matters for judging the current rally: a four-decade-old telecom supplier that has seen many cycles is not a young growth company — it is a cyclical whose business breathes with the capital budgets of network operators.

Since July 2024, RF Industries and its subsidiary C Enterprises had faced a class action in San Diego County Superior Court: a former employee accused the company of California labor-law violations — from unpaid working time and denied breaks to inaccurate wage statements — and in October 2024 expanded the case with penalty claims on behalf of the state (Private Attorneys General Act, PAGA).

On October 30, 2025 — the second-to-last day of the fiscal year — RF Industries signed a memorandum of understanding: an $855,000 settlement, "all-in and non-reversionary," meaning nothing can flow back to the company; the amount was fully accrued. For scale: that is more than eight times the net income of the entire fiscal year 2025 ($0.1 million). Per the quarterly report, the court hearing on preliminary approval was scheduled for August 7, 2026. This is not an existential risk — but at a company whose latest annual profit was a six-figure number, a single employment lawsuit visibly moves the earnings math.

While the stock gained roughly 250 percent in twelve months (data as of July 18, 2026), Note 1 of the quarterly report (10-Q) as of April 30, 2026, contains a paragraph you would not expect from a celebrated turnaround name: the interim statements are prepared assuming the company will continue as a going concern — and the propriety of that basis depends, "among other things," on future profitable operations, sufficient operating cash flow and the credit facility with Eclipse Business Capital.

This is not a formal going-concern warning with "substantial doubt" — but it is also not boilerplate that every micro cap prints in its quarterly reports. It fits the situation: $3.4 million in cash (October 31, 2025: $5.1 million), $6.1 million drawn on a secured credit line, operating cash flow slightly negative in the first half of fiscal 2026 (minus $47,000). The chart tells the story of a breakout — the footnote is a reminder of how narrow the financial foundation underneath still is.

In fiscal 2025 (ended October 31), RF Industries earned $0.8 million before taxes — and booked $0.7 million of that as income tax expense: an effective tax rate of 91 percent, as the annual report (10-K) dryly discloses. The reason is not a penalty tax but a balance-sheet legacy: because of the loss years 2023 and 2024, the company recorded a valuation allowance against its deferred tax assets — $3.8 million added in 2024 alone, another $0.8 million in 2025.

The punchline sits in the quarterly report (10-Q) as of April 30, 2026: after four consecutive quarters of pre-tax income, RF Industries reviews that allowance every quarter — and writes, verbatim, that releasing a significant portion of it "could result in a material income tax benefit in the period recognized." In everyday terms: during the loss years the company accumulated vouchers at the tax office that it has prudently valued at zero — if the recovery holds, those vouchers go back on the books, and a future quarter would look spectacularly profitable on paper. Whoever only reads the earnings line that day will mistake a bookkeeping entry for operating strength.

Toast does not just sell registers and process payments — through Toast Capital, the company is also in the lending business: per the annual report (10-K), restaurants get "fast and flexible funding" via loans issued by a partner bank that are generally repaid through a portion of their daily transactions. Toast underwrites with its own data models built on POS and payment data — whoever runs the restaurant's register knows its sales better than any credit bureau.

The model is clever, but it carries banking risks into a software company's books: the loan servicing revenue runs inside the "Financial technology solutions" line, and the filing explicitly names the credit risk of its "loan servicing activities." If the restaurant industry — a business with notoriously high mortality — hits a recession, Toast is exposed twice: payment volume falls and loan defaults rise. At a casual glance at the "restaurant software stock," this second business is easy to miss.

At Toast, what counts is not who holds the most capital but who holds the right shares: the unlisted Class B shares carry ten votes each, the listed Class A shares one. As of December 31, 2025, 66 million Class B shares were outstanding — a good tenth of all shares, but, per the annual report (10-K), roughly 55 percent of the voting power. Five-percent holders, directors and executive officers together also beneficially owned about 55 percent of the votes.

The filing spells out what that means: this group keeps control over board elections, mergers and every major decision — "even after they no longer have a service relationship with us." If you buy the Class A share, you buy economic participation with structurally limited say. For index providers and governance raters, precisely this construction has been a point of contention for years.

If you buy a point-of-sale system from Toast, you get it well below what it costs the company: per the income statement in the annual report (10-K), the "Hardware and professional services" line took in $180 million in 2025 — and caused $400 million in direct costs. A gross loss of $220 million, after $170 million in 2024 and $181 million in 2023. That is not an accident but a sales model: through the in-house "Easy Pay" lease, restaurants can minimize upfront costs and pay for the devices through a portion of their daily card transactions.

The logic behind it: every subsidized terminal ties a location to the software subscriptions and, above all, to payment processing — $5,037 million of the $6,153 million in 2025 revenue ran through that rail. The terminal is the bait, the card is the hook. If you compare Toast's gross margin (about 26 percent) with classic software companies, you should know about this built-in loss line.

Okta runs two share classes: the listed Class A carries one vote per share, the unlisted Class B ten votes per share. As of April 30, 2026, 167.7 million Class A shares stood against just 7.7 million Class B shares — a good 4 percent of the capital. The voting leverage turns that sliver into real power: the annual report puts the combined voting power of directors, executive officers and their affiliates at 32 percent (as of January 31, 2026).

For shareholders this means that mergers, charter amendments and board elections effectively run through the founder side around CEO Todd McKinnon — the annual report itself lists the structure as a risk factor and warns the concentration could deter takeovers and depress the share price. Whoever buys Okta buys a say with a built-in damper.

In the loss years, Okta's income statement carried a line you would not expect at a software company: gains from repurchasing its own debt. When the share and bond prices collapsed after 2021, Okta bought back its convertible notes below face value: in fiscal 2024 a total of $1,050 million in principal for $937 million in cash (gain: $106 million), in fiscal 2025 another $300 million in principal for $280 million (gain: $19 million) — together $125 million in income simply because its own IOUs were on clearance sale.

The punchline: fiscal 2025 produced Okta's first net profit, $28 million — but operations still lost $74 million that year. Only interest income ($106 million) and the notes-repurchase gain ($19 million) lifted the bottom line above zero. The chapter is nearly closed now: the 2025 notes were settled at maturity on September 1, 2025 with a final $510 million in cash, and only $350 million of the 2026 notes remained outstanding as of January 31, 2026 (due June 15, 2026).

Open Okta's balance sheet and the largest line item is not a data center and not software — it is goodwill: $5,487 million of $9,710 million in total assets (January 31, 2026), roughly 57 percent. Almost all of it comes from a single deal: in May 2021, at the height of the software boom, Okta acquired identity specialist Auth0 for approximately $5,671 million — paid almost entirely in its own stock (19.2 million shares valued at $5,175.6 million, near the then record price). Per the annual report, $5,290.1 million of that landed on the books as goodwill.

The curious part: although Okta shares at times lost more than 80 percent after 2021, this hope value was never written down — the annual impairment tests kept concluding that Auth0 delivers what was promised. For shareholders that means two things: the $6,999 million in equity consists mostly of this acquisition trust — and should Auth0's customer business ever seriously disappoint, the balance sheet would show it late, and then all at once.

The annual report (10-K) for 2025 contains a number that looks like a boom: backlog jumped from $33.2 million to $159.9 million within a year — almost five times over. The main driver, however, is not a sudden ordering frenzy but a change in invoicing: large multi-year contracts are, per the filing, increasingly billed annually in installments instead of fully up front. Contract portions not yet invoiced do not land in deferred revenue but in backlog — so the metric grows even when only the billing rhythm changes.

The same effect distorts what used to be the company's favorite yardstick, in the other direction: in the same report, Tenable explains it has retired calculated current billings as its steering metric — verbatim: "we have transitioned away from relying on calculated current billings to monitor performance of our business." Remember: when a software subscription company posts a spectacular jump (or slump) in backlog or billings, check first whether the business changed — or merely the cadence of the invoices.

If you think of Tenable as a purely Maryland company, you are missing a geographic concentration the company itself flags as a risk: per the annual report (10-K) for 2025, its research and development teams maintain "a significant presence in Tel Aviv, Israel" — and the filing warns verbatim: "Recent and ongoing hostilities in the region may have a material impact on our ability to deliver on our product roadmaps for these solutions."

That concerns precisely the growth franchises: for the companies acquired in recent years — among them Ermetic (cloud security, 2023, $243.3 million), Vulcan Cyber (risk management, February 2025, $148.5 million) and Apex Security (AI security, June 2025, $47.8 million) — the risk factors state verbatim that "companies we have recently acquired principally operate in Israel". The deep-dive analysis does not cover this cluster — which is why it sits here: whoever buys the turnaround thesis also buys a product pipeline whose engineering teams sit in a conflict region.

While Tenable's stock was searching for its bottom in early 2026, the company bought back its own shares more aggressively than ever: 6.1 million shares for roughly $130 million in the first quarter of 2026 alone, at monthly average prices of $22.71 (January), $20.73 (February) and $20.38 (March). The board had topped up the buyback authorization by another $150 million — to $700 million in total — in January 2026, in the middle of the slide. Since the program started in November 2023, $492.4 million had gone into 16.7 million of the company's own shares through March 31, 2026.

The balance-sheet context is what makes this remarkable: as of March 31, 2026, $360.3 million in cash and short-term investments stood against a term loan balloon payment of $350.6 million due July 7, 2028 — and equity has shrunk to $248.2 million through the buybacks (December 31, 2025: $326.4 million). As long as operating cash flow keeps delivering around $88 million per quarter, the math works. If it breaks, Tenable will have spent its reserve at prices the market had just judged too high.

In the final quarter of 2025 — months before the Fox takeover was signed on June 14, 2026 — four Roku executives adopted Rule 10b5-1 trading plans to sell shares, per the annual report (10-K, Item 9B): CEO Anthony Wood (through the Wood Revocable Trust, up to 384,000 Class A shares, plan adopted November 19, 2025, running through September 9, 2026), Roku Media president Charles Collier (642,753 shares), general counsel Christopher Handman (47,593), and devices chief Mustafa Ozgen (39,471).

Such plans are legal and common — they automate sales on pre-set terms precisely to avoid insider-trading accusations. The pattern is still worth noting: our in-house turnaround scanner counts 20 insider sales and not a single purchase for Roku in the twelve months through July 18, 2026 — the only one of the eight checklist signals the stock misses. While the market celebrated the turnaround and Fox negotiated, management stood on the sell side of its own stock.

The fine print of the merger agreement contains a remarkable asymmetry: if either side walks away — say, to accept a superior proposal — a mutual termination fee of $866,084,000 comes due. But if the deal fails on antitrust or investment-screening grounds — a final injunction, or missing regulatory approvals by the deadline — Fox owes Roku a reverse termination fee of $1,237,262,000.

And one more clause for connoisseurs: if it is the Fox shareholders of all people who vote down the required share issuance, Fox reimburses Roku's transaction expenses up to $70 million. The deadlines named in the 8-K: June 14, 2027, extendable to December 14, 2027, and at the outside March 14, 2028. Holding Roku stock therefore also means holding a regulatory lottery ticket: in the failure scenario, Roku would stand alone again — but with a consolation prize of over $1.2 billion added to an already full treasury.

In 2025, Roku reported its first annual net profit since the 2017 IPO: $88.4 million. Open the income statement in the annual report (10-K) and you find the punchline underneath: the operating result was still negative at −$5.6 million. The profit came from the line below — $101.4 million of other income, essentially interest on a cash pile of about $2.3 billion.

In plain terms: the actual business did not make money in 2025 — the savings account did. Only the first quarter of 2026 swung the operating line clearly positive (+$51.8 million). For turnaround hunters that is not a detail but the difference between "the turnaround is done" and "the turnaround is under way": a black zero made of interest income survives any recession — an operating turnaround still has to prove it.

If you plan to vote on the Fox takeover as a Roku shareholder, you should know one sentence from the annual report (10-K) for 2025: founder and CEO Anthony Wood controls a majority of the combined voting power even though he owns just 11.6 percent of the outstanding shares. The dual-class structure makes it possible: every Class B share carries 10 votes, every publicly traded Class A share one.

For the Fox deal, that means the vote is effectively over before it begins: per the 8-K filed June 15, 2026, Wood and allied shareholders — together holding roughly 55 percent of the voting power — have already committed by contract to approve the merger in a voting and support agreement. What is required is a simple majority of both share classes voting together. Class A holders may still cast their ballots — they just will not decide anything.

Between May 12 and June 11, 2026, Treace founder and CEO John T. Treace reported nine purchases totaling roughly 371,600 shares via insider filings (Form 4), at prices between $2.17 and $4.04 — about $1.1 million out of his own pocket, even though per the proxy statement (DEF 14A) he already holds 18.7 percent of the company.

The buying window sat directly after the stock's crash to near penny-stock territory and after the preliminary first-quarter figures with positive operating cash flow — a rare degree of skin in the game at a company whose quarterly revenue has shrunk twice in a row.

An unusual footnote hides in the quarterly report (10-Q) as of March 31, 2026: Treace Medical agreed with its primary legal counsel to defer up to $5 million of the legal costs arising in 2025/26 in the patent dispute with Stryker — bearing interest at 10 percent per year, repayable in twelve monthly installments starting January 2027.

The side clause is remarkable: if the share of costs paid on an ongoing basis misses certain thresholds, the firm may declare deferred amounts immediately due. The law firm has thus effectively become litigation financier and creditor of its own client at the same time — on terms you otherwise know from credit cards.

A fabless chip designer like MaxLinear owns no factories — its capital sits in heads and patents. How expensive that is shows in a remarkable number from the annual report (10-K) 2025: MaxLinear spent roughly $208.6 million on research and development — against total revenue of only $467.6 million. That is roughly 45 percent of revenue flowing into the development of new chips alone.

For scale: this R&D line by itself is five times as large as the entire revenue of the industrial & multi-market segment ($37.1 million in the same year). Whoever invests this much in the future can make big leaps when revenue picks up — but burns money just as fast when it does not. Exactly this cost-leverage mechanic explains why MaxLinear reported a $245 million net loss in the 2024 downturn although the operating business had not collapsed.

Whoever looks at the quintupled MaxLinear stock today hardly suspects that three years ago the company almost bought a group larger than itself. In 2022 MaxLinear agreed to acquire the Taiwanese memory-chip maker Silicon Motion for roughly $3.8 billion. In July 2023 MaxLinear pulled the ripcord, terminated the merger agreement and declared itself no longer obligated to close. Silicon Motion called that a breach of contract and turned to the arbitration court in Singapore.

The punchline sits in the balance sheet — more precisely, in what is missing from it. Silicon Motion demands the termination fee (roughly $160 million) plus damages "in excess of the termination fee". Yet to this day MaxLinear has reserved not a single cent for it: management does not consider an unfavorable outcome "probable", and an estimate of the amount is said to be impossible. With cash of just over $61 million, the company itself concedes its means for a damages event "may not be sufficient". A risk in the hundreds of millions that appears in no balance-sheet line: exactly the kind of thing you miss when you only watch the rising price.

Per the annual report (10-K) for fiscal year 2026, roughly 90 percent of Credo's revenue came from its ten largest customers — the three largest end customers alone accounted for 84 percent (33 plus 32 plus 19 percent); the year before, as much as 67 percent hung on a single contracting party. On top of that, all semiconductor wafers are manufactured exclusively by TSMC. As early as the beginning of 2023, the then-largest customer cut its demand forecasts — and fiscal year 2024 grew by only 4.8 percent.

Hyliion reports $3.475 million in revenue for 2025 — and every dollar of it comes from R&D services for the U.S. government, mostly under contracts with the U.S. Navy's Office of Naval Research. The quarterly report as of March 31, 2026, spells out the dependence: up to $11.2 million of potential revenue remains under the current contracts — followed by this sentence: "These contracts can be cancelled by the United States government at any time for, among other reasons, convenience." In plain English: the only paying customer may walk away at any moment — for convenience, among other reasons.

A footnote in the annual report shows how concentrated the business is: of the customer receivables outstanding as of December 31, 2025 ($0.3 million), "the majority" came from a single customer. For context: such termination clauses are standard in U.S. government contracts and not a Hyliion peculiarity. But for a company whose commercial product is only slated to launch by the end of 2026, the Navy's order book is the sole revenue bridge — and that bridge has a built-in trapdoor.

Share buybacks are what you expect from companies that earn more than they can invest — not from companies that have never sold a product. Hyliion did it anyway: in December 2023, in the middle of winding down its electric-truck business, the board authorized a $20 million repurchase program. By the end of 2025, 10,610,070 shares had moved into treasury for $14.1 million — an average price of about $1.33 per share. $6.1 million of the authorization remains open; the program is currently paused per the annual report.

The punch line: measured against the summer 2026 price level (as of July 18, 2026: around $4.90), buying back at the lows was, in hindsight, a remarkably good trade — the repurchased shares would be worth a multiple today. The flip side stands regardless: a pre-revenue company funding its development from a finite cash pile handed $14 million back to shareholders — money it now, per its own projection in the annual report, partly plans to raise again through roughly $10 million in equipment-backed financing.

If you think of Hyliion as a Texas garage start-up, you are missing the industrial giant in the background: the KARNO generator, on which the company's entire future rests, per the annual report "emerged out of General Electric's long-running R&D investments" in aerospace and metal additive manufacturing, and was acquired from GE's aviation business via an Asset Purchase Agreement in August 2022. The connection stayed close after the deal: every single additive printer Hyliion uses to produce the generator's components comes, per the annual report, from Colibrium Additive — the former GE Additive. Chief technology officer Joshua Mook spent 18 years at General Electric before joining Hyliion.

And GE is not just the intellectual parent but a co-owner: per the fundamental data profile, GE Aerospace held exactly 5,500,000 Hyliion shares (about 3.1 percent) as of March 31, 2026. For investors that cuts both ways: the technology has a more serious pedigree than is usual for an ex-SPAC — but Hyliion depends on a single supplier from that same GE universe for its most important production tool, which the annual report itself flags as a concentration risk.

The company behind the PENG ticker has changed identity several times in short order: until October 3, 2024, it was registered with the SEC as SMART Global Holdings, Inc. — a memory-module company that went public in 2017. The renaming to Penguin Solutions moved the AI computing brand into the corporate name. And on June 30, 2025, came the second cut: the redomiciliation from the Cayman Islands to Delaware, executed via a court-sanctioned scheme of arrangement under Cayman Islands law (Form 8-K12B). The "U.S. company" with the AI story was, formally, a Cayman entity until mid-2025.

There is also a ghost of the past in the numbers: in 2023 the company divested SMART Brazil — fiscal 2023 therefore ended with a $187.5 million loss ($195.4 million of it from the discontinued operation), and a deferred purchase-price installment of $24.3 million was still being collected in May 2025. If you hear the smooth "AI company all along" narrative, know this: today's Penguin Solutions is the product of a multi-year rebuild with divestitures, write-offs and two changes of name and legal home.

Look closely at the annual report (10-K) for fiscal 2025 and you find a quiet funeral inside the celebrated AI computing segment: the Penguin Edge product line (a legacy embedded-computing business) is being wound down — the report speaks of "winding down the manufacturing and discontinuing the sale of products" — and its goodwill has been written off in full: a $16.1 million goodwill impairment in fiscal 2025, in the company's own words "the full impairment of goodwill associated with our Penguin Edge business under our Advanced Computing segment."

That explains part of the weak segment numbers: Advanced Computing lost more than 20 percent of its revenue in the first nine months of fiscal 2026, partly because Penguin Edge revenue is falling away. For the AI story it means this: the segment that gives the stock its name and its imagination is currently burying a legacy — while the company's record numbers are being carried by the memory business.

South Korean carrier SK Telecom plays a triple role at Penguin Solutions that you rarely see spelled out this clearly: it is, first, a major shareholder — through its purpose-built vehicle Astra AI Infra it holds, per the quarterly report (10-Q) as of May 29, 2026, more than 10 percent of the company's voting interest, acquired via $200 million of convertible preferred stock (closed December 13, 2024). Second, it is represented on the board: Min Yong Ha, an SKT executive, is a member of Penguin Solutions' Board of Directors. And third, it is a customer: since May 2025, Penguin has been delivering solutions for SKT's AI data center initiatives — recognizing $33.9 million of revenue from it in the first nine months of fiscal 2026.

None of this is hidden — the filing discloses everything, and the transactions run through the audit committee. But it means that part of the growth fueling the AI story comes from a buyer who is also a co-owner with a board seat, and whose preferred shares earn a 6 percent cumulative dividend before common stockholders see anything. If you want to judge the quality of this revenue, you should know about the entanglement.

On January 7, 2026 — between the order-book turn and the price rally — Vishay's audit committee dismissed Ernst & Young (EY) as independent auditor and engaged Deloitte for fiscal year 2026. The current report (8-K, Item 4.01) stresses the usual: the outcome of a "competitive process," no disagreements with EY, no qualified opinions in prior years. EY finished auditing the 2025 accounts and remained in office through the issuance of its report.

An auditor rotation via tender is good governance practice and no alarm signal by itself. What is noteworthy is the timing: in precisely the year in which Vishay shoulders record capital spending, executes a large equity offering and sees its convertible notes become convertible, a new audit team sits over the books for the first time in decades — Deloitte's first look arrives with the annual report for 2026.

Whoever buys Vishay becomes a junior partner of a family: besides the regular common stock there is a Class B share carrying ten votes each (12.1 million Class B shares next to 123.7 million common shares as of February 11, 2026). The annual report (10-K) for 2025 states that Executive Chairman Marc Zandman — son of founder Dr. Felix Zandman — together with Ruta Zandman and Ziv Shoshani controls about 35 percent of the total voting power; the Class B holders can effectively decide substantially all shareholder votes, from board elections to change-of-control questions.

Add a classified board (directors removable only for cause) and the ability to issue preferred stock — the 10-K itself lists these as anti-takeover provisions. For long-term investors this is not automatically bad (the family thinks in decades, including 55 years of operations in Israel). But an activist pushing for quick value creation bites on granite here — and a takeover premium of the kind semiconductor shareholders enjoy elsewhere is effectively off the table against the family's will.

A side effect of the rally that hardly any momentum buyer has on the radar: Vishay's 2.25% convertible senior notes due 2030, $750 million in principal, carry a clause that makes them convertible once the stock trades sustainably above 130 percent of the conversion price. The effective conversion price is $30.16 per the quarterly report (10-Q), the threshold $39.21 — and exactly that happened: on July 6, 2026, Vishay notified holders via a current report (8-K) that the notes are convertible at the option of the holders during the calendar quarter ending October 3, 2026.

The mitigation sits in the fine print: Vishay must settle the principal in cash; only the value above par can be settled in shares, and the company hedged part of the dilution with capped call transactions. Still: the higher the stock climbs above $30.16, the larger the noteholders' claim in shares or cash — the rally has a built-in counterforce that nobody could see at $15.96 (the June 28, 2025 price).

You cannot fault Vishay's timing: a year earlier, on June 28, 2025, the stock stood at $15.96 per the cover page of the annual report (10-K). On June 29, 2026 — after tripling within three months — the company signed an underwriting agreement with J.P. Morgan for 15 million new shares at $50.00; the underwriters exercised the option for another 2.25 million shares in full one day later. Net proceeds: about $830.3 million — for "growth initiatives" and to pay down the credit facility, as the current report (8-K) puts it.

For existing shareholders that means the share count jumped by roughly 13 percent in one stroke (from about 135.8 million to about 153 million shares). Legitimate, even smart — a company coming off two loss years can hardly finance itself more cheaply. But it is also a quiet statement of the management's own view of the price: whoever sells no stock at $15.96 and collects $830 million at $50.00 evidently considers the higher price a good level to sell at.

A company founded in 1889 carries burdens no scanner metric will ever show: decades of pension promises. In fiscal 2024, Carpenter Technology drew a line and executed a buy-out annuity for its largest defined benefit plan — the obligations were transferred to an insurance company. The price of the farewell: a noncash settlement charge of $51.9 million that weighed on FY 2024 results.

The effect is visible now: net pension expense fell from $76.0 million (FY 2024) to $24.8 million (FY 2025), and for FY 2026 the company expects just $14.3 million, per the quarterly report. Part of the recent earnings jump is therefore not an operating miracle but the disappearance of a legacy cost — good for the risk profile, but a one-off effect you should strip out of the growth curve before extrapolating it into the future.

Roughly 43 percent of revenue in the first nine months of fiscal 2026 came from firm price sales arrangements, per the quarterly report (10-Q): the customer locks in price and volume, and Carpenter locks in the required raw materials — nickel, cobalt, titanium — through commodity forward contracts. The footnote has teeth: if a customer misses the agreed volumes or deviates from the consumption schedule, Carpenter may have to absorb the gains or losses on those forward contracts on a temporary basis.

Add LIFO inventory accounting and a built-in time lag: the raw-material surcharges Carpenter uses to pass nickel price swings through to customers are generally calculated from the previous month's published prices — so there is a systematic gap between surcharge revenue and the actual costs hitting cost of sales. In calm commodity markets, all of this is invisible. In wild ones — and nickel had its legendary moment on the London Metal Exchange in 2022 — quarterly margins can be distorted in either direction without anything changing in the underlying business.

Carpenter Technology's annual report contains a sentence that sounds like nobility: "We have paid quarterly cash dividends on our common stock since 1906." — the company has paid a cash dividend every quarter for 120 years, through two world wars and every crisis since. What the sentence omits sits right next to it: the quarterly rate has been an unchanged $0.20 per share across fiscal years 2023, 2024 and 2025 — while earnings per share jumped sixfold from $1.14 to $7.42 over the same span and the stock price multiplied. The result is a dividend yield of roughly 0.13 percent (data as of July 18, 2026) — one of the lowest we have ever come across.

The stated purpose of the $400 million buyback program is remarkably honest, too: "The primary use of this program is to offset dilution." — per the 10-K, its main job is not returning capital but offsetting the dilution from share-based compensation. If you are looking for a payout story here, you will find the opposite: a company that keeps essentially all of its free cash flow — currently, above all, for the melt-capacity expansion in Athens, Alabama.

Applied Materials has an open repurchase authorization of $13.2 billion (as of April 26, 2026) — and is barely using it: in the second quarter of fiscal 2026 the company bought back $400 million of its own shares, after $1,685 million in the same quarter a year earlier. For the full first half it was $737 million versus $2,999 million — a cut of roughly three quarters, precisely in the period in which the stock ran to record highs (up about 245 percent in twelve months, data as of July 18, 2026).

The filing gives no official reason. The price sensitivity is striking nonetheless: in fiscal 2025 — at much lower prices — the company still repurchased $4.9 billion of stock. You can read that as a quiet valuation verdict from management: the same team that hoovered up its own shares for years apparently no longer considers them a bargain at current levels. At the same time the dividend was raised 15 percent to $0.53 per quarter (March 2026) — returning capital, yes, but preferably on a schedule rather than at the top tick.

If you are cheering Applied Materials' 45 percent profit jump in the first half of fiscal 2026, read Note 3 of the quarterly report first: the company holds publicly traded equities with a cost basis of $965 million that were worth $2.248 billion as of April 26, 2026. Under U.S. accounting rules, the price swings of such positions run directly through the income statement — in the first half, that meant $1.157 billion of unrealized gains, more than a fifth of pre-tax income. The report does not say which stocks they are.

For perspective: operating income rose by exactly $10 million (up 0.2 percent) over the same half-year. Without the portfolio and without the tax rate falling from 25.2 to 13.0 percent, almost nothing of the profit jump would remain. This is not an accounting trick — the rules require it. But it means a sizable part of the reported record profit is stock-market luck on paper, and the same lever works in reverse in the next correction.

The quarterly report (10-Q) as of April 26, 2026, contains a footnote with real teeth: on February 11, 2026, Applied Materials settled with the U.S. Commerce Department's Bureau of Industry and Security (BIS) and paid $253 million to resolve an inquiry into "certain China customer shipments and export controls compliance" — shipments to China customers that allegedly violated export controls. The amount was paid in full during the second quarter of fiscal 2026 and dented the half-year margin of the core Semiconductor Systems segment.

More remarkable than the sum is the side agreement: the settlement includes a denial order that is merely suspended and will only be waived three years after issuance — provided Applied Materials completes internal audits of its export controls compliance program, plus training and reporting duties, on time. A denial order is the sharpest sword of U.S. export enforcement: it can simply prohibit a company from exporting certain products. For a company that generated about 30 percent of its fiscal 2025 revenue in China, that sword hangs directly over the top line — for three years.

Powell has paid a quarterly dividend for decades — and raises it with an almost comical stubbornness by one single cent per year (on the annual total, pre-split): $1.0400 per share in fiscal 2022, $1.0475 in fiscal 2023, $1.0575 in fiscal 2024, $1.0675 in fiscal 2025 — the consolidated statements of operations in the annual reports (10-K) line them up one below the other. After the 3-for-1 split, the board declared a quarterly dividend of $0.09 per new share on May 5, 2026.

What was once a solid income stock has thus become a footnote: at a price near $309 (data as of July 18, 2026), the annual rate of $0.36 yields about 0.1 percent — and the payout ratio has shrunk to roughly 7 percent of fiscal 2025 earnings. Powell would have room for far more ($544.9 million in cash and short-term investments, no drawn bank borrowings as of March 31, 2026) but prefers to park the money for capacity expansion and acquisitions such as Remsdaq. If you are buying this stock for the dividend, you have mistaken it for another one.

While the order book sets records, the insider register points the other way: per fundamental data as of July 18, 2026, zero reported insider purchases stood against eighteen sales. The SEC register matches: between mid-May and mid-July 2026 alone, ten insider filings (Form 4) and seven notices of proposed sales (Form 144) for Powell landed with the securities regulator. Among the sellers is CEO Brett A. Cope himself, who per a Form 4 dated July 10, 2026, sold 4,440 shares at $241.55 each on July 9 (holding 517,233 shares afterwards).

For perspective: insider sales on their own are no alarm bell — executives diversify, cover taxes on vesting equity or follow pre-set trading plans, and Cope remains the largest management shareholder with a good half-million shares. But the one-sidedness is a data point: after a multi-year price surge and a 3-for-1 split, not a single insider chose to buy at these prices for months. Whoever pays 60 times earnings for the stock is buying what the management is currently handing over.

On July 1, 2026, the compensation committee of Powell Industries approved a one-time special award of 36,000 restricted stock units (RSUs) for Brett A. Cope — President, CEO and, at the same time, Chairman of the Board. The purpose is stated openly in the 8-K filing: the award "is intended to incentivize Mr. Cope's continued service to the Company beyond the date Mr. Cope reaches age 60" — the date on or after which he becomes eligible to retire with immediate vesting of all outstanding equity awards. The grant is therefore deliberately backloaded: 25 percent vests in July 2027 and July 2028 each, and the remaining half only in July 2029 — and whoever leaves early forfeits the unvested portion.

The scale is remarkable: at the July 18, 2026 data cut-off, 36,000 shares carried a market value in the tens of millions of dollars — a very visible single package for a company with 3,143 full-time employees and roughly $95 million in annual SG&A. The episode shows two things: the board considers the CEO hard to replace (key-person risk), and the succession question has been postponed with money, not answered. Eight days after the approval, by the way, Cope reported the sale of 4,440 shares at $241.55 via an insider filing (Form 4).

Deep in the tax section of the 2025 annual report (10-K) sits an arrangement hardly any investor has on the radar: Teradyne enjoys a "tax holiday" in Singapore, granted by the Singapore Economic Development Board and tied to conditions on headcount and spending in the country. The effect is measurable: $21.6 million in tax savings in 2025, equivalent to $0.14 per diluted share — roughly 4 percent of the year's earnings of $3.47 per share.

The old agreement expired on December 31, 2025; in December 2025, Teradyne extended it on substantially similar terms — through December 31, 2035. The report itself warns that these savings "may not be achievable in subsequent years," for instance through changes in Singapore's tax laws or new global minimum tax rules. A small, quiet earnings building block with an expiration date and residual political risk — exactly the kind of footnote this section exists for.

Teradyne's handling of its own shares is a small masterclass in counter-cyclical capital allocation. In 2025 — when the second quarter brought shrinking revenue, a more-than-halved profit and a correspondingly cheap stock — the company repurchased 6.3 million of its own shares for $702.1 million, at an average price of $112.21, per the annual report (10-K). The buying was partly funded with $200.0 million drawn from the revolving credit facility — an unusual step for Teradyne, which otherwise operates virtually debt-free.

Then the AI cycle turned, the share price multiplied — and Teradyne all but stopped buying: in the first quarter of 2026, only $5.5 million went into repurchases, at an average of $229.00 per the quarterly report (10-Q); the $200 million revolver was repaid in full. Buy cheap on credit, stop when it gets expensive: that is exactly the behavior value textbooks preach and one rarely sees in practice — most companies buy more the higher their stock climbs.

In May 2024, Teradyne paid $524.1 million (483.1 million euros) for 10 percent of the Italian probe-card specialist Technoprobe — the company whose hair-fine contact needles connect tester and chip during wafer test. The stake is carried under the equity method: as of December 31, 2025, it stood at $537.1 million on the books. But the annual report (10-K) also discloses the market value of the same package: $935.7 million — up from just $389.5 million a year earlier.

Between book value and market value sits a hidden reserve of roughly $400 million that never shows up in Teradyne's income statement — which in 2025 actually recorded a $19.9 million loss from the stake, because Technoprobe's running result plus purchase-price amortization is passed through. If you read Teradyne's balance sheet, you should know both sides: the stake optically depresses profit — and is at the same time worth almost twice what the books show (as of December 31, 2025).

Since August 2025, Backblaze has had a board-approved share repurchase program of up to $10 million (running through August 1, 2026). The stated purpose in the annual report (10-K) is unusually candid: the program is intended to offset dilution resulting from stock-based compensation — funded, of all things, from the proceeds that come in when employees exercise their stock options and contribute through the employee stock purchase plan. By the end of 2025, 256,549 shares worth about $2.0 million sat in treasury.

The orders of magnitude do not match, though: stock-based compensation cost about $26.4 million in 2025 alone — more than two and a half times the entire repurchase authorization — and the weighted share count rose from 36.0 to 56.2 million within two years. The buyback is therefore less a return of capital than a drop against the dilution bill: with one hand the company buys back in small size what it hands out in large size with the other.

In November 2025, Backblaze launched a "2025 Restructuring and Transformation Plan" — and among the roughly $2.5 million in restructuring charges for the year, the annual report (10-K) explicitly lists, next to severance costs, an impairment charge related to the exit from its own corporate headquarters facility. What remains is a lease for about 24,000 square feet of office space in San Mateo (running through 2029), of which roughly 12,000 square feet are already subleased and the remaining 12,000 are available for sublease — the workforce of about 320 employees is predominantly remote anyway.

The punchline is delicious: a company whose business model is that other firms stop running their own infrastructure now applies that principle rigorously to itself — it leases data centers in California, Arizona, Virginia, Amsterdam and Toronto, owns no real estate, and its own office has become a sublease listing. For context: there were restructuring charges in 2023 ($3.6 million) and 2024 ($4.9 million) as well — at Backblaze, the "one-time item" is more of a regular guest.

The $335 million contract that carried Backblaze into the momentum scanners has a flip side spelled out in the 8-K filed June 23, 2026: alongside the Master Strategic Agreement, Backblaze issued its customer CoreWeave two warrants for a combined total of up to 4,194,876 shares — an "Initial Warrant" for 3,053,314 shares that vests in twenty equal quarterly installments of 5 percent over five years as long as the contract remains in effect, and an "Additional Warrant" for 1,141,562 shares whose tranches are tied to contracted storage capacity. The exercise price is $7.60 per share, derived from a volume-weighted average price formula — with expiration dates in 2032 and 2035, respectively.

Measured against the roughly 60.0 million shares outstanding (as of April 28, 2026), that is potential dilution of about 7 percent — and a remarkable role reversal: the flagship customer is now also a shareholder with a fixed-price entry. For CoreWeave it is a built-in rebate paid in equity; for existing holders it means part of the celebrated contract value is being handed back through new shares. A registration rights agreement additionally obliges Backblaze to file a resale registration statement for the warrant shares within 60 days of issuance.

As it became clear that exclusivity for the pain drug Nucynta was running out, Collegium chose a strategy well known in the industry but baffling to outsiders: it struck a deal with Hikma for authorized generics — Hikma has been selling generics of Collegium's own brand since February 25, 2026 (Nucynta IR) and March 11, 2026 (Nucynta ER), and under the supply agreement Collegium supplies Hikma's entire requirements. The company therefore manufactures the competing product to its own brand itself and collects at both ends: in the first quarter of 2026, "Nucynta ER AG" ($1.4 million) and "Nucynta IR AG" ($1.3 million) appear as revenue lines for the first time.

The punchline followed on February 2, 2026: Grünenthal — the German licensor from which Collegium holds the Nucynta rights — sued Collegium and Hikma for patent infringement in federal district court in New Jersey. The planned generic launch infringes two Grünenthal patents on Nucynta ER, the complaint says; Hikma is demanding indemnification from Collegium. Collegium counters that it holds "all necessary rights" for the authorized generics. It is a remarkable constellation: licensor versus licensee while both earn money on the same drug — and an object lesson in how contested the final months before a patent expiry really are.

Collegium's new credit facility of December 2025 — a $580 million term loan, a $300 million delayed-draw tranche (drawn in May 2026 for the Azstarys acquisition) and a $100 million revolver — officially runs until December 23, 2030. But deep in Note 14 of the annual report sits a springing maturity clause: if more than $50 million of the 2.875 percent convertible notes due 2029 is still outstanding on November 18, 2028 and liquidity is below $350 million (minus any note repurchases), the maturity of the entire facility springs forward to November 18, 2028 — roughly two years earlier.

Why that matters: the $241.5 million convertible only converts comfortably into shares if the stock trades well above the conversion price of roughly $36.56. Otherwise Collegium has to repay it in cash in 2029 — and for precisely that scenario the banks have secured the right of way. Translated: if the share price stays below the conversion threshold for too long, two maturities that sit three years apart on paper move together. There is nothing forbidden or unusual about such a clause — but anyone judging this company's debt load should run the calendar by this footnote, not by the cover page.

When Collegium acquired Jornay PM maker Ironshore in September 2024, the fair value of the consideration came to roughly $306 million. Since May 2025 that acquisition has carried a remarkable price tag of its own: David Lickrish, as legal assignee of the liquidated North Sound Pharmaceuticals (NSP), has initiated arbitration against the Ironshore subsidiary IPD — with compensatory damages the annual report (10-K) for 2025 states as "in excess of $500,000". The figure is stated in thousands of U.S. dollars: what is being demanded is therefore more than $500 million — more than one and a half times the entire Ironshore purchase price.

The allegation: before the Collegium acquisition, IPD is said to have violated a license and assignment agreement with NSP and forced the company into liquidation. Collegium rejects the claims and says it intends to defend itself "vigorously"; the company explicitly offers no assessment of the outcome or the potential loss. The legacy is nonetheless already visible on the balance sheet today: $19.85 million of the Ironshore purchase price remains locked in escrow (recorded as restricted cash), in part because of this proceeding — and the final installment to the former Ironshore equity holders has not yet been paid out.

Anyone who sees SmartStop only as the landlord of 177 owned storage facilities misses the second business: as of December 31, 2025, the REIT managed 273 third-party facilities with roughly 140,000 units and 20.4 million square feet — more floor space than its own portfolio (13.9 million square feet). The platform consists of the self-sponsored fund REITs (SST VI, SSGT III and SST X, launched in January 2025) and, since October 1, 2025, of the acquired third-party manager Argus Professional Storage Management: 221 management contracts across 27 states, roughly 400 employees, purchase price about $23.4 million — plus an earnout of up to $11 million tied to 2028 platform revenue.

The business is small but capital-light and growing fast: $19.2 million of platform revenue in 2025 (after $11.4 million the year before) plus $12.5 million of reimbursed costs — and it hands SmartStop a pipeline on the side: the sponsor already manages the fund REITs' properties before they potentially migrate into its own portfolio. For investors it is a double-edged extra: recurring fees without property risk — but also conflicts of interest that the annual report itself lists as a risk factor, because executives simultaneously hold offices at the managed vehicles.

As a sponsor, SmartStop earns management and transaction fees from non-traded sister REITs such as Strategic Storage Trust VI (SST VI). The fine print of the annual report (10-K) for 2025 hides a remarkable construction: under a "Sponsor Funding Agreement", SmartStop covered, from November 2023 through June 2025, the front-end sales load for the share sales of its own fund REIT SST VI — the sponsor thus subsidized the raising of investor money on which it subsequently earns fees.

In return, SmartStop received "Series C Units" in the SST VI operating partnership. The punch line: these units convert into full-value units only if SST VI reports a net asset value of at least $10.00 per share — calculated after deducting the units to be converted. When SST VI reported exactly $10.00 in August 2024, not a single unit could be converted, because the conversion itself would have pushed the value below the threshold. On top of that, SmartStop must book the excess of its payments over the value of the units received as a reduction of its own management fees. The agreement was terminated as of June 30, 2025 — it remains an object lesson in how much capital circularity can sit between a sponsor and its fund vehicles.

Competitors do not usually lend each other money — yet at SmartStop, of all companies, the industry giant Extra Space Storage (NYSE: EXR) sat on the capital-provider side for years: through a subsidiary, Extra Space subscribed to $200 million of Series A convertible preferred stock of SmartStop starting in October 2019, paying 6.25 percent and, from October 2024, 7.0 percent. Only the IPO ended the arrangement: on April 4, 2025 — one day after the IPO closing — SmartStop repaid roughly $203.6 million out of the offering proceeds.

That did not end the entanglement, though: in December 2024, SmartStop bought a self-storage property in Ladera Ranch — its own California hometown — from Extra Space, and part of the purchase price was financed on the spot via a $42 million loan from Extra Space Storage LP (fixed 5.0 percent, due December 2027, secured by the property). Within a few years, the largest competitor was thus preferred shareholder, seller and secured lender all in one. Everything disclosed, everything at market rates — but it shows how tightly knit the self-storage industry is, and that SmartStop's capital sources before the IPO were expensive: nobody pays a 7 percent preferred dividend voluntarily when cheaper alternatives exist.

Millrose's starting capital was not cash but land: at the spin-off, Lennar contributed lots worth about $5.5 billion (roughly 87,000 homesites) plus about $1 billion of cash; three days later Millrose paid another roughly $859 million for the land of homebuilder Rausch Coleman. Who determined those values? The seller itself. The risk section of the annual report (10-K) states verbatim: "We have not obtained independent appraisals or fairness opinions as to the value of our real estate assets, including those acquired in the Spin-Off from Lennar and the Rausch Transaction".

The same report concedes that environmental assessments do not exist for every property and that the company relies on its counterparties for information about the homesites — above all on Lennar, which is simultaneously its largest customer. The book values may be perfectly correct; they were just never checked by a neutral party. For a stock trading below book value, that is not a footnote — it is the core question.

Millrose Properties manages land worth $9.2 billion (March 31, 2026) — and employs nobody to do it. The annual report (10-K) is explicit: all personnel, up to and including the company's own executive officers, are provided and paid by the external manager Kennedy Lewis; Millrose books no payroll expense at all. In exchange flows the management fee of 1.25 percent per year on "Tangible Assets" — $87.8 million in the 2025 stub year, and $28.2 million in the first quarter of 2026 alone (annualized more than $110 million).

The curiosity sits in the fee base: under the contract's definition, for roughly the first 13 months after the spin-off the cash balance counted as part of Tangible Assets — only afterwards is cash carved out. For a while, the manager literally earned a fee on uninvested money. It fits the sober legal nature of the relationship, which the 10-K itself names: "contractual, not fiduciary" — the manager's liability is limited, and Millrose indemnifies it against certain claims. Externally managed REITs are a well-known, often criticized model on Wall Street — here you get it in its purest form.

At the February 2025 spin-off Lennar kept roughly 20 percent of Millrose — and swapped it back in November 2025 through an exchange offer for its own shares: 33,298,754 Millrose shares came back, and in return Lennar retired 8,049,594 of its own shares. Since then, the annual report (10-K) says, the former parent owns only a "de minimis" stake — practically nothing.

What was not sold along with the shares: the special rights. The Founder's Rights Agreement gives Lennar an evergreen capital priority right (Lennar may reserve part of Millrose's available capital exclusively for its own future land deals), a most-favored-pricing clause on option rates (if another builder gets a lower rate, Lennar may match it for future deals), an approval right over any new manager should the Kennedy Lewis contract end — and per the 10-K, Millrose may not even take on debt above a 1:1 ratio to equity "unless Millrose obtains the prior approval of Lennar". A shareholder that is no longer a shareholder but still co-governs: worth knowing before you read the 10 percent dividend as an ordinary REIT coupon.

On July 2, 2026, Lam Research chief Tim Archer exercised options on 30,000 shares at an exercise price of $30.03 — and sold every share the same day at $390.01, thirteen times as much. He still holds more than a million shares afterwards; at the same time the company was buying back its own stock at ever higher prices ($105.67 → $210.57 per quarter in fiscal year 2026).

Lam Research hangs on both ends of the trade conflict: Washington regulates what Lam may sell to China — and Beijing, per the quarterly report (10-Q) as of March 29, 2026, controls the export of rare earth elements that Lam needs for its own manufacturing. The relaxation expressly applies only "in part" and until November 2026, unless extended.

Ernst & Young had audited Lam Research's books since 1981 — it says so in black and white in the audit opinion of the annual report (10-K) for 2025. On September 8, 2025, the audit committee resolved the dismissal of E&Y and engaged KPMG for fiscal year 2026.

Per the current report (8-K), there were no disagreements or reportable events — a routine change, then, but one that had not happened in four decades.

An easily overlooked second leg: alongside selling its power modules, Vicor licenses its technology for recurring fees. A side business has turned into an earnings block: licensing revenue rose from $15.9 million (3.9 percent of revenue) in 2023 to $57.4 million (14.1 percent) in 2025 — almost a quadrupling in two years. Because license fees cause hardly any costs, they are highly profitable and lift the reported gross margin. The flip side: such payments can be lumpy and dependent on individual agreements or litigation — and they explain why Vicor enforces its patents so aggressively (see the $45 million settlement in 2025). For investors it is a double-edged find: a real, high-margin source of earnings, but not as reliable as classic product revenue.

Vicor has two share classes: the listed Common Stock (ticker VICR, one vote per share) and an unlisted Class B share carrying ten votes each. Almost all Class B shares belong to founder Dr. Patrizio Vinciarelli, who is at the same time Chairman of the Board, CEO and President. As of March 31, 2026, he held 27.1 percent of the Common and 94.1 percent of the Class B shares — together 79.1 percent of the voting power, with only about a quarter of the capital. Vicor is therefore a "controlled company" under Nasdaq rule 5615(a)(7) and exempt from key governance requirements: it needs neither a majority-independent board of directors nor an independent compensation committee. For minority shareholders that means: they carry the capital risk but have effectively no say — and a hostile acquirer or activist can never "clean up" at Vicor.

On paper it is a dream year: Vicor's net income jumped in 2025 from $6.1 million to $118.6 million. But whoever reads the annual report finds two one-time effects that together make up about $64 million — roughly 54 percent — of this "record". First, a patent settlement: "In the second quarter of 2025, we received $45 million as a patent litigation settlement", less $5.1 million in legal fees. Second, an accounting effect without any cash: at the end of 2025, Vicor released a tax valuation allowance of $43,648,000 — for years the company had not trusted its own deferred tax assets ever to be used; now it reversed that judgment, which flowed straight into the profit as a tax benefit. The operating core business contributed only about $37 million (operating margin about 9 percent). Whoever takes the reported net margin of 29 percent for the normal state of affairs considerably overestimates Vicor's earnings power.

Old Dominion does not rent, Old Dominion buys: of the network's 260 service centers, the company owned 240 outright as of December 31, 2025; the balance-sheet line "Land and structures" stands at $3,523.4 million at cost — nearly two thirds of a full year's revenue, tied up in docks, terminals and land. Historically, the company says it spends 10 to 15 percent of revenue per year on capital expenditures, explicitly building ahead of future growth.

This strategy has a flip side the annual report itself names: "… prior capital investments based on our projections may contribute to excess capacity that could negatively impact our profitability." That is exactly what has been happening since 2023: the network is built for more freight than shrinking demand delivers — depreciation keeps running, the operating ratio keeps climbing. In an upturn the empty space becomes operating leverage; until then, Old Dominion pays the storage fee on its own bet on the future.

Tucked into the human-capital chapter of the annual report (10-K) for 2025 is a number you would not expect at a freight company: "Since 1988, we have provided a no-cost opportunity for qualified employees to become drivers through the 'Old Dominion Driver Training Program.'" As of December 31, 2025, 3,439 active drivers were graduates of this school — roughly 33.3 percent of the entire driver workforce of 10,320.

Why this is more than a staffing footnote: the trucking industry notoriously suffers from driver shortages and high turnover. Old Dominion puts the ten-year turnover rate of its own driving-school graduates at roughly 7.7 percent — below the company-wide driver average of roughly 10 percent, which is itself remarkably low for the industry. More than 26 percent of drivers have logged over one million accident-free miles. The free driving school is a quiet moat: it produces loyal, safe drivers in an industry where exactly that is scarcest — and it shows up in no financial metric.

Old Dominion is a company with a market value of a good $40 billion, an S&P 500 member — and at its core still a family business. The annual report (10-K) for 2025 carries its own risk factor for it: "David S. Congdon, John R. Congdon, Jr. and their affiliate family members beneficially own an aggregate of approximately 10% of the outstanding shares of our common stock."

The two are cousins — one Executive Chairman of the board, the other a regular board member — and grandsons of the founding generation: Old Dominion was founded in 1934 and has been a Virginia corporation since 1950. The 10-K notes dryly that the family "may be able to significantly impact the outcome of all matters involving a shareholder vote," and that its interests may differ from those of other shareholders. For investors this cuts both ways: a founding family with billions invested in its own company rarely thinks in quarters — but whoever buys the stock always rides shotgun when it comes to votes.

The Grasberg district in Indonesia delivers 98 percent of Freeport-McMoRan's gold and almost a third of its copper — but the operating company, PT Freeport Indonesia (PTFI), belongs to the group only to 48.76 percent. The majority of 51.24 percent has been held since the 2018 transaction by Indonesia's state — via the state holding MIND ID and a regional entity. FCX fully consolidates PTFI and runs the operations — but the majority owner is Jakarta.

Even more remarkable is the price of the future: the mining rights (IUPK) run through 2031 and are extendable to 2041 under conditions. For the extension beyond 2041, the annual report (10-K) for 2025 states: "We expect to maintain our ownership interest in PTFI of approximately 49% through 2041 and hold approximately 37% beginning in 2042, following the transfer of an additional interest to an Indonesia state-owned enterprise." The extension of the best mine is thus paid for with a further transfer of ownership — a detail that shows up in no quality metric.

In the risk factors of the annual report (10-K) for 2025, Freeport-McMoRan describes at length how dependent the company has become on digitalization and automation — including AI in operating processes and automated ore loading systems underground. And then comes a sentence you would not expect from a company of this size: "We do not maintain cyber risk insurance, and the lack of insurance coverage could adversely affect our cash flows and overall profitability in the event of a cybersecurity incident that has a material adverse effect on our business."

A company whose most important mine runs on driverless trains and increasingly networked operational technology therefore carries cyber damage entirely on its own books. That may be a deliberate cost-benefit decision — cyber policies for heavy industry are expensive and full of gaps. But it is a footnote hardly any investor has on the list when thinking about the risks of a mining company.

The notes to the annual report (10-K) for 2025 contain a line that does not fit a pure copper, gold and molybdenum company at first glance: oil and gas properties are carried "net of accumulated amortization and impairments of $27.5 billion". It is the accounting grave of the 2013 oil adventure, when Freeport-McMoRan spent billions acquiring oil and gas companies and later had to write most of it off.

The ghost still costs money today: 2025 brought further charges of $118 million for the legacy oil and gas properties (2024: $222 million) — mostly impairments and adjustments to plugging and abandonment obligations for old wells and platforms. Whoever buys FCX today as a pure copper bet also inherits the wind-down bill of a failed strategic detour that lies more than a decade in the past.

Deep in the tax chapter of the annual report (10-K) for 2025, Rambus clears out a claim it had carried on its books for years: on license payments from South Korea — by far its most important revenue country — the local authorities withhold taxes whose refund Rambus had recorded as a receivable since 2018. In the third quarter of 2025 came the reversal: "… we determined it is not more likely than not that the withholding taxes paid in South Korea are recoverable." The entire long-term tax receivable of $114.9 million was reduced to zero.

It did not hit earnings — a matching tax liability of the same size sat on the books and was written off in parallel, so the net tax effect was zero. The episode remains remarkable all the same: it shows how much tax friction sits inside the core business with Korean major customers, and that the company itself, after years of waiting, wrote off a nine-figure refund claim. Whoever admires the spotless balance sheet finds here the footnote with the scars.

In the annual report (10-K) for 2025, Rambus lists its patent licensees — and alongside Western heavyweights such as AMD, Broadcom, NVIDIA and Qualcomm, there sits, as if it were the most natural thing, CXMT: ChangXin Memory Technologies, the state-backed hope of China's memory-chip industry. While Washington regulates exports of leading-edge semiconductor technology to China ever more tightly (the 10-K devotes its own risk section to export controls), patent royalties from China's DRAM build-up program flow to Silicon Valley at the same time.

The constellation is curious twice over: first, Rambus earns from precisely the competitor that aims to take market share from its established customers Micron, Samsung and SK hynix — the buyers of Rambus's chip business. Second, the entry shows how a patent-licensing model bridges geopolitical divides as long as only rights, not goods, change hands: a patent cannot be held up at customs. Whether that bridge holds through a further escalation of export and sanctions rules is another question.

The balance sheet as of December 31, 2025 contains an unassuming line with symbolic power: "Retained earnings (accumulated deficit): 76,795" — for the first time in the filings at hand, a positive figure. Just one year earlier, as of December 31, 2024, that line showed an accumulated deficit of minus $153.7 million; at the end of 2023 it was roughly minus $334 million. Translated: 35 years after its founding, the company has arithmetically refilled the accumulated losses of its history — the expensive development and litigation years — entirely with profits.

For investors this is more than accounting folklore: the sign flip marks how recently today's earnings power emerged. Only the combination of a growing chip business and stable patent royalties turned the notorious deficit into a surplus within two years (net income 2024: $179.8 million, 2025: $230.5 million). Anyone filing the stock away as a long-established cash machine should know: this machine has been running in its current form for only a few years.

The late annual report of 2023 had a legal epilogue: a shareholder filed a so-called derivative complaint — a suit brought by a shareholder on the company's behalf against its own officers. Per the annual report (10-K) for 2025, the matter wound through SEC filings for a good 18 months before the court granted final approval of the settlement on January 22, 2026. The sober verdict in the mandatory filing: "There was no material financial impact to results of operations or financial position of Arrow as a result of the settlement."

What the case reveals is the true cost of such episodes: no damages, no fine — but two and a half years of management, lawyer and auditor attention. And a personnel footnote: Arrow's auditor today is Crowe LLP; in 2023, when the internal-control assessment triggered the filing jam, the audit chair was still held by KPMG.

The deposits chapter of the annual report (10-K) for 2025 contains a number that makes Arrow Bank's funding model special: $787.1 million of municipal deposits (prior year: $684.8 million) — roughly a fifth of $3.94 billion in total deposits. Towns, counties and school districts across upstate New York park their tax money with the community bank from Glens Falls. Such balances are large, price-sensitive and must be collateralized under New York law: $276.1 million of deposits were secured with pledged securities as of December 31, 2025.

The second number from the same chapter fits the picture: an estimated $917.6 million of deposits above the FDIC insurance limit (including $57.5 million of intercompany balances) looks at first glance like a Silicon Valley Bank-style concentration risk — but softens considerably once you note how much of it is collateralized or intercompany. What remains: whoever holds the stock also holds the confidence of upstate New York's municipal treasurers.

If Arrow Financial makes you think only of checking accounts and loans, the annual report (10-K) for 2025 lists a whole side business you would miss: the group owns the insurance agency Upstate Agency, LLC with nine offices of its own, selling property and casualty insurance and selling and servicing group health care policies and life insurance. Of the company's 578 full-time equivalent positions (as of December 31, 2025), 40 sit inside the insurance agency — and in the first quarter of 2026, insurance commissions contributed $2.1 million to non-interest income.

Beyond that, the annual report names an in-house REIT (real estate investment trust) and a registered investment adviser among the indirect subsidiaries — structures you would expect at a money-center bank rather than at a 38-branch community lender between Albany and the Adirondacks. For investors this matters twice: fee, fiduciary and insurance income makes revenue a little less hostage to interest rates — and in the margin-squeeze year of 2024, that cushion was exactly what was needed.

Open company file CIK 0000027419 in the SEC's EDGAR database and you find, under "Former Names," an entry many Target shareholders do not know: until April 12, 1999, the company was called Dayton Hudson Corporation — named for its department store roots, Dayton's of Minneapolis and Hudson's of Detroit. The first Target store opened back in 1962 as the department store group's discount offshoot, but only in 1999 did the most successful subsidiary take over the parent's name.

For dividend collectors that means: the famous streak — "We have paid dividends every quarter since our 1967 initial public offering," as every quarterly report still states — began under a different name. Trace the 54 consecutive years of increases back and you land in the 1970s of a department store group that no longer exists under that name. The Dayton's and Hudson's stores were renamed Marshall Field's in 2001 and sold in 2004 — what remained is the discount offshoot that today generates $104.8 billion in sales and whose nameplate is younger than its dividend history.

Between diapers and detergent, Target is growing a business that has little to do with shelves: Roundel, the in-house retail media network, sells advertising space on Target's website, app and other channels to vendors and marketplace sellers. Advertising revenue recognized within net sales jumped from $649 million to $915 million in fiscal 2025 (ended January 31, 2026) — up 41 percent — while merchandise sales fell 2.0 percent. In the first quarter of fiscal 2026 the pace accelerated: $246 million versus $163 million, up 51 percent; the 10-Q attributes the 24.6 percent growth in non-merchandise sales "primarily" to Roundel.

And the reported figure is only half the story: depending on the contract, Target books Roundel revenue not as sales but as a reduction of cost of sales or SG&A expenses — the real advertising business is bigger than the revenue line suggests. The annual report now devotes a dedicated risk factor to the ad business, explicitly warning about competitors' "artificial intelligence-enabled advertising solutions." A retailer whose highest-margin growth business is advertising — worth knowing before you value Target as a pure store chain.

If you take Target's GAAP earnings of $8.13 per share for fiscal 2025 (ended January 31, 2026) as the operating truth, you are missing a line in the fine print: the company booked $593 million of net pretax gains from settlements of credit card interchange fee litigation — the processing fees U.S. retailers have been fighting over with the card networks for years. The windfall lifted GAAP earnings by $0.97 per share; adjusted, Target earned just $7.57 per share, a decline of 14.5 percent instead of the optical 8.2 percent.

The punchline: a retailer whose operating result shrank for the third year in a row got the cushion that made its accounts look friendlier from, of all places, its payment processors — roughly a sixth of the $3.7 billion in net income came out of the courtroom. One quarter later the same effect flipped the optics the other way: because the prior-year quarter contained the windfall, GAAP EPS fell 24.5 percent in the first quarter of fiscal 2026 — while adjusted earnings rose 31.6 percent.

West Pharmaceutical supplies practically the entire pharmaceutical industry — and yet depends increasingly on a single name the 2025 annual report (10-K) does not disclose: "one of these customers individually accounted for more than 10% of consolidated net sales, at 15.8% or $485.9 million." A year earlier the same line item stood at 12.3 percent ($356.4 million) — a jump of 36 percent in twelve months, cutting across both segments (components and contract manufacturing). The ten largest customers together account for 47.6 percent of revenue.

The filing does not say who the customer is; it does say what drives the growth: contract manufacturing grew "primarily … [due to] self-injection devices for obesity and diabetes" — the GLP-1 business. The boom carrying West's comeback is thus concentrating its revenue at the same time: the better the weight-loss pens sell, the bigger the single line item whose loss the risk section explicitly lists as a danger.

West's highest-margin products carry brand names like FluroTec (fluoropolymer-coated stoppers) and Crystal Zenith (polymer vials and syringes). What the risk section of the 2025 annual report (10-K) discloses: these technologies do not belong to West but to Japanese partner Daikyo Seiko — West owns 49 percent of Daikyo, yet licenses the processes under contracts that, per the filing, expire in 2027: "Our rights to these products and processes are licensed pursuant to agreements that expire in 2027." If they are terminated early or not renewed, the business "could be adversely impacted."

For context: the partnership goes back decades, West even hedges its Daikyo stake with a dedicated $130 million cross-currency swap, and walking away would hurt both sides. Still, the mechanism is remarkable: a $24 billion company whose high-value lineup partly rests on borrowed technology has to quietly renew, within two years, what investors have long assumed to be its own property. The renewal is silently priced into the stock — it is not yet in the contracts.

Scroll through West Pharmaceutical's 2025 annual report (10-K) to the market-risk chapter (Item 7A) and you find a position you would sooner expect at an airline: as of December 31, 2025, the company held call options on 184,075 barrels of crude oil with maturities through June 2027 at a weighted-average strike price of $72.94 per barrel. The reason sits right next to it: the rubber stoppers and seals West earns its living from are made of synthetic elastomers — "derived from the petroleum refining process." Many supply contracts carry escalator clauses that trigger surcharges when oil prices rise; the options are the insurance against exactly that.

A second document shows how seriously to take the oil connection: in November 2023, West filed an amended Global Master Supply Agreement with ExxonMobil Product Solutions Company — a framework supply contract deemed material, with volume tables running through 2028 (the figures themselves are redacted). A medtech dividend aristocrat whose feedstock comes from an oil major and whose price risk is hedged with crude options: not a warning sign, but a rare look at what pharmaceutical packaging is actually made of.

Deep in Note 15 of the annual report (10-K) for fiscal 2025 sits a legacy case few investors have on their radar: on April 24, 2020, Franklin Templeton's Indian subsidiary closed six fixed income mutual fund schemes holding roughly $3.4 billion of client money (INR 25,648.3 crore) to redemptions and wound them up — in the middle of the Covid liquidity crunch. Unitholders and authorities went to court, and after a forensic audit the regulator SEBI opened proceedings against the Indian subsidiaries and individual employees.

The remarkable part is the outcome so far: by September 2023, INR 27,508.1 crore (about $3.3 billion) had been distributed to unitholders — more than the funds were worth on the day of the closure announcement, because the wind-down ran into a recovering market. Yet the matter is not over: parts of the appeals remain pending before the Supreme Court of India, and individual SEBI proceedings continue, per the annual report. As context for the Western Asset turbulence, the case is a useful precedent — it shows that legal risks at an asset manager of this size take years to truly leave the building.

The annual report (10-K) for fiscal 2025 lists the executive biographies — and under Executive Chairman Gregory E. Johnson (64) sits a side job you would not expect at a trillion-dollar asset manager: "Chairman of the San Francisco Giants, a professional baseball organization, since November 2019." The chairman of the board of a $1.7 trillion asset manager also chairs a Major League Baseball club.

The rest of the org chart is just as remarkable: CEO Jennifer M. Johnson (61) is Gregory's sister, and Vice Chairman Rupert H. Johnson, Jr. (85) is their uncle — a director without interruption since 1971. The company, founded in 1947 by the current CEO's grandfather and named after Benjamin Franklin, is thus run by the same family in its third generation; the annual report spells out the kinship explicitly under "Family Relationships." For investors it cuts both ways: long-term continuity on one side — and a power bloc of one family, baseball side job included, on the other.

Run the numbers in the statements of stockholders' equity in Franklin Resources' annual report (10-K) for fiscal 2025 and you find a sequence that clashes with the image of the unshakeable dividend payer: in fiscal 2024 the company declared $670.1 million in dividends — against only $464.8 million of net income. In fiscal 2025 it was $688.4 million of dividends against $524.9 million of income. For two consecutive years the GAAP payout ratio sat above 130 percent; per share, a $1.28 dividend most recently stood against $0.91 of earnings.

Nothing is broken because of it: the gap came mostly from non-cash impairments of fund management contracts — $389.2 million in fiscal 2024 and $226.6 million in fiscal 2025, largely on the scandal-hit bond subsidiary Western Asset. On adjusted earnings ($2.22 per share in fiscal 2025) the ratio was 58 percent, and in the first half of fiscal 2026 GAAP earnings ($0.95 per share) covered the dividend ($0.66) again. But the finding stands: precisely in the years when the 45-year raise streak was supposed to prove its reliability, it was paid out of the balance-sheet cushion — not out of reported earnings.

Nucor's famous pay-for-performance culture is not just brochure material — it is quantified in the annual report. The company funds a Profit Sharing and Retirement Savings Plan whose contributions track profitability. The series in Note 17 of the 10-K for 2025 reads like a business-cycle barometer: $611 million in 2023, $298 million in 2024, $256 million in 2025. Workforce compensation breathes with the steel cycle — in good years Nucor teammates earn well above industry average, in weak years the variable share shrinks without the company resorting to mass layoffs.

For investors this is a double find. First, the system acts as an automatic cost buffer — a sizeable compensation block shrinks by itself when profits fall, cushioning margins in a downturn. Second, it explains part of the swing in marketing, administrative and other expenses that can distort quarter-over-quarter comparisons: when pre-tax earnings jump (as in the first quarter of 2026), accruals for profit sharing and bonuses jump with them. Anyone comparing Nucor's cost ratios with conventional industrial companies should know about this built-in cycle amplifier.

Read the raw materials segment description in the annual report (10-K) for 2025 and, between scrap brokerage and direct-reduced-iron plants, you find a half-sentence few investors would expect from a steel company: the segment also includes "our natural gas production operations" — plus the industrial gas business Universal Industrial Gases. In a small way, Nucor is also an energy and gases company: natural gas fires the direct reduced iron (DRI) facilities in Louisiana and Trinidad that turn iron ore into a scrap substitute, and every steel mill consumes industrial gases like oxygen and argon daily.

Strategically this is no accident but part of the company's raw materials doctrine: control as many upstream stages of steelmaking as possible — from scrap broker DJJ (North America's leading ferrous scrap broker, with shredders able to process roughly 6,800,000 tons a year) through its own DRI plants to energy itself. For investors, the takeaway is this: a small slice of the "steel company" Nucor earns (or loses) its money on gas prices rather than steel prices — and in segment reporting, this second business disappears almost entirely into the raw materials line.

On May 14, 2025, Nucor reported a cyberattack to the U.S. securities regulator, the SEC — as a "Material Cybersecurity Incident" under Item 1.05, the most serious disclosure category. The amended filing (8-K/A) of June 20, 2025 records what happened: a threat actor illegally accessed the company's information technology systems, access to portions of the applications supporting operations at some facilities was temporarily limited — and Nucor "temporarily and proactively halted certain production operations at various locations" in an abundance of caution. When a company that melts steel at over 2,900 degrees Fahrenheit idles furnaces because of an IT breach, it shows how deeply production control now depends on software.

The second uncomfortable line of the amendment: the investigation determined that the attacker exfiltrated "limited data" from the IT systems — copied and carried it off. Nucor activated its incident response plan, took systems offline, restored data from backups, brought in outside forensic experts and notified federal law enforcement. The company did not expect a material impact on the full year; the annual report (10-K) for 2025 accordingly no longer lists the incident as a drag on results. For investors, the finding still matters: even a company with fortress-balance-sheet ratings can become an industrial-shutdown case within days — and the 8-K/8-K/A disclosure chain worked here like a textbook example.

In the first quarter of 2026, Caterpillar drastically accelerated its share repurchases: $5.0 billion went into its own stock per the quarterly report (10-Q) — almost as much as in all of 2025 ($5.2 billion). The company signed accelerated share repurchase (ASR) agreements worth $4.50 billion with banks and advanced the full amount up front. Its cash pile fell by half, from $10.0 billion to $4.1 billion — within three months.

The purchase prices are the remarkable part: the monthly averages disclosed in the 10-Q were $626.54 (January), $659.85 (February) and $700.06 (March 2026) — after $559.93 and $588.28 in November and December 2025. Caterpillar is buying more aggressively the more expensive its own stock gets, and now pays a large multiple of book value (equity as of March 31, 2026: $18.7 billion across 460.6 million shares, about $41 per share). As long as profits grow, the effect per share is a booster — if the cycle turns, the company will have bought at record prices and spent the reserve doing it.

The June 2026 dividend filing (8-K) contains a sentence that is historic in its casualness: "Caterpillar has paid a cash dividend every year since the company was formed and has paid a quarterly dividend since 1933." The company was formed in 1925 — so the annual payout has survived the Great Depression, World War II, the oil crises, Caterpillar's own near-death experience in the early 1980s, the 2008 financial crisis and the pandemic. The quarterly streak began in the middle of the Depression.

The younger streak that makes Caterpillar an Aristocrat — 32 consecutive years of higher annual dividends — is almost juvenile by comparison. For scale: the June 10, 2026 increase of 8 percent to $1.63 per quarter costs the company roughly $3 billion a year — against $9.5 billion of free cash flow from the industrial business (MP&E) in 2025, that is routine, not a stretch.

If you think of Caterpillar as a pure machinery maker, you are missing almost half the balance sheet: the financing arm Cat Financial accounted for roughly $41.6 billion of the company's $95.6 billion in total assets as of March 31, 2026. It finances the yellow machines for dealers and customers — and runs at leverage you would normally expect from a bank: the quarterly report (10-Q) as of March 31, 2026, discloses a covenant leverage ratio of 8.03 to 1 (maximum allowed: 10 to 1) and an interest coverage ratio of 1.53 to 1 against a contractual minimum of 1.15 to 1.

That is standard practice for a captive finance business and agreed with the lenders — but it means a sizable part of the company lives off lending with thin safety buffers, not off selling machines. If Cat Financial misses one of these covenants, the filing notes, the bank syndicate can terminate its commitments and cross-default clauses can accelerate other borrowings. If you hold Caterpillar as a dividend rock, know that a bank levered 8 to 1 is working underneath the rock.

While U.S. corporations fill their annual reports with AI chapters, Porsche allows itself a remarkable minimalism: in the entire 2025 Annual and Sustainability Report, the term "Künstliche Intelligenz" (artificial intelligence) appears exactly one single time — not with products, not in production, but in human resources development: as one of eight "future competence" fields of employee training ("Künstliche Intelligenz und Software"). Neither as a revenue source nor as a named business risk does AI play any role in the sports car maker's reporting — in the year 2025, that is almost a curiosity in itself.

Why Porsche could not sell a combustion-engined 718 Boxster/Cayman in Europe for months in 2025 is recorded in the fine print of the management report: the sales gaps were caused by "die Angebotslücken beim 718 Boxster/Cayman und beim Macan mit Verbrennungsmotoren aufgrund der europäischen Vorschriften zur Cybersicherheit" — supply gaps for the combustion 718 and Macan due to European cybersecurity regulations. UN rules R155/R156, binding for all new EU vehicles since July 2024, require certified software-update and cyber-management systems — a retrofit that no longer paid off for the run-out model generations. The 718 lost about 21 percent of its volume in 2025 (18,612 after 23,670) — a curious lesson that these days paragraphs, not just emissions, decide the fate of engines.

Barely noticed between the quarterly figures: in March 2026 the supervisory boards of Porsche AG and Volkswagen AG cleared the sale of Porsche's stakes in the Rimac Group, in Bugatti Rimac and in Bugatti International Holding; the sale agreement followed in April 2026. Since March 31, 2026 the holdings sit on the balance sheet as "held for sale" under IFRS 5 — €411 million in assets waiting only for regulatory approvals. It closes a chapter: Porsche first invested in Croatian electric-hypercar pioneer Mate Rimac in 2018 and folded the Bugatti brand into the Bugatti Rimac joint venture in 2021. The sale fits the new "leaner, faster" doctrine — and shows how consistently the group is clearing out its electric prestige projects.

The retrospectively consolidated SPCX financials for 2023 contain one of the most expensive renamings in business history: an impairment of $3,775 million, per the prospectus "primarily related to the impairment of the Twitter brand following its rebranding to X." Without this item, the group's 2023 operating loss of $(3,505) million would have nearly disappeared.

For SPCX investors this is more than an anecdote: it shows how much paid-for substance sat in the Twitter acquisition that is now part of the listed story via common-control accounting — and how quickly brand value can vanish when a single individual decides it should.

Between rockets, satellites and GPU clusters, the SPCX balance sheet carries a line item few would expect: 18,712 Bitcoin, cost basis $661 million, fair value $1,637 million as of December 31, 2025 (prior year: $1,749 million). The decline ran through the 2025 income statement as an "unrealized loss on digital assets."

The curious part: Tesla has discussed its Bitcoin position prominently for years — but that Elon Musk's rocket company has quietly been sitting on a billion-dollar Bitcoin stash never appeared in any shareholder disclosure until the IPO prospectus. Against the group's $100 billion cash pile the position is small — for earnings volatility it is not: Bitcoin price swings now run through SPCX's income statement quarter after quarter.

Part of xAI's AI data centers does not belong to SPCX at all: three leases for compute equipment run through entities of Valor Equity Partners — the investment firm of Antonio Gracias, who also sits on the SPCX board. The payment obligations per the prospectus: "aggregate cash payments of $6,986 million," "$6,633 million" and "$6,587 million ... over the life of the lease" — together a good $20.2 billion, guaranteed by SpaceX. $885 million was paid in 2025, another $1,917 million in January through April 2026 alone.

All disclosed, all legal — but an investor should know: on the largest cost block of the fastest-growing segment, a board member sits on both sides of the table. The charter even explicitly renounces certain corporate opportunities in favor of Musk and individual directors ("we renounce certain corporate opportunities").

In January 2026, five months before the IPO, Elon Musk received two share awards totaling 1,302 million Class B shares — worth roughly $176 billion at the $135.00 IPO price. The "SpaceX CEO Award" (1.0 billion shares) vests, per the prospectus, only on market-cap milestones between $500 billion and $7.5 trillion — and, verbatim, on "the Company's establishment of a permanent human colony on Mars with at least one million inhabitants." The additional "AI CEO Award" (302.1 million shares) is tied, among other things, to "non-Earth-based data centers capable of delivering 100 terawatts of compute per year."

The accounting punchline: because both performance milestones are classified as "improbable," SPCX has not booked a single dollar of compensation expense for them. If the improbable happens, roughly 10 percent of additional shares sit waiting as a dilution overhang for Class A holders.

NPK International's growth figures read brilliantly — 27 percent more revenue, 39 percent more rental revenues. Yet the management discussion (MD&A) of the 2025 annual report (10-K) names the cause with unusual candor: the jump was "primarily attributable to our success on larger-scale, longer-term projects with a key utilities customer" — essentially driven by larger, longer-term projects with one key utilities customer. Combine that with the customer concentration (the three largest customers = 44 percent of revenue, contracts cancellable on short notice) and a concentration risk emerges that hardly any investor would expect from a casual glance at the pretty growth rate: the momentum for which the market currently pays a P/E of roughly 34 stands and falls to a large degree with a single customer relationship. If this customer throttles its project pipeline, the growth story loses its engine overnight.

Whoever files the sale of NPK's oilfield fluids business to private-equity firm SCF Partners away as "divested and forgotten" overlooks a curious detail in the notes of the 2025 annual report (10-K). NPK is not merely a creditor of residual receivables toward the buyer — the company even holds an interest-bearing loan (a "note receivable") of $5.0 million that pays 12.5 percent per year and does not come due until March 2030. A double-digit interest rate in the environment of 2025/26 is strikingly high — it betrays that the paper was judged riskier than a normal bank claim. For investors who know NPK only as a clean grid build-out lessor, this is a side find with substance: a piece of the old oilfield business still sits in the balance sheet as a credit relationship — including an attractive, but not risk-free, interest rate.

Concentra (NYSE: CON) looks broadly positioned — 628 centers in 41 states. But the distribution is anything but even: per the annual report (10-K), roughly 16 percent of the centers sit in California and 16 percent in Texas, plus 6 percent in Florida, 5 percent in Pennsylvania and 4 percent in Colorado.

Because reimbursement for workplace injuries (workers' compensation) is set by each state individually through official fee schedules, almost a third of the network thus hangs on the reimbursement rules of just two states. An unfavorable fee reform in California or Texas would hit Concentra far harder than the map with 41 states suggests — a geographic cluster risk that easily drowns in the "nationwide market leader" narrative.

Concentra (NYSE: CON) was spun off from Select Medical in 2024 as an "independent" company — and Select has not held a single share since November 2024. Yet the founding family of the old parent still sits firmly in the saddle: Robert A. Ortenzio, co-founder of Select Medical, is chairman of Concentra's board and, per the proxy statement (DEF 14A), holds roughly 6.4 percent of the shares. In addition, the Estate of Rocco A. Ortenzio — the second Select co-founder — reported a stake of 5.0 percent via a beneficial ownership filing (SC 13D).

Together the family thus controls a good tenth of the "independent" company and provides the board chairman. Nothing untoward in itself — founder ties can mean stability. What remains remarkable is how much umbilical cord to the former parent a spin-off can keep when on paper it is a clean cut.

Read AMN Healthcare's proxy statement (DEF 14A) and you find a compensation detail that casts the strike story in a new light. 70 percent of the top managers' annual cash bonus hangs on a single financial metric: "Adjusted EBITDA" (an adjusted operating result). The twist sits in the adjustment. Per the proxy, AMN expressly strips out "goodwill impairment loss, long-lived assets impairment loss, and gain on sale of disposal group" — that is, precisely the goodwill impairments of the boom years (2024 and 2025 together more than $330 million) that caused the reported net loss. The balance-sheet burdens of the expensive acquisitions thus weigh on the share price, but not on the bonus metric. And for 2026 the second effect kicks in: the one-time strike revenue of $721.9 million, which AMN itself calls "unpredictable", flows unfiltered into the same Adjusted EBITDA — mechanically lifting the bonus base. Shareholders evidently applaud only tepidly by now: say-on-pay approval fell to roughly 78 percent in 2026 — after a five-year average of 93 percent. None of this is impermissible, and impairment add-backs are industry standard. But it is a governance detail no investor expects behind the green momentum chart: the metric that pays the boss blanks out the costliest legacy burden and benefits from the strike windfall.

AMN Healthcare advertises a dedicated "Technology & Workforce Solutions" segment and positions itself as a technology-enabled staffing provider. All the more surprising is a find in the 2025 annual report: AMN sold its hospital scheduling software "Smart Square" and booked a gain on the deal ("gain on sale of disposal group") of roughly $39.1 million. Of all things, a software building block that should belong to the future story was divested — while the technology segment as a whole shrank (2025 revenue: −12 percent) and at the same time carries the largest remaining goodwill position on the balance sheet. That is no scandal, but a telling contrast between the tech narrative and the actual portfolio decision. For investors who file AMN under "healthcare tech", it is a side finding with substance.

Lay AMN Healthcare's cash flow statements side by side and you find a textbook case of procyclical timing. In the boom years, when the coffers overflowed thanks to COVID record bill rates, AMN spent the money on share buybacks: roughly $576.8 million in 2022 and another $424.7 million in 2023 — together a good billion dollars. The annual report puts the average price of the 2023 buybacks at $96.90 per share. Today the stock trades roughly three quarters below its all-time high — that is, at about a third of those buyback prices. The pattern is familiar: buybacks happen when the coffers are full and the price is high; when it later falls, the money is gone. AMN pays no dividend and today carries net debt of roughly $769 million. No investor who only sees the green momentum chart would expect that a billion was burned near the top here.

In the point-of-sale financing programs of the Atlanticus brand Curae, merchants pay a fee of up to 65 percent of the purchase price, per the annual report (10-K), so that even 0-percent promotions remain profitable for the lender. Together with annual percentage rates of 19.99 to 36 percent, that explains a return on equity of roughly 21 percent — against annualized charge-offs of 17.1 percent and $406 million of gross charge-offs in the first quarter of 2026 alone.

In November 2024, a Schedule 13D on Allient surfaced in the SEC's EDGAR system — filed by Juniper Investment Company, LLC of New York (signed by John A. Bartholdson). The stake covers roughly 5.2 percent of the shares outstanding. The twist sits in the form type: a 13D is the filing of an active investor who reserves the right to influence the company — as opposed to the passive 13G that index funds, for instance, use. True, the statement of intent in the document is worded cautiously ("for the purpose of making an investment"), but the mere choice of the active form is a signal: someone did not buy in merely in passing. For investors who know Allient only as a momentum chart, this is a side finding with substance — on the shareholder register stands an investor keeping its options open.

Whoever reads Allient's proxy statement (DEF 14A) closely stumbles over a family detail that easily drowns in the success story. In the section on executive officers it says expressly: "There are no family relationships between any of our directors or executive officers except that Mr. Stephen R. Warzala is the adult son of Mr. Richard S. Warzala." Richard S. Warzala (born 1953) is at once board chairman, chief executive officer and president and has steered the company since 2006. His son Stephen (born 1983) is president of the Allient Defense unit and chief growth officer — responsible, of all things, for two sensitive areas: the defense business and growth. Nothing about this is illegal or unusual for a family-shaped industrial company, and insiders together hold roughly 15 percent of the shares. But it is a governance detail worth knowing: at Allient, responsibility stays partly in the family.

In the supply-chain panic of 2022/2023, customers paid Bel Fuse premiums to get components delivered faster. These "raw material expedite fees" were real but fleeting money: in the Power segment alone they brought in roughly $14.9 million in 2023 — and shrank to $0.1 million in 2024 as supply chains normalized. Part of the apparent margin strength of the boom years was thus a one-off effect that disappeared as quickly as it had come. Whoever judges Bel's margin trajectory should know this one-timer — it explains a good part of the optical decline from 2023 to 2024.

In its $325.6 million purchase of defense supplier Enercon (Israel) at the end of 2024, Bel Fuse initially took over only 80 percent. For the remaining 20 percent, a contractual put/call option applies that Bel intends to exercise by early 2027. The twist sits on the balance sheet: Bel has booked a provision as a "redeemable noncontrolling interest" — initially $72.4 million, valued via Monte Carlo simulation with an assumed EBITDA volatility of 51 percent. If Enercon performs well, the remainder purchase gets more expensive; if it performs poorly, impairments loom on the $215 million of goodwill. Either way, the "acquisition" is not yet finished on the balance sheet — a detail that easily drowns in the cheering over the defense story. On the side, Bel acquired Ethernet specialist dataMate for roughly $16 million in March 2026: the group remains a serial acquirer.

Since its 1998 recapitalization, Bel Fuse has had two share classes: the Class A (BELFA) with one vote per share and the Class B (BELFB) with no voting rights at all. Whoever buys "the Bel Fuse stock" almost always buys the much larger, more liquid Class B — and gets no say in anything. Voting runs exclusively through the roughly 2.1 million Class A shares (just 26 holders of record), controlled by the founding family around Chairman Daniel Bernstein and a handful of holders such as the Gabelli/GAMCO group. As compensation, the charter locks in that the non-voting Class B receives at least 5 percent more dividend per share than the Class A ($0.07 versus $0.06 per quarter). A curious trade: a little more money for zero say — one that many buyers of the $3.5 billion stock never even notice.

Deep in the tables of the annual report (10-K) for 2025 sits a number that never makes a headline: of the 212,718 auto loans in the company's own portfolio ($3.78 billion), 99,830 accounts had received at least one payment extension as of December 31, 2025 — together $1.69 billion of remaining balances, or about 45 percent of the portfolio. 58,326 accounts have been extended twice or more.

An extension pushes the due payment back by one month; CPS allows up to two per year and eight over the life of a loan and classifies them as "insignificant delays". The statistical effect: extended loans do not count as delinquent — the reported 14.8 percent delinquency-plus-repossession ratio is therefore the ratio after this relief valve has been applied. For comparison: at the end of 2023 the extension total stood at $1.24 billion. The tool is industry-standard and fully disclosed — but anyone judging CPS's credit quality should know that nearly every second dollar in the book has already had to catch its breath.

Consumer Portfolio Services refinances its loan book almost entirely through securitizations and bank lines — but at the very bottom of the capital structure sits a curiosity: "subordinated renewable notes" of $29.0 million (December 31, 2025) that CPS continuously sells to retail investors under a prospectus. Average interest rate in 2025: 9.8 percent — subordinated meaning that in a crisis the securitization holders, the warehouse banks and the residual financiers get paid first, and the note buyers last.

What makes it remarkable is who else buys: the annual report documents that the audit committee has given general approval for purchases of these subordinated notes by the company's own executive officers — on the same terms available to the public. You can read that as a vote of confidence: whoever puts personal money into the company on a subordinated basis believes in it. Or as a hint at how expensive unsecured capital is for a sub-prime lender whose unrestricted cash stood at $6.3 million at year-end.

Share buybacks are considered shareholder-friendly — but the annual report (10-K) for 2025 of Consumer Portfolio Services describes an unusual variant: in June 2024 the company bought 50,000 of its own shares directly from CEO Charles E. Bradley, Jr. (at the previous day's closing price of $8.98), and in September 2024 another 70,000 shares at $9.85 — more than $1.1 million that flowed straight to the boss rather than to anonymous sellers on the exchange. In 2025 the pattern repeated with director William B. Roberts: 100,000 shares in September (at $8.62) and another 100,000 in December 2025 (at $8.69), roughly $1.7 million in total.

The spicy part sits in the related-party section: the repurchases from the CEO and the December tranche from Mr. Roberts were "not preapproved" — contrary to the company's own policy for transactions with insiders. The audit committee blessed the deals after the fact ("subsequently ratified"). None of this is illegal, and the prices matched the market. But it shows how short the corridors are in a company whose CEO has served since January 1992 — and whose 2025 say-on-pay vote drew only 62 percent approval.

That a washing-machine maker from Ripon, Wisconsin — population about 7,500 — operates its own corporate aircraft is remarkable enough. The proxy statement (DEF 14A dated April 27, 2026) adds the punchline: CEO Michael Schoeb's employment agreement guarantees him "the use of our corporate aircraft for personal use for up to 25 flight hours per year" — a clause carried over from his 2015 contract into the new agreement signed for the IPO.

The honest coda sits right next to it: "During our 2025 fiscal year, Mr. Schoeb made no personal use of our corporate aircraft." A boardroom perk that gleams on paper and stays on the ground in practice — too small a governance finding for the analysis, too good a footnote to throw away.

Alliance Laundry does not just sell washers — it finances them. An in-house financing organization lends primarily to laundromat operators buying company-branded equipment through the distributor network. The receivables flow into purpose-built, bankruptcy-remote special-purpose entities and a trust that refinance themselves through securitization facilities — in May 2025 the equipment facility's lender commitment was raised to $500.0 million, alongside a $120 million trade-receivables facility.

As of March 31, 2026, the balance sheet carried $620.3 million of "Asset backed borrowings — owed to securitization investors" — on top of the $1.3 billion Term Loan. The financing business contributed about $49.6 million of revenue in 2025 and ties customers to the brand twice over. But it also means nearly a fifth of the group's debt belongs to a built-in bank whose credit risks (laundromat operators!) live in the footnotes — hardly what anyone expects behind the ticker of a machinery maker.

Whoever underwrites an IPO earns underwriting fees. Note 25 of Alliance Laundry's annual report (10-K) for 2025 discloses a constellation you rarely see spelled out this plainly: majority owner BDT — which sold 18.8 million shares for its own account in the very same offering — also served as one of the underwriters of its own IPO, collecting about $2.8 million in fees ("BDT served as one of the underwriters of the IPO for which it received underwriting discounts and commissions of approximately $2.8 million").

It is not the only double role. For arranging the 2024 Term Loan — the loan that funded the $900 million dividend to BDT and management — BDT collected a $5.2 million arrangement fee. And per the same note, entities affiliated with BDT hold a controlling interest in a raw-materials vendor from which Alliance Laundry bought $7.2 million of supplies in 2025. Seller, loan arranger, underwriter, vendor — all disclosed, all legal. But you should know how many chairs one party occupies on the other side of the table before you sit down at it.

Nearly half of World Acceptance belongs to a single camp: Prescott General Partners, LLC and its affiliates beneficially owned about 46.3 percent of the common stock as of March 31, 2026. The annual report itself warns in the risk factors that a small number of shareholders "may exert significant influence" over everything put to a vote — from board elections to takeover questions.

On September 3, 2025, that turned into a remarkable transaction: after approval by the audit committee, the company repurchased 347,064 of its own shares for $60.0 million directly from Prescott affiliates, in a privately negotiated transaction at the day's closing price ($172.88). For every other shareholder this means two things: the company's buyback purse financed the anchor shareholder's partial exit — and the already thin float of a company with only about 4.6 million shares got thinner still. None of this is illegal, and all of it is disclosed. But whoever invests here should know that the most important address on the shareholder register is not the stock exchange.

When CEO R. Chad Prashad resigned effective April 10, 2026, "in order to pursue other opportunities", both sides agreed to treat the resignation as a termination by the company without cause — with everything that entails: $1,260,000 in severance, payable over 24 months, accelerated vesting of all time-based equity awards, payments under the supplemental retirement plan, an 18-month health-insurance subsidy — and, as a detail at the edge of Note 19: title to his company car and a mobile phone.

The succession is just as remarkable: Janet L. Matricciani took over as interim CEO — the same executive who already ran World Acceptance from 2015 to 2018, with earlier stops at Capital One and Countrywide Financial. After Prashad's departure the board shrank from seven to six members. Together with the simultaneous change of chief operating officer (J. Tobin Turner succeeded the retiring D. Clinton Dyer in February 2026), the firm re-staffed its two most important operating posts within weeks — in the middle of a fiscal year with a 61 percent profit decline.

Behind the ticker WRLD you would expect installment loans — not this: in fiscal 2026, World Acceptance prepared roughly 91,000 U.S. income tax returns and netted about $40.4 million from it (prior years: $36.5 and $29.1 million). The tax service runs in nearly all of the 1,009 loan branches, including interest- and fee-free advances on the expected tax refund ($500 to $7,000) — a growing second business that also deepens customer ties in the core lending trade.

The punchline sits in Note 2 of the annual report (10-K) for fiscal 2026: it was precisely in this side business that the company found an error in revenue recognition. Fees for the "Refund Assurance Plan" — a kind of tax protection plan with a three-year term — had been booked as revenue entirely at the time the return was prepared, instead of ratably over the 36-month coverage period. Cumulative effect for the years before fiscal 2024: $1.7 million, recorded as an adjustment to retained earnings as of April 1, 2023. The company deems the error immaterial — so there was no formal restatement of earlier reports, only revised comparative figures.

Effective January 1, 2026, Lantheus sold its legacy SPECT business (TechneLite, Cardiolite, NEUROLITE, among others) to SHINE Technologies — total consideration per the annual report: "up to $155.0 million". The quarterly report (10-Q) as of March 31, 2026 breaks down what actually changed hands on closing day: $31.4 million in cash — the rest consists of an installment note of $67.2 million, a seller note of $14.5 million, $12.3 million of deferred purchase price and $5.1 million of net contingent earnout receivables. Fair value at the closing date: $130.5 million.

Translated: Lantheus largely financed its own buyer and will carry that buyer's ability to pay as a risk on its books for years — while the "$155 million" headline suggests the money is in. That fine print matters when judging the $59.3 million book gain from the sale that flattered the first quarter of 2026.

Tucked into the annual report (10-K) for 2025 is a line of business you would not immediately expect from a maker of radioactive imaging agents: Digital Solutions. Through its subsidiary EXINI, Lantheus sells AI medical software — including aPROMISE, marketed in the United States as PYLARIFY AI, which performs standardized quantitative assessment of PSMA PET/CT images in prostate cancer, and the Automated Bone Scan Index, which classifies tumor burden on bone scans using an artificial neural network. Both are FDA cleared and carry a CE mark.

The logic is elegant: the software makes the readouts of the company's own imaging agents better and ties clinics into the Lantheus ecosystem — and the company openly states that it may use clinical imaging data to develop further AI solutions. On the revenue side this still sits inside "Strategic Partnerships and Other Revenue" ($59.4 million in 2025, the smallest category) — a second business in its infancy, but one with real regulatory clearances instead of mere AI prose.

Put the Lantheus balance sheets of year-end 2024 and year-end 2025 side by side and you can watch one number melt: cash fell from $912.8 million to $359.1 million. No loss-making business is behind it — the company generated $390.1 million of operating cash flow in 2025 — but a deliberate overhaul offensive: $268.9 million net for the contract manufacturer Evergreen Theragnostics, $306.7 million net for Life Molecular Imaging (the Alzheimer's diagnostic Neuraceq) and roughly $300 million for its own shares.

What stands out is the pace: within a single fiscal year, more than half of the cash cushion was redeployed — from "money in the bank" into "bets on the post-PYLARIFY era". On top of that, a $575 million convertible note (2.625 percent) comes due in December 2027. The balance sheet remains solid ($1.09 billion of equity), but the character of the company changed noticeably in twelve months: less cushion, more wager.

In the "Human Capital Management" section of the annual report (10-K) for 2025 — the place where companies usually write about workforce and training — Fortinet abruptly places a geopolitical sentence: "We do not own any manufacturing or research and development activities in China."

That a U.S. security company states this explicitly is no accident: Fortinet sells to governments and critical infrastructure, whose buyers now scrutinize supply-chain origins closely. Co-founder and CTO Michael Xie and CEO Ken Xie are U.S. entrepreneurs born in China — the sentence reads like a pre-emptive answer to a question that gets asked regularly in Washington. For investors it is a small, telling signal of how deeply geopolitics now reaches into the filings of the security industry.

Among the usual cyber risks in the annual report (10-K) for 2025 sits a risk factor you would sooner expect at a property company: "Our real estate investments, including construction, acquisition, development or leasing of new data centers …" — complete with warnings about impairments, environmental liabilities and construction risks. Behind it lies a deliberate strategy: Fortinet prefers to buy and build offices and data centers rather than rent. Net property and equipment grew to $1,619 million in 2025, and the holdings in Europe, the Middle East and Africa nearly tripled within a year, from $73.3 million to $211.3 million.

For investors this is remarkable twice over: first, it ties up capital that other software companies return to shareholders or spend on cloud rent. Second, Fortinet's 10-K explicitly names the continued "capital expenditures in data centers and real estate" as one reason the operating margin is expected to decline in 2026. A cybersecurity investment with an attached property developer — hardly anyone would have guessed that from a casual glance at the 80 percent gross margin.

The balance sheet as of December 31, 2025 contains a line that, at first glance, does not fit a company with $1.85 billion of annual profit: an accumulated deficit of minus $507.9 million. The reason is not a hidden loss-making business but capital returns on a grand scale: since inception of its repurchase program, Fortinet has bought back 267.3 million of its own shares for $8.51 billion — more than the company has retained in earnings over the years.

The consequence: on $10.4 billion of total assets, only $1.24 billion of equity remains (an equity ratio of roughly 12 percent) — and every equity-based metric turns into a distortion. Return on equity computes to more than 130 percent, not because the business got that much better, but because the denominator has nearly vanished. Anyone viewing Fortinet through a quality lens with an ROE focus should know this mechanic.

Sezzle was incorporated in Delaware on January 4, 2016, and is based in Minneapolis — yet for years the U.S. company was listed exclusively in Australia: per its SEC filings, Sezzle historically financed itself through capital raises on the Australian Securities Exchange (ASX), and only since the 2023 ASX delisting has Nasdaq been its sole trading exchange.

The 10-K for fiscal year 2023 still carries the traces on its cover page: the public float is valued there at an ASX closing price of 21.57 Australian dollars (June 30, 2023, before the 6-for-1 stock split of March 2025), and just 5,633,172 shares were outstanding — only the split turned that miniature share count into an ordinary one. Whoever looks at the Nasdaq ticker SEZL today would never guess that this U.S. company spent its first stock market years in Sydney.

As of December 31, 2025, Sezzle employed just 201 people across the United States and Canada — against $450.3 million in revenue and $133.1 million in net income for fiscal year 2025, that is roughly $2.2 million of revenue and $660,000 of profit per head.

The rest of the work runs, per the annual report (10-K), through so-called professional employer organizations (PEOs) and independent contractors — with the "highest concentration" of these contract workers in Colombia and Mexico. A billion-dollar U.S. fintech whose extended workbench sits in Latin America and whose core staff would fit on a single office floor: you will not find that in any marketing deck, but it is in the mandatory filing.

Sezzle has been a Delaware public benefit corporation since June 2020 — a legal form whose board must, by charter, balance profits against a stated public benefit. Sezzle's declared mission per the annual report (10-K): "to financially empower the next generation."

The same reporting season delivers the counter-calculation: in fiscal year 2025, only $102.2 of $450.3 million in revenue (23 percent) came from merchants and partners. The rest — $348.1 million, or 77 percent — comes mostly from the consumers themselves: $131.9 million of consumer fees inside transaction income (prior year: $60.3 million), $99.4 million of subscription revenue and $116.8 million of income from other sources, per the filing "largely derived from consumer fee income," above all late payment fees. A year earlier, the consumer share stood at 68 percent. The mission is written into the charter — the margin into the fee schedule.

Almost unnoticed, NRC Health tidied up its charter in June 2026: per the 8-K, the annual meeting of June 23, 2026 removed the supermajority voting requirements from Article 6 of the certificate of incorporation, deleted the restrictions on removing directors without cause, and lowered the threshold for stockholder action by written consent from unanimity to the ordinary voting requirement.

For a company in which family trusts hold 46.8 percent of the shares, that is a remarkably shareholder-friendly direction — the new rules formally strengthen all shareholders, not just the largest. The amended charter ("Amended and Restated Certificate of Incorporation of NRC Health") took effect on June 24, 2026, and is the first to carry the company's new name.

Founder Michael D. Hays has transferred almost all of his shares to family trusts for estate-planning purposes. The ownership table in the proxy statement shows the result: the Common Property Trust holds 38.2 percent; together with further family vehicles (Amandla LLC, CPT LLC) the family bloc reaches 46.8 percent — and the table lists Patrick E. Beans as beneficial owner of 10,575,634 shares, or 46.9 percent.

The curious part sits in footnote 7: only 35,003 of those shares belong to Mr. Beans directly. The rest he controls as manager of the family LLCs, as trustee, and as "Special Holdings Direction Advisor and Protector" of the various Hays trusts — with direct voting and dispositive power. Anyone voting at NRC Health should know: effective control over nearly half the shares is bundled with one trustee whose own economic stake is tiny.

In June 2025, NRC Health brought in Trent Green, the former head of Amazon One Medical, as its new chief executive. The proxy statement puts his total 2025 package at $11,961,900 — $697,115 in salary, a $4,503,333 bonus and $6,755,000 in stock awards. For comparison: the company's entire net income for 2025 was $11.6 million. A single compensation package weighed as much as the whole company's annual profit.

The annual report (10-K) spells out the consequences: selling, general and administrative expenses rose $9.9 million in 2025, "primarily due to $6.6 million in bonuses related to our executive leadership transition, and $3.0 million in stock compensation related to new executive leadership compensation arrangements" — helping push the operating margin from 25 to 16 percent. The footnote-worthy contrast: founder and Chairman Michael D. Hays has drawn an unchanged $127,400 annual salary since 2005, per the same proxy.

Everspin's licensing and services revenue carried an item since early 2025 that you would not expect at a memory-chip minnow: a paid development contract "for the development of an AI technology application" — presumably MRAM-based; the filing gives no details. The agreement propped up the highly variable licensing line in 2025.

In the quarterly report (10-Q) as of March 31, 2026, it appears only as a farewell note: licensing and services revenue collapsed by 63.4 percent to $0.8 million, "primarily due to the conclusion of an agreement for the development of an AI technology application." The lesson for investors: Everspin's licensing, patent and services revenue ($6.9 million in 2025, 12.5 percent of total revenue) hangs on a handful of individual contracts — every expiry immediately tears a visible hole.

Since August 2024, Everspin has been collecting milestone payments from a "strategic award" for a long-term plan to provide manufacturing services for aerospace and defense segments — worth, per the annual report (10-K) for 2025, up to approximately $14.6 million over 2.5 years. Who provides the money is not stated; the filing says only that the award is "not in the ordinary course of the Company's business and hence not a contract with a customer" — it is booked as other income by analogy to the revenue rules.

That very line currently decides Everspin's sign: in 2024, $6.1 million of other income stood against a $7.1 million operating loss (net: +$0.8 million); in 2025 it was $4.4 million against −$6.5 million operating (net: −$0.6 million); in the first quarter of 2026, $2.2 million of award income against −$2.7 million operating (net: −$0.3 million). The 2.5-year span from August 2024 runs out in early 2027 — after that, the factory has to earn what the award currently contributes.

On May 11 and May 29, 2026 — in the middle of the price rally — Everspin director Lawrence G. Finch sold, per Form 4, 310,091 shares in three transactions at prices between $26.03 and $34.29, roughly $8.5 million in total. Holdings afterwards: zero shares. The filing shows no pre-arranged 10b5-1 trading plan; the shares were held in a family trust.

On July 14, 2026 came the second act: per Form 8-K, Finch declared his resignation from the board of directors, including the audit committee, effective August 4, 2026 — explicitly "not the result of any disagreement with the Company." All of it is legal and properly reported. But the sequence — first the complete exit near the high, then the departure from the oversight body — belongs on the list of things an investor should know before chasing the momentum.

Globe Life is the incarnation of the American heartland insurer — and yet the business description in the 10-K for 2025 carries a surprising half-sentence. The by far largest distribution organization, the American Income Life Division, counts "11,920 average producing agents in the U.S., Canada, and New Zealand."

Canada and New Zealand appear nowhere else in the investment story — American Income grew historically by serving union members and working families, and the company quietly exported that model to two more countries. A small geographic second life that hardly anyone expects when hearing "insurer for middle America."

Tucked into the risk factors of the annual report (10-K) for 2025 is a capital move hardly any investor has on the radar: through an intercompany reinsurance agreement initiated in 2025, Globe Life ceded "approximately $1.2 billion of our life statutory reserves" from Liberty National, Globe Life And Accident and American Income to a subsidiary called GL Re (as of December 31, 2025).

Such intercompany reinsurance is common in the industry and serves capital and reserve management (statutory reserves are allocated more efficiently). But it is a good example of how much moves behind an insurer's corporate facade without anything changing in its outward appearance — $1.2 billion of reserves migrate from the left pocket to the right pocket of the same group.

Hardly any company names the sender of an attack — Globe Life does. In the legal-proceedings footnote of the annual report (10-K) for 2024, the allegations behind the shareholder derivative suits "derive from the Fuzzy Panda short seller report." On April 11, 2024, the short seller Fuzzy Panda published a report attacking the sales practices of the independent agents — the stock dropped in its wake.

A year later, the 2025 annual report no longer names the author, referring neutrally to "certain short seller reports." Anyone who wants to know who set the ball rolling finds it in black and white only in the prior-year report.

How jumpy a refiner's earnings are is captured by two first quarters side by side: in the first quarter of 2025, HF Sinclair earned practically nothing — diluted EPS was minus $0.02. One year later, in the first quarter of 2026, it jumped to $3.56 per share ($648 million of net income) — more than all of fiscal year 2025 ($3.08).

That number is both an opportunity and a warning: it shows how quickly a turning refining margin multiplies earnings — and how little a single strong quarter says about earning power "through the cycle." Whoever annualizes a Q1 2026 makes the same mistake as someone who extrapolated the future from the record year 2022.

In the segment table of the annual report (10-K) for fiscal year 2025 sits a number you would not expect from a "green" business of the future: the Renewables segment (renewable diesel) reports a negative gross margin — $991 million of revenue against $1,027 million of cost of sales, for minus $129 million at the gross level alone. And it is no fluke: gross margin was also red in 2024 (minus $86 million) and 2023 (minus $128 million).

The reason is a margin that depends heavily on government incentives: renewable diesel only pays off as long as the sale price plus federal and state low-carbon fuel incentives exceeds the expensive feedstocks (such as soybean oil). Per the annual report, the 2025 U.S. law OBBBA curtailed exactly those green funding programs — a segment of the future that hangs not on demand, but on subsidy policy.

Anyone looking up HF Sinclair on the New York Stock Exchange types in a symbol unusually playful for an oil company: DINO. Behind it lies no modern marketing gag, but the 1930s: the Sinclair fuel brand has advertised for generations with "Dino," a green apatosaurus — the idea being that crude oil dates back to the age of the dinosaurs. When HollyFrontier acquired Sinclair Oil in 2022 and formed the new holding company, the dinosaur wandered from the billboards into the stock symbol.

That is more than a footnote: the Marketing segment with the Sinclair gas stations is, per the annual report (10-K) for fiscal year 2025, the only part of the company whose operating income has grown steadily over the past three years — from $37 million through $48 million to $73 million. The green dinosaur is thus, of all things, the smallest but most reliable workhorse in an otherwise highly cyclical company.

Mercury General was a reliable dividend payer for decades — per the annual report (10-K) for fiscal year 2025, the company paid cash dividends continuously since its November 1985 public offering and raised the dividend per share year after year — until 2022. Then the streak broke: "As a result of challenging business conditions, the Company reduced the dividend per share during 2022 for the first time since 1985."

The trigger: in 2022 Mercury plunged into a record loss of $512.7 million with a combined ratio of 109.5 percent, as auto-repair and parts costs rose faster than regulator-approved premiums. Since then the quarterly dividend has stood at $0.3175 (declared February 13, 2026) — the company pays out roughly $70 million a year. A 37-year record hardly any investor had on the clock ended quietly in a single line of a filing.

Subrogation — the right to recover the cost of a claim from whoever caused it — is normally a matter for the legal department. On the Palisades fire, Mercury took a different route: per Note 12 of the annual report (10-K) for fiscal year 2025, the company sold its Palisades-fire subrogation rights to a third party in June 2025 — for a guaranteed percentage of losses of roughly $48 million plus a possible "Upside Recovery" above a threshold.

That turns a fire loss into a tradeable financial product: Mercury swapped an uncertain, years-long litigation prospect for an immediate, booked amount — and handed the litigation risk to a financial investor. On the larger Eaton fire the company chose the opposite path, pursuing about $538 million of subrogation itself against utility Southern California Edison. That a California auto insurer showcased two opposite strategies for handling fire subrogation in a single year is one of the more curious footnotes of 2025.

Anyone picturing a down-to-earth Los Angeles auto insurer when they read "Mercury General" skims past a curious line in the human-capital section of the annual report (10-K) for fiscal year 2025: alongside roughly 4,300 employees in the United States, the company runs a technology subsidiary called Mercury Shanghai with "approximately 80 employees working at a leased office space in Shanghai, China."

Fittingly, in 2023 Mercury hired a Chief Data & Analytics Officer who had previously co-founded a data-analytics company for eight years and led business analytics at LinkedIn. So a family business that has been writing California auto policies since 1961 quietly operates a small data workshop on the other side of the world — something hardly any investor has on the radar when they hear "California auto insurer."

Since the inception of its repurchase program, The Cheesecake Factory has bought back a total of 59.9 million of its own shares for roughly $1,983 million — nearly two billion dollars — per the annual report (10-K) for fiscal year 2025. For comparison: the company’s reported equity as of December 30, 2025, was just $436.4 million.

The two only seem to clash at first glance: repurchased shares sit as treasury stock reducing equity, so a company that profitably retires shares for years can show optically thin book equity and still be healthy. It does explain why the price-to-book ratio sits around 8.7 (data as of July 17, 2026): the book value has been bought away. In fiscal year 2025 alone, another $153.9 million went to buybacks and $55.2 million to dividends — capital return is not a sideshow at CAKE, it is policy.

Comparable sales at the core Cheesecake Factory brand rose a scant 0.1 percent in fiscal year 2025 — that is $2.2 million on a brand doing billions. It gets interesting when you unpack that zero: per the annual report (10-K), a 4.3 percent increase in menu pricing lifted the average check while guest traffic fell 1.9 percent.

Put differently: the flagship holds its revenue because the remaining guests pay more, not because more guests come. For a restaurant chain that is the most uncomfortable of all metrics — because price increases have a ceiling, while traffic erosion rarely has a visible floor. Hear "record revenue" and you picture full tables; the footnote describes slightly emptier tables at higher prices.

Read only the headline and you see a record year: The Cheesecake Factory’s revenue rose 4.7 percent to $3,751.8 million in fiscal year 2025, and income from operations climbed from $178.3 million to $187.3 million. Yet net income fell from $156.8 million to $148.4 million. The reason sits two lines lower in the income statement: a "loss on extinguishment of debt" of $15.9 million.

In February 2025 the company issued new convertible senior notes of $575.0 million maturing in 2030 and used them to essentially retire the old notes due 2026. Such refinancings are routine — but they are not free: the book loss on early retirement landed straight in earnings and pushed the net margin from 4.4 to 4.0 percent. Higher impairments ($23.0 million versus $13.6 million) added to it. A lesson that "revenue at a record" and "profit at a record" are two different sentences.

Anyone who thinks of AAR as a pure metal-and-oil business overlooks a subsidiary hardly anyone has on the radar: Trax, a maintenance-software provider for airlines and repair shops acquired in 2023. The annual report (10-K) describes it as a "leading provider of maintenance software for airlines, other aircraft operators and MROs" — with apps "that are in process of automating MRO workflows with artificial intelligence."

This is not an AI story to sell, but a quiet efficiency lever in AAR’s own toolbox — a software offshoot inside a company the market values almost entirely as a parts and repair business. That is exactly why Trax anchors our company-specific AI rating: AAR uses AI operationally but does not (yet) sell it as a product of its own.

Settlements with regulators cost money — but rarely as visibly as at AAR. The $55.6 million settlement with the U.S. Department of Justice and the SEC over violations of the U.S. anti-corruption law, the FCPA, did not just depress pre-tax profit in fiscal year 2025. It struck a second time, because fines are generally not tax-deductible: "In fiscal 2025, our effective income tax rate was 67.9% as the majority of the FCPA settlement charge was nondeductible for income tax purposes resulting in no income tax benefit."

For context: the regular U.S. federal rate is 21 percent. That a profitable industrial company hands nearly 68 percent of its pre-tax profit to the tax authorities is the quiet extra penalty of a corruption case — the amount sits in the income statement, but the tax penalty hides in a footnote.

The third quarter of fiscal year 2026 looked like the big breakthrough: $68.0 million of net income after a $8.9 million loss in the prior-year quarter. But read the quarterly report (10-Q) line by line and, just above the interest line, you find two items no investor keeps in mind: a $35.7 million bargain purchase gain and a $9.8 million gain on the sale of the headquarters building.

A bargain purchase gain arises when a buyer pays less for a business than its assets are worth net of liabilities — a book gain with no cash coming in. AAR states the reason plainly: "We believe the acquisition resulted in a bargain purchase gain as the seller was highly motivated to divest the business." The seller was HAECO Americas, acquired on November 3, 2025. Together with the building sale, roughly $45 million of the $93.1 million pre-tax income came from one-off items — fantasy profits for the quarter, but no blueprint for the next one.

The property notes of the annual report contain a sentence easy to skim past: as of June 30, 2025 and 2024, Twin Disc owned one permanently idle facility with a book value of $3.0 million. The plant produces nothing but stays on the balance sheet.

On its own the amount is small. As a footnote, though, it fits the bigger picture of a 108-year-old industrial company with sprawling, widely distributed fixed assets — plants in the United States, Belgium, Canada, Finland, Italy, the Netherlands and Switzerland. Anyone reasoning about asset value should know that part of it no longer works.

Five days before this research, on June 30, 2026, Twin Disc signed a new credit agreement that replaces the February 2025 facility: a $30 million term loan (maturing June 30, 2031) plus a revolving line of up to $60 million. Alongside Bank of Montreal, JPMorgan Chase now joins as a second lender. The freshly acquired Canadian subsidiary Kobelt serves as a guarantor.

The price of that flexibility sits in the fine print of the 8-K: the loan is secured by substantially all of Twin Disc's and Kobelt's personal property — receivables, inventory, machinery, intellectual property — and the company additionally pledged 65 percent of its equity interests in certain foreign subsidiaries. For a company whose entire profit currently comes from abroad, that is a notable amount of collateral handed to the banks.

When Twin Disc acquired the Finnish gearbox and drivetrain specialist Katsa Oy on May 31, 2024, something rare in M&A happened: the fair value of the acquired net assets ($29.608 million) was higher than the purchase price of $25.884 million. Under the accounting rules, Twin Disc recorded the $3.724 million difference as a "gain on bargain purchase" — income from buying below value.

That paper gain flowed straight into earnings and lifted fiscal year 2024 net income (of $11.0 million attributable to Twin Disc) by more than a third. It is not genuine operating profit but a purchase-accounting effect — and a quiet hint that a buyer smelled a bargain. Investors comparing Twin Disc's earnings line across the years should strip out this one-off.

In its annual report (10-K) for fiscal year 2025 Jabil advertises its own "procurement intelligence platforms and AI-driven orchestration systems" — AI-assisted procurement and orchestration platforms for its own supply chain. Fittingly, the company had acquired the procurement advisory firm ProcureAbility in November 2023. Two years later the same filing records the ending: Jabil sold ProcureAbility during fiscal year 2025 for roughly $60 million in cash — and booked a $97 million pre-tax loss in the process.

An acquisition offloaded at a clear loss within two years is, by itself, no catastrophe. What is notable is the simultaneity: while one half of the company markets "AI-driven" procurement intelligence as a capability, the other half parted, at a loss, with exactly the procurement know-how it had bought. A small reminder that between the AI narrative and the capital allocation, worlds can lie even at a $30 billion company.

A sentence in the footnotes of the annual report (10-K) for fiscal year 2025 is one hardly any investor has on the radar: in fiscal year 2025 Jabil sold $11.4 billion of trade receivables under receivable-sale programs and received $11.3 billion in cash for them. The sold receivables disappeared from the balance sheet, and the cash inflow ran through operating cash flow.

Factoring — selling open invoices to third parties to get paid earlier — is common in contract manufacturing; what is unusual is the sheer scale. The amount sold but not yet collected at the balance-sheet date, where Jabil retains "continuing involvement" and thus risk, rose within a year from $367 million to $927 million. Anyone admiring Jabil's lean balance sheet should know: part of that leanness comes from billions of receivables never appearing on the balance sheet at all.

The name GEO Group stands for buildings with fences. But the second leg is digital: through its subsidiary BI Incorporated, the company runs electronic monitoring for ICE (the ISAP program, renewed effective October 1, 2025) — ankle monitors, GPS, case management. On December 22, 2025, a new piece was added that hardly any investor has on the radar: BI was awarded an ICE contract for "skip tracing" services.

The annual report describes it dryly as "enhanced location research with identifiable information, commercial data verification, and physical observation" — that is, determining the current whereabouts of people in immigration proceedings through data matching and physical observation. GEO is therefore not only a landlord of beds but also a provider of people-finding services — a second business that serves the same political demand as the detention beds and raises the same reputational and ESG questions.

In its Secure Services segment, GEO reported it was marketing 5,896 idle beds with a net book value of roughly $180.9 million at six shuttered facilities as of December 31, 2025; the reentry segment adds another 750 empty beds. That sounds like ballast — but against the backdrop of the OBBBA windfall it is an option: idle, already-depreciated capacity can be reactivated for ICE faster and more cheaply than a new build.

Exactly that played out in June 2025, when GEO activated its company-owned, 1,868-bed D. Ray James facility in Folkston, Georgia, via a contract modification with ICE. For investors the quiet reserve is therefore double-edged: in the upswing it is leverage on detention demand — but if the political wind turns, it is $181 million of book value that produces no rent and whose sale, per the report, could occur below carrying value.

It is well known that several major banks pledged after 2019 not to finance private detention operators. Less known is a side effect the annual report (10-K) for fiscal year 2025 records in a single sentence: some of those same institutions did not merely stop lending — they also ended their equity research coverage of GEO: "Some of these same institutions have ceased their equity analyst coverage of our company."

That is a double withdrawal: refusing both to finance a stock and to cover it narrows refinancing and capital-market visibility at once. For investors it is a quiet quality filter of an unusual kind — it is not the balance sheet that drives the analysts away, but the industry. Which is exactly why, per fundamental data (as of July 17, 2026), only a handful of analysts still follow GEO; the thin coverage is no accident but a documented consequence of the ESG exclusions.

Indivior was created in 2014 as a spin-off of the British consumer group Reckitt Benckiser, incorporated under English law as "Indivior PLC" and long listed on the London Stock Exchange. The annual report (10-K) for 2025 now documents the complete farewell: after market close on January 23, 2026, a court-approved scheme of arrangement made the new Indivior Pharmaceuticals, Inc. (Delaware) the parent company; Indivior PLC became the subsidiary "Indivior Limited".

The punchline sits a few pages further on: as part of streamlining its international business, Indivior is discontinuing the sale of its products in the United Kingdom of all places — alongside Ireland, Sweden, Israel, Finland and Italy. Canada, Australia and France continue; Germany is served without local operations. Only the active-ingredient plant in Hull (which makes the buprenorphine for SUBLOCADE and SUBOXONE) and support functions in Slough remain British. A company with a British birth certificate that will file as an American, report as an American — and no longer sell a single product in the country it came from.

The quarterly report as of March 31, 2026 says it plainly: current liabilities exceed current assets by $111 million, and total liabilities exceed total assets by $144 million — Indivior has negative stockholders' equity, a "stockholders' deficit" in U.S. GAAP terms. At year-end 2025 the deficit was $98 million; the first quarter's buybacks ($126 million including costs) deepened the hole again.

What stands out is how calmly the report explains it: the negative working capital reflects the timing of rebate payments to insurers relative to the collection of receivables — the line "Accrued rebates and product returns" stood at $582 million at the end of 2025, by far the largest item on the liabilities side. For investors this means two things: the price-to-book ratio is meaningless for this stock (technically in the triple digits), and the company's resilience rests not on equity but entirely on the reliability of SUBLOCADE's future cash flow.

In the first quarter of 2026, Indivior stacked two capital moves you rarely see in the same report: the company repurchased 3,974,153 of its own shares at an average of $31.45 ($125 million out of the $400 million program announced in February 2026) — and in March 2026 simultaneously placed a $500 million convertible senior note with a coupon of just 0.625 percent and a conversion price of $41.66.

Translated: shares are collected at the bottom while an exchange right is sold at the top. If the stock holds above $41.66, the repurchased shares — and more — can eventually re-enter the market through conversion; the quarterly report already applies the if-converted method to diluted share counts (129 instead of 124 million shares in the denominator once the stock trades above the conversion price). The proceeds, by the way, repaid the old $333 million term loan in full — including an $18 million loss on debt extinguishment in the quarterly income statement.

American Bitcoin did not reach the Nasdaq in September 2025 through its own IPO but through the shell of an existing company: per Hut 8's annual report, the subsidiary was merged with Gryphon Digital Mining in a stock-for-stock transaction — American Bitcoin's owners received roughly 98 percent of Gryphon's shares, after which Gryphon was renamed "American Bitcoin Corp." and the ticker changed to ABTC.

Tracing that shell's history in the SEC's EDGAR database (CIK 1755953) turns up a curiosity: the vehicle previously operated as MTech Acquisition Holdings (an acquisition vehicle, 2018), then as Akerna Corp. — a maker of compliance software for the cannabis industry —, then from February 2024 as Bitcoin miner Gryphon Digital Mining, and since September 2025 as American Bitcoin. Not decisive for the valuation; quite useful for understanding: the "new" Bitcoin platform lives inside a stock-market shell that changed its name and business four times in seven years.

American Bitcoin presents itself in the annual report as a "Bitcoin accumulation platform" — the stated goal is more Bitcoin per share. The footnotes of the quarterly report (10-Q) as of March 31, 2026, show the other side of that coin: to buy new mining machines from manufacturer Bitmain, the company pledges its Bitcoin. In October 2025 it replaced a $46.0 million cash deposit with a pledge of 391 Bitcoin; in February 2026 it pledged roughly 314 Bitcoin covering 80 percent of the purchase price of about 11,298 S21 XP miners. As of March 31, 2026, a total of 3,090 Bitcoin sat as collateral with Bitmain — nearly a fifth of the group's entire holdings of 16,331.

The construction has a catch that the filing records dryly: if American Bitcoin does not redeem the pledged Bitcoin within the agreed window at a "mutually agreed upon price," the redemption right lapses — the Bitcoin then belong to the machine maker. Economically, "we accumulate Bitcoin" thus becomes "we trade Bitcoin for machines, with a buy-back option." In the first quarter of 2026, the group already derecognized Bitcoin with a carrying value of $81.2 million to settle miner purchase liabilities.

The related-party footnote of the annual report (10-K) for 2025 contains a transaction you need to read twice: on December 30, 2025, the boards of directors of Hut 8 and American Bitcoin consented to the purchase of 23,199,205 Class B shares of the mining subsidiary American Bitcoin by a limited liability company (the filing calls it the "LLC Purchaser") — purchase price: $1.40 per share, roughly $32.5 million in total.

The piquant detail: per the filing, that LLC is solely managed and controlled by Asher Genoot and Michael Ho — Genoot is CEO of Hut 8 and Executive Chairman of American Bitcoin, Ho is Chief Strategy Officer of Hut 8 and CEO of American Bitcoin. Both sit on both boards that consented to the deal, and per the 10-K they hold an indirect financial interest in the purchasing LLC. Formally, everything is disclosed and approved. But if you buy Hut 8 stock for its American Bitcoin stake, you should know: when the subsidiary's shares are handed out, the company's own chiefs sit at the table — on both sides.

Cohu’s restructuring program of February 2025 — consolidating the sites in La Chaux-de-Fonds, Switzerland, and Kolbermoor, Germany, into lower-cost regions — cost roughly $10.1 million in fiscal 2025. But the footnote holds an item with the opposite sign: because the Swiss headcount reductions cut off future entitlements under the local pension plan, the company recorded pension curtailment gains of $2.2 million — income that follows directly from the terminations.

In accounting terms this is correct (the pension obligation shrinks when entitlements lapse), but it remains a remarkable mechanism: the same overhaul that triggered $8.3 million of severance also produced a book gain out of the retirement plans of those affected. Whoever reads only the headline restructuring number misses that part of it was counter-financed by the pension fund of the dismissed.

The risk section of Cohu’s annual report (10-K) for 2025 contains a passage you would sooner expect from a software giant: an analysis of the presidential executive order "Ensuring a National Policy Framework for Artificial Intelligence", signed on December 11, 2025 — including the Department of Justice "AI Litigation Task Force" it establishes, tasked with suing states over their AI laws. Cohu concedes it is "not yet clear" on what legal grounds that task force may challenge state-level AI regulation.

The reason for the unusual passage: with the acquisition of AI software provider Tignis (January 2025, $36.6 million), the test-equipment maker brought — in the report’s words — "new compliance requirements for data privacy and algorithmic transparency" into the house. A hardware cyclical poring over Washington AI regulation in its risk chapter — a curious testament to how quickly a $36 million acquisition becomes a new field of law.

When Cohu placed its $287.5 million convertible note in September 2025, it spent $31.4 million on so-called capped call options — an insurance policy meant to soften the dilution of existing shareholders if the note is one day converted into shares. The fine print in the annual report (10-K) for 2025: this insurance only works up to an "initial cap price" of approximately $41.02 per share — double the then-current stock price of $20.51.

What looked like a generously sized ceiling in September 2025, the market took out within months: the stock traded near $67 in mid-July 2026 (data as of July 17, 2026) — a good 60 percent above the cap. Translated: dilution is hedged for the stretch from $27.18 (the conversion price) to $41.02, and unhedged for everything above it. The better the stock performs, the larger the unprotected part of the bill becomes — an irony hardly any investor has on the radar who only sees the rally.

The receivables footnote of the annual report (10-K) for 2025 contains a data series hardly anyone reads: the allowance for credit losses grew from $4.8 million at the end of 2023 to $18.3 million at the end of 2024 and $32.8 million at the end of 2025 — with $14.5 million newly added in 2025 alone, after $13.5 million the year before.

The series turns explosive in combination with two other disclosures: four customers accounted for 62 percent of outstanding receivables at the end of 2025, and 99.6 percent of revenue is earned in mainland China, where chip fabs often pay for their tools in full only after final acceptance. When an equipment maker has to set aside ever more for potential payment defaults while its receivables clump at a handful of customers, that is an early-warning indicator the price chart does not show.

ACM Research is incorporated in Delaware, headquartered in Fremont, California, and listed on the Nasdaq. The staffing table in the annual report (10-K) for 2025 tells you where the company actually lives: of 2,513 full-time employees as of December 31, 2025, 2,357 worked in mainland China and the Taiwan region, 127 in Korea — and exactly 29 in the United States. That is about 1.2 percent of the workforce.

The division of labor is equally clear: per the report, the Fremont headquarters hosts "strategic planning, marketing and finance activities"; development and production sit in Shanghai (including the Lingang production center). Of the 1,228 research-and-development employees, the overwhelming majority is based in Asia. For investors this is not a footnote but the core question of the stock: the Nasdaq ticker ACMR is the packaging — the contents are a Chinese equipment maker.

In September 2025, ACM Shanghai, the operating subsidiary of ACM Research, placed 38,601,326 new shares at 116.11 yuan apiece with investors in mainland China — net proceeds of roughly $623.0 million. The money did not flow to the Nasdaq holding but to the subsidiary; per the annual report (10-K) for 2025, such proceeds are "generally … not available" for distribution to ACM Research.

The price of the capital injection: ACM Research's stake in ACM Shanghai fell from 81.5 to 74.6 percent. The dilution staircase since 2019: 100 → 91.7 percent (pre-IPO placements), → 82.5 percent (STAR IPO 2021), → 81.5 percent (option exercises), → 74.6 percent (private offering 2025). Whoever holds ACMR shares has watched their claim on the operating business shrink for years — without the count of their own shares changing at all. A quarter of the group's profit now belongs to the minority shareholders in Shanghai: in 2025, $27.8 of $121.9 million in net income was attributable to them.

In 2023, Blue Bird founded the 50/50 joint venture Clean Bus Solutions with infrastructure investor Generate Capital: school districts would no longer have to buy electric buses and charging infrastructure but rent them as a turnkey service — "fleet as a service." The idea sounded like the answer to the electric story’s core question: how are cash-strapped school districts supposed to pay for buses that cost a multiple of a diesel?

The market’s answer was sobering. Per the quarterly report (10-Q) as of March 28, 2026, the joint venture was unable to generate business on a timeline likely to produce profitable returns: in the fourth quarter of fiscal 2025, Blue Bird wrote its stake down by $7.4 million to zero; in October 2025 both partners voted to dissolve the venture, and by the end of 2025 the wind-down was largely complete. What remains is a finding that appears in no e-mobility deck: without subsidy money, the U.S. school market so far barely buys (or rents) electric buses.

Whoever looks at Blue Bird’s quarterly numbers in the fall of 2026 should know this footnote: the company is winding down its defined benefit pension plan — in April 2026, first benefits of $13.0 million were settled via lump-sum payments, with the remainder to be transferred to a group annuity insurer or the federal pension insurer PBGC, all funded from plan assets.

The catch is an accounting effect: under U.S. GAAP (ASC 715), a plan settlement forces the entire actuarial loss parked in equity through the income statement in one go. Per the quarterly report (10-Q) as of March 28, 2026, that is roughly $28.1 million — non-cash, but earnings-effective in the third quarter of fiscal 2026. Whoever then reads a headline about a "profit collapse" will know: no new money vanished, it had merely been hiding in equity for years.

With the closing of the Micro Bird acquisition on April 1, 2026, the Canadian Girardin family reached a remarkable accumulation of roles at Blue Bird. It sold the second half of the small-bus joint venture for $201.8 million — 70 percent paid in stock —, has since held 7.88 percent of Blue Bird per Schedule 13D, and Steve Girardin joined the board as a director (term through 2029, with a successor clause for his brother Dave). A board election agreement obliges the family holding to vote all shares in line with the board’s recommendations.

The real surprise sits in the amended 8-K/A of May 4, 2026: through its company Superbird Capital, the family is at the same time an authorized Blue Bird dealer — with roughly $205 million in aggregate gross revenues in fiscal 2025 and roughly $146 million in the first half of fiscal 2026. That equals about 14 percent of consolidated revenue in fiscal 2025 — and a good fifth in the first half of 2026. And through its real estate firm Valiant, the family leases Micro Bird sites for roughly $3 million in annual rent to the company that now owns those operations. All disclosed, approved by the audit committee, declared arm’s length — but whoever buys the stock should know that supply, distribution, rent, ownership and a board seat here partly run through the same hands.

When AZZ acquired the coil coater Precoat Metals for $1.28 billion in 2022, the investment firm Blackstone (via BTO Pegasus Holdings) helped with $240 million of convertible preferred stock carrying a 6 percent dividend. The exit, recorded in the annual report (10-K) for fiscal year 2025: on May 9, 2024, AZZ redeemed the 240,000 preferred shares for $308.9 million — calculated as face value times a contractual "Return Factor" of 1.4, less the $27.1 million of dividends paid to date. The $75.2 million redemption premium was recorded as a deemed dividend and cut fiscal 2025 earnings available to common shareholders by $2.50 per share.

The redemption was funded by an equity raise: in April 2024, AZZ sold 4.6 million new shares at $70.00 ($322.0 million gross). A good two years later the same stock trades around $160 (data as of July 10, 2026) — Blackstone locked in its 40 percent total return by contract, while common shareholders carried the dilution right before the share price doubled. Expensive bridge financing, cleanly disclosed — and a lesson in what "flexible acquisition financing" ultimately costs.

The very footnote of the annual report (10-K) for fiscal year 2026 that explains AZZ's record profit — a $261.8 million proportionate gain from the partial sales of the AVAIL joint venture — contains a paragraph hardly any investor has on the radar: AVAIL recorded a prior period adjustment for accounting errors within its Brazil operations. AZZ's 40 percent share of the adjustment: approximately $9.6 million, booked in the fourth quarter of fiscal 2026 — $8.4 million of it relating to prior periods. After its own analysis, management concluded the adjustment was not material to previously issued financial statements.

The piquant part is the setup: AZZ is a minority holder in AVAIL without decision-making authority — the risk section of the same report explicitly warns of the "lack of sole decision-making authority." The stake that delivered two-thirds of the group's net income in fiscal 2026 thus also brought a $45.9 million impairment, a transfer pricing change and accounting errors in Brazil in the very same year — and after all of it still sits on the books at $20.0 million.

Host’s income statement carries a line that turns the portfolio’s weather exposure into numbers — as income: "Net gain on insurance settlements". After Hurricane Ian (2022), the Maui wildfires (2023) and Hurricanes Helene and Milton (2024), Host booked net insurance gains of $86 million (2023), $110 million (2024) and $24 million (2025) — amounts on the order of up to a seventh of the respective year’s net income.

For fairness: these gains are no trick but the accounting flip side of real damage — insurers reimbursed more than the destroyed assets were carried at, plus business-interruption compensation. But they make three years of GAAP earnings harder to read: per the annual report, the $86 million decline in insurance gains was a main reason the 2025 operating margin fell — operating profit dropped even though the hotel business grew. Whoever values Host on net income is always also valuing last year’s claims settlements.

The most spectacular sale of the year — the Four Seasons Resort Orlando at Walt Disney World Resort and the Four Seasons Jackson Hole for a combined $1.1 billion — carries a footnote hardly any investor is likely to have on the radar: the buyer, BDT & MSD Partners, is related to Host’s C-suite. The annual report (10-K) for 2025 states verbatim: "Teddy Overton, stepson of our Chief Executive Officer, James Risoleo, is a Principal at BDT & MSD Partners and worked on the transaction on behalf of BDT & MSD Partners."

Fairness requires the rest of the passage: Host explicitly classified the sale as a related-party transaction, CEO Risoleo did not participate in the negotiations per the report, and the board of directors reviewed and approved the deal — and a $1.1 billion price with a $242 million book gain hardly suggests a friends-and-family discount. Still, it is remarkable: the largest single deal of the year, the one that doubled earnings per share, was struck with the employer of the CEO’s stepson — and lives in a footnote of the consolidated accounts.

Between room rates and banquet revenues, a line item appears in Host’s accounts since 2025 that you would not expect from a lodging REIT: condominium sales. On a five-acre parcel right next to the Four Seasons Resort Orlando at Walt Disney World Resort, Host builds Four Seasons-branded condominiums and villas — and sells them. In fiscal 2025, 16 units sold brought in $99 million of revenue and $17 million of net income — an average of a good $6 million per unit. The first quarter of 2026 added another $26 million of revenue, and per the quarterly report, completion of 40 villas was slated for June 2026.

The curious part: Host sold the resort itself in that same quarter, as part of a $1.1 billion package to BDT & MSD Partners — but keeps building and selling the residences next door. A small, high-margin second business that shows up in no RevPAR statistic and will keep distorting quarterly comparisons.

Host Hotels & Resorts owns 76 hotels with roughly 41,700 rooms — including properties with well over 1,000 rooms and names like Ritz-Carlton, Four Seasons and Grand Hyatt. And according to the annual report (10-K) for 2025, the company itself employs exactly 162 people, all in the United States, with an average tenure of 14 years. None of them works a front desk: the report states plainly that Host neither employs nor manages its hotels’ workforce — "We do not directly employ or manage employees at our consolidated hotels" — with a single exception: at the three hotels in Brazil, Host is formally the employer but likewise leaves direction and supervision to the managers.

That is not an accident but the business model: as a REIT, Host leases its properties to its own taxable subsidiaries and has them operated by Marriott, Hyatt and other brands. Tens of thousands of people work in hotels Host owns — none of them is on Host’s payroll. For investors, that means: you are buying a property owner with a head office, not a hotel operator.

NVE's annual report (10-K) for fiscal year 2026 contains a number you have to read twice: as of March 31, 2026, the company had exactly 47 shareholders of record — plus "several thousand" beneficial holders whose shares sit in street name at banks and brokers. Shares outstanding: 4,837,166. Shares authorized: just 6,000,000 — the board could not issue more without amending the articles of incorporation. The share count has been effectively frozen for years: not a single new share was issued in fiscal 2026 and no option was exercised; only 40,000 options are outstanding, and total stock-based compensation expense was $93,663.

For investors this miniature structure cuts both ways: zero dilution — your slice of the pie stays the same size — but also a thin market in which small orders can move the price (the stock's average daily range is around 7 to 8 percent per the scanner, data as of July 17, 2026). A curiosity on the side: the buyback program in place since 2009, $7.5 million in total authorizations, has sat untouched for years — "dividends are a more efficient method," NVE writes.

NVE's effective tax rate fell to 14.7 percent in fiscal 2026 (ended March 31, 2026) — well below the 21 percent U.S. statutory rate. The main driver is spelled out in the annual report (10-K): the tax provision included $1,067,993 of advanced manufacturing investment tax credits — CHIPS-era semiconductor manufacturing credits NVE claimed for expanding its production in Eden Prairie. Together with R&D credits, they cut the rate by 6.4 percentage points (prior year: 1.1).

The report itself says this tailwind is ending: with the production expansion complete and equipment purchases expected to "decrease significantly," NVE expects the credits to shrink "significantly" in fiscal 2027. Anyone extrapolating the $15.2 million net income should know that roughly a million of it was a tax effect, not business — no small footnote for a company whose dividend already exceeds its earnings.

On December 17, 2025, NVE filed a document with the SEC whose header looks like a typo: a Form 8-K/A ("Amendment No. 11") — a retroactively amended current report — with an event date of January 1, 2006. The explanation: the most important customer contract in company history, the Supplier Partnering Agreement with pacemaker maker Pacesetter (then St. Jude Medical, now part of Abbott Laboratories), was signed on January 3, 2006, and has been renegotiated twelve times since. Instead of opening a new report each time, NVE has been extending the same 8-K for two decades — amendment by amendment.

The latest Amendment No. 12, dated December 12, 2025, extends the contract through December 31, 2027, and raises prices: "Amendment No. 12 to the Supplier Partnering Agreement was executed on December 12, 2025 extending the Agreement term through December 31, 2027 and increasing pricing for 2026 and 2027." Given that 37 percent of company revenue now hangs on this perpetually amended document (annual report 10-K for FY 2026, Note 10), the curious filing history is arguably the most important footnote of the entire stock.

Plexus reported net income of $172.9 million for fiscal year 2025 — up 54.6 percent from $111.8 million the year before. The headline reads "record profit," and operationally there is something to it: operating income rose 20.7 percent and operating margin climbed 80 basis points to 5.0 percent. But a second, quieter driver sits further down the income statement.

Interest expense fell from $31.5 million (FY 2023) to $28.9 million (FY 2024) to $11.6 million (FY 2025) — a relief of a good $17 million versus the prior year alone, landing one-for-one in pre-tax income. Plexus paid down debt over that period, which is healthy. But interest savings are a one-time tailwind that cannot repeat every year — unlike operating margin. Anyone extrapolating the 55 percent earnings jump into the future should back the quiet helper out first.

Buried in the notes of the annual report (10-K) for fiscal year 2025 is a line hardly any investor has on the radar: Plexus sold $705.0 million of trade accounts receivable in fiscal year 2025 (prior year: $854.7 million) to two bank programs (MUFG and HSBC) and received $698.1 million in cash for them. On roughly $4 billion of revenue, that is nearly one in six invoice dollars that comes not from the customer but, in advance, from a bank.

Accounting-wise the maneuver is elegant: the sold receivables disappear from "Accounts receivable" and show up, per the report, as "cash provided by operating activities" — making operating cash flow look better than it would without factoring. The price for it, the sale discount, lands quietly in the "Miscellaneous, net" line. None of this is illegal or hidden; receivable sale programs are a common tool in the low-margin contract business for turning working capital. But whoever admires Plexus’ cash-flow strength should know that part of it is time borrowed from the bank.

Plexus is a Wisconsin corporation headquartered in Neenah, and to many investors "U.S. contract manufacturer for defense and medical devices" sounds like American value creation. The risk factors of the annual report (10-K) for fiscal year 2025 correct that picture: "Operations outside of the U.S. in the aggregate represent a majority of our net sales and operating income, with a particular concentration in Malaysia" — the majority of revenue and operating income is earned outside the United States, with a particular concentration in Malaysia.

The numbers behind it: of $4.03 billion in revenue in fiscal year 2025, $2.39 billion came from the Asia-Pacific (APAC) segment and only $1.22 billion from the Americas region. And per the report’s human-capital disclosure, 58.3 percent of the more than 20,000 employees sit in the APAC region. Whoever buys a PLXS share is buying less a Midwest location than a manufacturing base in Southeast Asia — with everything a single country brings in currency, trade and geopolitical risk.

Anyone who wants to see how tightly Millicom and its controlling shareholder are intertwined finds the answer in a side note on expansion: in early 2026, a joint venture with NJJ — the holding company of Xavier Niel — acquired Telefónica Móviles Chile (closed February 10, 2026, roughly $50 million plus earn-outs of up to $150 million). In the Q1 2026 release, management commented that it had begun to "apply the Millicom playbook in Chile," led by NJJ.

That cuts both ways: on one hand, access to the Niel universe gives Millicom expansion opportunities an ordinary free-float company would not have. On the other, it blurs the line between the listed company and the owner's interests — the very line the annual report itself warns about when it lists Xavier Niel's influence as a risk factor. When the same circle that holds 42.2 percent of the voting shares also leads the foreign expansion, an investor should at least know whose playbook is being run here.

The risk section of the annual report (20-F) for 2025 contains a footnote you rarely see stated so plainly in a Latin American telecom: during fiscal year 2025, Millicom determined that the Bolivian boliviano (BOB) could no longer be freely exchanged into other currencies ("lacked exchangeability"), and therefore had to use an estimated exchange rate to convert its Bolivia operations.

The consequence is in the same passage: the estimated rate "has affected our results of operations in Bolivia," and Bolivia represents roughly 6 percent of total revenue. In the segment breakdown, that translates into a Bolivia revenue decline of about 41.9 percent year over year — a lesson that in emerging markets the business does not have to shrink for reported revenue to collapse: it is enough for the local currency to lose its convertibility. For a company whose costs are mostly in U.S. dollars while its revenue comes in local currencies, that is not a footnote curiosity but the structural background hum.

The annual report (20-F) for fiscal year 2025 records how deep the trough was in one dry clause: "No dividend distributions were made in 2023" — not a single dividend flowed in 2023. Two years later the picture is different: for 2025, Millicom paid out $5.50 per share in total — a regular dividend of $3.00 (in four quarterly installments of $0.75) plus a $2.50 special dividend in two tranches, approved by the board on August 6, 2025.

The timing is no accident: the special payout came in the same year the tower sale poured roughly $727 million of one-off gain into the books and equity free cash flow hit a record $916 million. For 2026, the board again proposes a regular dividend of $3.00 per share (payable July 2026 through April 2027). A company that jumps from "nothing at all" to "regular plus special" in two years earns both: credit for the discipline — and the question of how much of it comes from the running business and how much from the family silver.

Behind the clunky segment name "Performance Materials" hides a raw-material moat that hardly any investor associates with an Ohio materials company: per the annual report, Materion operates the world's largest bertrandite ore mine and refinery — "This segment operates the world's largest bertrandite ore mine and refinery, which is located in Utah, providing feedstock hydroxide for our beryllium businesses and external sale." From that ore Materion extracts beryllium hydroxide, the feedstock from which most beryllium-containing materials are made at the Elmore, Ohio plant.

That makes Materion vertically integrated from ore to finished component for aerospace, defense and energy — a rare setup that carries strategic weight in an age of supply-chain sovereignty and critical minerals. In 2025 the company invested $26.3 million in mine development alone, more than double the prior year ($12.2 million).

Beryllium is Materion's strategic treasure and its oldest liability at once: inhale beryllium dust and you can develop the incurable chronic beryllium disease (CBD) — "severe cases of CBD can cause disability or death." For decades Materion (and predecessor Brush) was therefore a defendant in personal-injury suits. All the more striking is the dry sentence in the annual report (10-K) for 2025: "As of December 31, 2025 there were no pending beryllium cases."

That is good news almost nobody has on the radar — and no free pass either: the report expressly warns that an unfavorable outcome or adverse media coverage could encourage new litigation, and that CBD concerns can depress demand for beryllium-containing products. The liability has not vanished; it is latent — it is sleeping.

Skim Materion's balance sheet and you see a materials company with $1.8 billion in revenue — and miss that a large share of the metal it processes never appears on the books at all. Materion works with consignment metal: precious metals, copper and nickel sit physically in its plants and get processed, but remain owned by the consignors, who charge fees for it. The notional value of this off-balance-sheet metal was $579.7 million as of April 3, 2026 (December 31, 2025: $526.2 million), per the quarterly report (10-Q).

The clever twist is a built-in vise: "The owners of the precious metals and copper charge a fee that fluctuates based on the market price of those metals" — the fee rises with the metal price. The very same price increase that visually inflates Materion's net sales also drives the consignment fees up; in the first quarter of 2026 that line item alone rose $3.9 million. A rising gold price is therefore not a pure tailwind for Materion — first it means higher costs and a larger off-balance-sheet exposure.

The preliminary proxy statement (PRE 14A, filed July 16, 2026) for the annual meeting on September 22, 2026, shows how directly executive pay tracks the steel cycle at Friedman Industries: President and CEO Michael J. Taylor received total compensation of $2,709,607 for fiscal year 2026 (ended March 31, 2026) — after $700,734 the year before, an increase of nearly four times, driven by a $1,221,635 bonus and $718,234 in stock awards in the very year the steel cycle pushed net income to $19.5 million. In the lean fiscal year 2025 (net income $6.1 million), his bonus had been just $21,635.

The footnote to the compensation table adds a charming detail rarely seen at a listed company: the bonus figures include "bonuses based on Company performance and Christmas bonuses, each of which is paid at the discretion of the Board of Directors" — no formula, no disclosed targets. For a shareholder of a cyclical company, that cuts both ways: pay that swings with the cycle is honest in its way, but fully discretionary bonuses at peak earnings are exactly the moment when boards tend to be at their most generous.

Concentration risk has many faces — at Ironwood (Nasdaq: IRWD), three of them meet at once. First, the product: per the annual report (10-K), revenue from the linaclotide partnerships makes up "substantially all" of total revenue. Second, the geography: in 2025, 97.7 percent of revenue came from the United States, just 2.3 percent from the rest of the world. Third, the partner: in the U.S. it is not Ironwood that sells but AbbVie — Ironwood merely books its share of the joint net profit.

Ironwood's entire earnings position thus hangs on a single chain: one compound, one home market, one marketing partner. Each link is stable on its own — LINZESS is established, the U.S. is the largest pharma market, AbbVie is a reliable giant. But there is no second strand to catch a tear, whether from payer price pressure, a contract change with AbbVie or the generics from 2029. For the casual glance at a profitable pharma company, this triple concentration is easy to miss.

Ironwood's (Nasdaq: IRWD) shareholder register contains a name that makes biotech boards sit up: Sarissa Capital Management — the investment firm of activist investor Dr. Alexander Denner. Unlike the usual index giants, Denner is not merely a shareholder: per the proxy statement (DEF 14A) he has sat on the board since 2020 and chairs its nominating and governance committee. Sarissa holds roughly 5.6 percent (9.19 million shares as of March 31, 2026) and has been engaged with an activist ownership filing (Schedule 13D) since 2019.

Denner was once a close associate of famed investor Carl Icahn and served on the board of Biogen, among others. For investors, such an activist cuts both ways: on one hand he tends to push for disciplined capital allocation, a clear strategy or value-creating steps — a possible catalyst. On the other, a board seat is no guarantee of a breakthrough, and the hard structural questions (the 2029 patent cliff, the debt wall) are not solved single-handedly by a prominent name either.

Whoever buys VectivBio for roughly $1 billion expects a fat asset on the balance sheet. At Ironwood (Nasdaq: IRWD), the opposite happened. Because the purchase was classified as an asset acquisition rather than a business combination, and the drug candidate apraglutide had "no alternative future use," practically the entire purchase price — roughly $1.1 billion — moved through the income statement immediately and in full as research expense (in-process R&D) in 2023.

The consequence: an operating loss of $945.4 million and a net loss of roughly $1.03 billion in 2023 alone. That is why Ironwood's balance sheet today shows neither meaningful goodwill nor large intangible assets. There is a curious flip side that deserves a fair mention: there is no impairment risk left — the purchase price has long been expensed. Should apraglutide ever be approved, the payoff would land on a cost basis of nearly zero.

Anterix’s board approved a share repurchase program of up to $250 million in 2023. A find in the annual report (10-K) for fiscal year 2026 shows how sparingly the company used it most recently: $226.7 million remains open — across all of fiscal year 2026, Anterix bought back only 43,175 shares for about $1.0 million.

That is a telling side note: in the very year the stock rose about 260 percent and climbed from roughly $20 to roughly $80, management throttled the buybacks to a minimum — and the insiders switched from buyers (in the fall at $20–22) to sellers (in the summer at around $80). A company that buys its own shares when they are cheap and pauses when they get expensive is behaving rationally. For investors following the scanner signal "institutions + CEO buy", it is also a quiet hint at how management itself views the current valuation.

In July 2024, the investment firm Heard Capital LLC, run by founder William E. Heard, filed a Schedule 13D with the U.S. securities regulator — the form an investor uses to signal openly activist intentions. A few weeks later, in August 2024, Heard was already sitting as an independent director on Anterix’s board. The attacker had become an insider. The 2025 annual meeting was contested nonetheless: on June 30, 2025, Anterix filed a DEFN14A — the code for a contested (non-management) proxy fight over board seats.

Hardly any investor who only sees the scanner hit "insiders buy" would expect a power struggle behind the governance: first the 13D activist, then the board seat, then a contested vote — and in parallel a complete change at the top from Rob Schwartz to Scott Lang. Anyone filing Anterix away as a quiet spectrum landlord is missing how much was moving in the engine room in 2024/2025.

Manufacture electrical components for 130 years and you leave traces. Two former CTS sites — Asheville, North Carolina, and Mountain View, California — sit on the National Priorities List of the U.S. Superfund program for the country's most demanding contamination cleanups. Asheville has been governed by a consent decree with the U.S. Environmental Protection Agency (EPA) since March 2017; in February 2023 the agency additionally sought $9.955 million in past response costs from the three potentially responsible parties, an amount the Department of Justice later adjusted to $8.288 million.

The annual report (10-K) for 2025 records the current state: on October 3, 2025, CTS presented a settlement offer as part of pre-litigation mediation — "the mediation is ongoing," the outcome open. Accrued: $6.575 million (estimated range: $6.575 to $7.169 million). For the company that is manageable — but it is a lesson in how old industrial firms must keep accounting for their past: per the report, the sites generate no revenue at all anymore, yet keep costing money and lawyers.

The annual report (10-K) for 2025 documents an awkward discovery under "Immaterial Correction of Prior Period Errors": the books of naval-sonar specialist SyQwest, acquired in July 2024 for $128.0 million, carried errors in the calculation of revenue and cost of goods sold — both before and after the acquisition date. The errors migrated from the target's books straight into the consolidated financial statements.

The correction table quantifies the damage to the already audited fiscal year 2024: revenue down $1.0 million, net earnings trimmed from $58.1 million to $55.5 million (minus $2.6 million), diluted earnings per share from $1.89 to $1.81. CTS classifies the errors as "immaterial" — formally defensible under SEC standards, but 4.5 percent of a year's profit is a remarkable footnote for the acquisition that is supposed to carry the company's defense story. In the fourth quarter of 2025, further "immaterial" follow-up corrections from the same source were recorded.

The equity section of the quarterly report (10-Q) as of March 31, 2026, contains a curiosity worth reading twice: CTS has issued 57.7 million shares, of which 29,047,141 sit in the company vault as treasury stock — more than the 28,624,587 shares still outstanding. Over the decades, the company has bought back more than half of all shares it ever issued.

The price tag sits on the balance sheet as of December 31, 2025: $543.7 million in cumulative cost of treasury shares — almost as much as the entire remaining equity of $551.8 million. And the pace is picking up: $41.3 million (2023), $43.0 million (2024), $56.7 million (2025); in November 2025 the board authorized a new $100 million repurchase program. The dividend stays symbolic next to that: $0.16 per share per year, most recently about $4.8 million in total payout. CTS pays its shareholders almost exclusively through its own demand for its own stock.

The annual report for fiscal year 2025 explicitly calls R.J. O'Brien & Associates, acquired on July 31, 2025, "the oldest futures brokerage in the U.S." — bought by a group whose own roots, per the same report, go back to Saul Stone in 1924. Two of the oldest names in American futures trading now sit under one roof.

The payment method is remarkable too: besides roughly $651.9 million in cash, StoneX settled the purchase price with 3,085,554 shares reissued from treasury stock — repurchased shares given a second life as acquisition currency. On top came $125.7 million of assumed RJO debt and a new $625 million bond at 6.875% due 2032. The oldest brokerage in the U.S. was thus paid for with a three-part chord of cash, canned shares and fresh debt.

Whoever pictures broker screens at StoneX overlooks the warehouse: as of March 31, 2026, the balance sheet carried $1,038.8 million of physical commodities inventory — $437.2 million of it at fair value. The group does not just trade derivatives on gold, grain and energy; it buys, stores and delivers the goods itself — precious metals holdings and agricultural commodities included.

This physical business is exactly what inflates reported revenues to $132.4 billion in fiscal year 2025 — and it hands StoneX a risk exotic for a financial firm: in the quarter ended March 31, 2026, $8.0 million of bad debts came out of the Global Metals business — client receivables from metals trading, not from the securities business. A broker that is also a merchant carries both kinds of risk.

Stock splits have become rare — StoneX pulled off two 3-for-2 splits within twelve months. The first was completed on March 21, 2025; the second, approved by the board on February 3, 2026, followed on March 20, 2026 — each executed as a stock dividend: every holder received one additional share for every two held, with cash paid in lieu of fractions. 100 shares held before March 2025 turned into 225 within a year.

The quarterly report justifies the latest split as making stock ownership "more accessible to employees and investors." Economically, splits change nothing, of course — but as a signal they are a rarity: only a board expecting further rising prices treats itself to two splits back to back. For context: StoneX has never declared a cash dividend in its history.

Among all the young names in the momentum leader lists, this sentence from the annual report (10-K) for 2025 reads like a relic: “We have operated in the transmission and distribution industry since 1891” — founded as L.E. Myers; the SEC’s EDGAR database still lists the historical company name “MYERS L E CO GROUP.” Electrical services for commercial and industrial construction were added in 1912. The stock that showed a gain of 171 percent over twelve months as of July 10, 2026, is thus one of the oldest names in the entire scanner universe.

The man at the top fits the picture: CEO Rick Swartz joined MYR in 1982 as a project foreman, per the annual report, worked his way up through superintendent, project manager and district posts, and has led the company since January 2017. A 135-year-old trade business, run by a man with 44 years of service — that is the kind of continuity no momentum metric captures, and yet it explains why utilities have let the same company climb their poles for generations.

About 85 percent of MYR Group’s 7,200 craft workers are union members — and through its collective bargaining agreements the company pays into multiemployer pension funds tied to more than 300 IBEW locals. The footnote on this (Note 15 “Employee Benefit Plans” in the annual report 10-K for 2025) contains a number hardly any investor has on the radar: into the Southern California IBEW-NECA Pension Trust Fund alone, $51.6 million of contributions flowed in 2025 — more than the company’s entire 2024 net income ($30.3 million). And of all funds, this one sits in the yellow zone under the Pension Protection Act (“endangered,” 65 to 80 percent funded), with a funding improvement plan and a contribution surcharge imposed.

The structural risk of such plans is spelled out in the filing itself: if an employer exits an underfunded fund, a withdrawal liability based on the underfunding comes due, and the obligations of departing employers can shift onto those remaining. MYR notes that some plans it contributes to have even been classified as “critical,” while stating it is not currently aware of related liabilities. For investors, the finding stands: part of the retirement burden of this business model sits not on MYR’s balance sheet but in third-party funds — and their health helps decide what the 2026–2028 wage rounds will cost.

In 2024 and 2025, MYR Group bought back its own shares aggressively: $75 million in each year, about 1.28 million shares in total at average prices of $116.54 and $117.33. On July 30, 2025, the board approved a new $75 million program — and then: nothing. Item 5 of the annual report (10-K) for 2025 dryly records that not a single share was purchased in October, November or December of 2025, that the full $75 million remained available, and that the program expired on February 4, 2026.

The context makes the footnote piquant: in exactly that window the stock ran toward its all-time high — as of July 10, 2026 it traded around $487, more than four times the company’s own last purchase price. Buyback programs are authorizations, not obligations, and restraint after a near-tripling can be read as discipline. But as a price signal, the footnote is unambiguous: the leadership team that knows this business best last found its own stock worth buying around $117 — and afterwards no longer did. MYR Group pays no dividend either.

In November 2024 the board authorized a share repurchase program of $300 million running through December 2029, executed via an automatic Rule 10b5-1 trading plan. In fiscal year 2025 — at prices roughly between $30 and $46 — the company bought aggressively: $97.1 million went into its own shares (657,000 under the prior program, 1,785,000 under the new one).

Then the stock took off — and the machine throttled down to homeopathic doses: roughly 171,000 shares for $6.8 million in the first half of fiscal year 2026, and in the rally quarter from January through April 2026 only about 3,000 shares for $0.1 million (quarterly report 10-Q as of April 4, 2026). You can read that as price discipline: management itself was evidently unwilling to pay the prices the market has been quoting since spring 2026. Whoever buys today pays them.

On March 9, 2026, the U.S. securities regulator, the SEC, sent Kulicke & Soffa a comment letter on its quarterly report (10-Q) as of January 3, 2026. One trigger: the Advanced Solutions segment had reported negative selling, general and administrative expenses for the prior-year quarter — a curiosity the report nowhere explained. The company’s response of March 19, 2026 lifts the curtain: behind the credit balance sits primarily the $71.1 million reimbursement from the cancellation of “Project W” by an unnamed mega-customer, plus $1.7 million from a supplier settlement.

The SEC also faulted the revenue commentary for not quantifying its drivers and the MD&A for not discussing segment operating income. Kulicke & Soffa promised improvement on all three points in future filings. Not an accounting scandal — but a rare glimpse of the regulator ordering more precise numbers work, in the very year the stock became a momentum star.

Tucked into the risk factors of the annual report (10-K) for fiscal year 2024 is an incident hardly any investor has on the radar: in May 2024, Kulicke & Soffa detected unauthorized access attempts into its network and servers — and determined that the threat actor had “accessed and acquired” data, explicitly including source code, engineering information, business partner data and personally identifiable information.

The company deems the incident immaterial to its operations and financial condition — an assessment you may or may not take at face value. What remains remarkable: for a machine builder whose competitive edge sits in decades of process know-how, source code and engineering data are the family silver. There was no current report under Item 1.05 of Form 8-K (the mandatory disclosure for material cybersecurity incidents) — the find exists solely as an example inside the risk factors.

Astronics has two share classes: the Nasdaq-listed common stock (ATRO) and a Class B stock (ATROB, over the counter) with ten votes per share. Per the annual report (10-K) for 2025, roughly 31.9 million common shares and 3.8 million Class B shares were outstanding as of February 19, 2026 — the ten-fold voting power turns about 10 percent of the shares into roughly half the votes.

The curious part: as of the record date there were more Class B holders of record (1,707) than holders of the common stock (1,588) — the super-voting paper sits broadly scattered in legacy holdings, not with a single founding clan. Selling forfeits the privilege: on sale, Class B automatically converts one-for-one into common stock; only gifts and inheritance preserve the ten votes. Dividends on Class B may also only flow if the common stock receives at least as much — currently Astronics pays none anyway (dividend-free for the past three years, no plans per the 10-K).

In fiscal year 2025, $10.4 million of tariff expense weighed on Astronics' cost of products sold, per the annual report (10-K). Then things got curious: on February 20, 2026, the U.S. Supreme Court struck down certain tariffs imposed under the IEEPA emergency statute. The U.S. administration followed up with a new global 10 percent tariff under Section 122 of the Trade Act — which the U.S. Court of International Trade struck down in turn on May 7, 2026.

The quarterly report (10-Q) as of April 4, 2026, notes dryly that it remains uncertain what impact these decisions will have — "including the process and availability of obtaining refunds of amounts previously paid for the IEEPA tariffs". Translated: somewhere in the balance sheet slumbers a potential multi-million-dollar refund claim whose fate hangs on courts and agencies — a small lottery ticket hardly any investor has on the radar.

On December 3, 2024, Astronics raised $165 million through a convertible note with a 5.5 percent coupon maturing in 2030. Then the stock price roughly tripled — and the note turned into a trap: because the paper ran deep into the money, the company repurchased 80 percent of it ($132 million in principal) in the third quarter of 2025, nine months after issuance. The price, per the quarterly report (10-Q) as of April 4, 2026: roughly $285.8 million in cash in total — more than double the principal repurchased. The 2025 income statement was left with a $32.6 million loss on settlement of debt, more than the entire net income of the year ($29.4 million).

The buyback was funded with a new, now zero-coupon convertible of $225 million (due January 2031, conversion price $54.87) plus $85 million drawn on the credit facility. For another $26.9 million, Astronics bought capped calls that soften dilution up to a stock price of $83.41 — and at the scanner run of July 17, 2026, the stock already traded above that cap. The lesson in the footnote: convertibles are cheap as long as the stock goes nowhere — they get expensive precisely when everything goes right.

The liquidity chapter of the annual report (10-K) for 2025 contains an involuntary valuation marker: during 2025, Benchmark repurchased roughly 0.7 million of its own shares for a total of $26.8 million — at an average price of $38.22 per share; $122.7 million of the buyback authorization remained open at year-end. As of the July 17, 2026 data cut-off, the same share cost about $97.

You can read that as a success story: management bought low. You can also read it as a yardstick: the business the company itself valued at $38 per share in 2025 has barely changed operationally since — 2025 revenue was flat on the prior year, gross margin stood at 10.2 percent, and the first quarter of 2026 brought 7 percent growth. What has changed, almost exclusively, is the price other people pay. The interesting question is whether the company keeps buying at current prices — the remaining $122.7 million of authorization now buys only about half as many shares as it would have a year earlier.

While the Benchmark stock was doubling, the leadership rebuilt itself almost completely — documented in a series of Forms 8-K (mandatory filings for material events): on September 2, 2025, CEO Jeffrey Benck announced his retirement effective March 31, 2026; his successor is David Moezidis, previously 25 years at EMS competitor Flex, who per his February 19, 2026 employment agreement starts at a $900,000 base salary plus incentives. On December 29, 2025 came the retirement of chief technology officer Jan Janick effective January 16, 2026. And back in October 2025, after the unexpected passing of director Robert Gifford, Michael Slessor joined the board — the sitting CEO of semiconductor test specialist FormFactor.

None of this is improper on its own; the successions were announced in an orderly fashion, and the filings stress there were no disagreements. But taken together it means: the team responsible for the run the stock just had is no longer the team that must deliver the expectations now priced in — the new CEO took over on March 31, 2026, inheriting a share price that already demands perfection.

Note 1 of the annual report (10-K) for 2025 carries a heading few investors ever read: "Immaterial Correction of an Error." In the fourth quarter of 2025, Benchmark discovered errors in its own income tax calculation — and retrospectively corrected the comparative figures for 2024 and 2023. The result of the cleanup: opening retained earnings for 2023 had been overstated by $11.2 million; 2024 tax expense had been understated by $2.2 million, 2023 tax expense overstated by $4.6 million, with deferred tax positions shifting by millions more.

The company classifies all of it as "not material," which means no earlier report has to be formally withdrawn. The clustering is still remarkable: the annual report for 2024 was followed by a 10-K/A — a formal amendment — just three days after filing, and the 2025 profit collapse itself traces chiefly to a 59.6 percent effective tax rate. Anyone comparing multi-year Benchmark figures is now comparing corrected numbers with as-reported ones in places — and the tax line has proven to be the most error-prone spot in this company's books.

In the Global ECS segment, Arrow has entered into non-cancellable multi-year purchase obligations running through 2032, per the quarterly report (10-Q) as of April 4, 2026 — in exchange, the company was designated exclusive partner for certain products. In the first quarter of 2026, Arrow recorded a loss on one of these contracts, "due to lower profit expectations on a certain underperforming contract," and warns verbatim that there could be "additional losses in the coming quarters on certain agreements" — the long-term performance of the deals cannot be reasonably estimated at this time.

The size of the total package sits in the annual report (10-K) for 2025: purchase obligations of $21.3 billion — non-cancellable inventory purchase orders and future payments under IT distribution arrangements, $11.4 billion of it due within twelve months. That is more than one and a half times the market value (roughly $12.2 billion, data as of July 17, 2026). The annual report explicitly lists the risk that sales may not be sufficient to cover these obligations. For a company whose business model has traditionally been to trade other people's inventory flexibly, this is a remarkable role change: Arrow is increasingly taking its own demand risk onto the books.

The 8-K filed September 17, 2025, records an unusually hard cut: CEO Sean Kerins separated same-day, effective September 16, 2025, as director, president and CEO; the board shrank from ten to nine seats with his exit. Per the separation agreement, Kerins receives no severance payments, and his unvested equity and cash incentive awards were forfeited as of the effective date — remarkably austere terms for what is presented as an orderly transition.

Two details make the filing worth reading: first, Kerins agreed to advise the company for at least six months — at $61,000 per month. Second, Arrow used the very same 8-K, under Item 2.02, to preemptively reaffirm its third-quarter 2025 outlook — investors were evidently not supposed to read the leadership change as a profit warning. The filing gives no reason for the separation; the customary formula that the departure was "not the result of any disagreement" is absent for the CEO — though it does appear in the filing two weeks earlier announcing the change of chief accounting officer (8-K of August 29, 2025).

The balance sheet as of December 31, 2025, holds a curiosity you have to read twice at a trading company: accounts receivable jumped from $13.0 to $19.7 billion within one year (+51 percent) — while revenue grew only 10.5 percent. Mirror-image, accounts payable leapt from $11.0 to $17.4 billion, and total assets swelled from $21.8 to $29.1 billion. For scale: the market value stood at roughly $12.2 billion most recently (data as of July 17, 2026) — the receivables line alone is more than one and a half times that.

The explanation sits in the liquidity section of the annual report (10-K): per the company, the swings in receivables and payables are primarily tied to the supply chain services of the components business — Arrow acts as an intermediary, collecting from the customer and remitting to the supplier, so both sides of the balance sheet inflate in step. The cash-flow consequence is still real: operating cash flow fell to $64 million in 2025 — after $1,130 million the year before — as inventory was also built up for the expected market recovery. A company with $571 million of book profit that barely generates operating cash: not an alarm bell, but a textbook lesson in how growth in distribution costs money before it makes any.

Deep in Astrana's corporate structure sits a curiosity: the consolidated physician entity Allied Physicians of California (APC) itself holds 6,132,802 Astrana shares, per the annual report (10-K) for 2025 — as "excluded assets" for the benefit of its physician shareholders. A vehicle that is consolidated in the company's own financial statements is thus simultaneously one of the parent's largest shareholders; on the balance sheet, those shares count among the 10,571,011 treasury shares.

In 2025, Astrana repurchased a total of 967,058 of its own shares for $25.6 million — including 300,000 shares for roughly $10.6 million directly from APC. The company is, step by step, buying its own stock back from an entity it consolidates. All disclosed and permissible — but a vivid example of how intricate California's "friendly PC" construction of physician corporations and management holding companies gets in practice.

A small study in wording, with timestamps: in the annual report (10-K) for 2025, filed on March 12, 2026, Astrana describes itself as a "leading physician-centric, technology-powered, risk-bearing healthcare company." In the quarterly report (10-Q) as of March 31, 2026, filed on May 8, 2026, the same self-description suddenly reads: "leading physician-centric, AI-powered, risk-bearing healthcare management company."

Nothing discernible changed in the business model between the two filings — revenue still comes from capitation and care services, not from selling AI. The company does, per its annual report, use machine learning models and predictive analytics in its platform. But the relabeling within eight weeks is a neat little time capsule of how quickly "technology" becomes "AI" in 2026 when investors are listening.

With the Prospect acquisition, the network-and-administration specialist Astrana bought a real acute care hospital for the first time: Foothill Regional Medical Center in Tustin, California — roughly 109,000 square feet of surgery, orthopedics, intensive care and skilled nursing, now its own reporting unit ("Care Delivery – FRMC").

The risk-factor section of the annual report (10-K) for 2025 is disarmingly honest about it: "We have no experience owning or operating a hospital" — and the company expects significant operational challenges, from clinical risks and labor shortages to competition from Hoag and UC Irvine Health. A company whose core business is coordinating outpatient care is now learning the most expensive corner of healthcare from the inside — something hardly any investor has on the radar when hearing "physician network platform."

The annual report (10-K) for 2025 records a sequence you have to read twice: on July 1, 2025, Astrana closed the $674.9 million acquisition of the Prospect businesses — and on July 7, 2025, six days later, the seller entities (the "Prospect PhysicianCo Entities") filed a voluntary petition under Chapter 11 of the Bankruptcy Code.

The second punchline sits in the same risk-factor section: under a letter agreement dated July 1, 2025 — closing day — the escrow account for non-assumed liabilities and the recourse against the sellers (with limited exceptions) were eliminated. Astrana itself writes that in a worst case it would have "limited to no recourse" and might have to absorb the costs of the insolvent sellers' breaches to protect its own business interests and relationships.

The annual report (20-F) for 2023 contains a sentence you rarely read in the papers of a listed company: the most recent annual general meeting was held on June 25, 2021 — NESR held no shareholder meeting in 2022 or 2023, "due to the ongoing activities associated with the restatement" of previously issued financial statements. This was possible under the legal framework of the British Virgin Islands, where NESR is incorporated: the board may delay or postpone general meetings at its discretion.

For shareholders it meant two years without a forum to vote on the accounting crisis or hold the board to account — precisely while the audited 2020 figures were being restated and the 2021 annual report ran roughly 20 months late. Since the return to normality the picture has flipped: the control weaknesses were declared remediated (June 30, 2025), and since January 1, 2026, NESR is subject to full U.S. domestic-issuer obligations, including proxy rules that keep future meetings on a much tighter leash.

On May 26, 2026 — after the share price had multiplied within twelve months — NESR filed an automatic shelf prospectus (S-3ASR) with the SEC registering 28,257,859 shares held by legacy holders for sale "in one or more offerings": Olayan Financing Company (17.3 million), Al Nowais Investments (4.8 million), Mubbadrah Investments (3.9 million), plus 1.8 of the 3.2 million shares held by CEO Sherif Foda and smaller stakes. Together that is roughly 28 percent of all 100.85 million shares outstanding.

A registration is not a sale — it is the loaded revolver in the holster. That it is not merely decorative shows in the parallel filings: seven notices of proposed sale (Form 144) arrived in May 2026 alone, and board member Yousif Al Nowais reported share sales on Form 4 as late as the end of June 2026. In fairness: the anchors date back to the 2017/2018 SPAC founding era, and a larger free float would improve the stock's notoriously thin tradability — but whoever buys the stock for its momentum should know that the best-informed insiders are now free to stand on the sell side.

The goodwill chapter of NESR's first annual report on Form 10-K contains a remarkably candid calculation: for the annual impairment test on October 1, 2025, management estimated the fair value of its two reporting units at a combined $1.8 billion — while the stock market valued the entire company at roughly $1.0 billion that same day (closing price of $10.39 across 100,777,759 shares). Management explained the gap in part by noting that the share price did not reflect "certain non-public information" available to management at the time.

The twist: exactly that information — including the announcement of a major contract award in Saudi Arabia after October 1, 2025 — later became public, and per the report the stock price "increased materially during the remainder of 2025." Management explicitly reads this as corroborative evidence for its own estimate rather than hindsight bias. For investors, the punchline cuts both ways: goodwill of $645.1 million — about a third of total assets — hung in the fall of 2025 on a valuation the market only came around to months later.

On January 6, 2026, ASP Isotopes completed the acquisition of South African helium and LNG producer Renergen (Virginia Gas Project, 94.5 percent of operating company Tetra4) — paid for with 14,270,000 of its own shares worth roughly $92.9 million. Less than six months later, on June 25, 2026, the company signed a merger agreement that carves Renergen right back out of the group via the holding company "Noble Africa": through a merger with ENDRA Life Sciences (Nasdaq: NDRA) — a microcap whose business to date has been thermoacoustic ultrasound imaging for fatty liver disease.

After closing, Noble Africa is expected to trade on Nasdaq as a standalone helium company under the ticker NOBA; in parallel, Noble is raising roughly $50 million from investors ($6.57 per unit), and ASP Isotopes receives 55.5 million Class B units for contributing Renergen. Call it value creation or financial acrobatics: an asset of this size that is bought, re-hung and spun out again within half a year makes the group's numbers extraordinarily hard to compare for outsiders.

The annual report (10-K) for 2025 contains a sentence you have to read twice: on August 29, 2025, Paul Mann, Chairman and Chief Executive Officer of ASP Isotopes and also Chairman of the Board of Managers of subsidiary Quantum Leap Energy, purchased, "as an individual investor," $2.5 million of shares and warrants of Skyline Builders Group Holding (Nasdaq: SKBL) — on the same terms as the investment Quantum Leap Energy was making for the group in parallel. Effective January 1, 2026, Skyline's board additionally appointed Mr. Mann Executive Chairman.

For fairness: the report discloses all of this, none of it is illegal, and the warrants carry a 4.99 percent exercise cap. But the constellation remains remarkable — the acquirer's CEO sits on both sides of the table: as the group executive deploying shareholders' capital into Skyline, and as a private investor personally participating in the same deal. A few months later, Skyline was deconsolidated again via an exchange agreement. Anyone valuing ASP Isotopes should keep such double roles in view.

Open the revenue-by-geography table in ASP Isotopes' annual report (10-K) for 2025 and you find a line nobody expects at a nuclear and quantum technology company: Hong Kong, $18.2 million — out of $23.8 million of total consolidated revenue. The reason: in August 2025, of all subsidiaries it was the nuclear-fuels unit Quantum Leap Energy that acquired a 79 percent voting interest in Hong Kong builder Skyline Builders Group Holding (Nasdaq: SKBL), which per the report mostly performs road and drainage works for the city government. Two construction customers thereby accounted for roughly 32.2 percent and 13.7 percent of total consolidated revenue.

The chapter ended as quickly as it began: effective March 29, 2026, Skyline was deconsolidated again through a securities exchange agreement — with a book gain of roughly $20.8 million and a retained stake of about 8.6 percent. In the quarterly report (10-Q) as of March 31, 2026, the previous year's largest revenue segment is already presented as "discontinued operations." Anyone reading ASP Isotopes' revenue series should know: the jump from $4.1 million to $23.8 million in 2025 was three-quarters asphalt, not atoms.

The fundamental data on Lattice stock shows a striking insider pattern as of the July 17, 2026 data cut-off: 20 sale transactions by officers and directors stand against zero purchases. At the same time, institutional investors trimmed on balance (6 funds adding versus 12 reducing). The counterparty with the biggest checkbook was the company itself: in the first quarter of fiscal 2026, Lattice bought back 165,913 of its own shares for $15.0 million — at an average of $90.41, under a $250 million program announced in December 2025.

To be fair: with a compensation model that carried $116.3 million of stock-based compensation in fiscal 2025, scheduled insider sales after a doubling are normal, not scandalous. But the division of roles remains remarkable: the people who know the company best exclusively cashed out into the rally — the buying was done with corporate money by the balance sheet.

In the quarterly report (10-Q) as of April 4, 2026, Lattice reports a 64 percent revenue share for "Greater China." The accompanying footnote immediately relativizes its own map: the attribution follows the ship-to location, and — verbatim — "Products shipped to Hong Kong may subsequently be transferred to mainland China or other destinations, and products shipped to mainland China may similarly move through intermediary locations."

Translated: the 64 percent is a logistics figure, not a map of end demand — a substantial part runs through distributor warehouses and contract manufacturers in the region whose end customers may sit elsewhere. For investors that cuts both ways: the true China dependence may be smaller than the headline 64 — or new export controls could strike exactly where the supply chain is densest. Not even the filing knows precisely.

Read the Lattice annual report (10-K) for fiscal 2025 all the way to the footnote on contract assets and you find a hidden second business: the FPGA maker's contract assets, per Note 3, "relate primarily to our rights to consideration for licenses and royalties due to us as a member of the HDMI Founders consortium."

The background: through Silicon Image, acquired in 2015, Lattice belongs to the founding companies of the HDMI standard — the plug on practically every TV, console and projector. A central licensing agent collects the fees and distributes them to the founders. An FPGA specialist quietly collecting on one of the most widespread consumer standards in the world: not a load-bearing revenue pillar, but a curiosity hardly any investor has on the radar.

TD SYNNEX’s Hyve unit grew a reported 49 percent in the second quarter of fiscal year 2026 (March through May 2026). A subordinate clause in the quarterly report (10-Q) reveals that the actual surge was far larger: a growing share of sales runs through a "customer-owned procurement model" — the customer buys the components itself, TD SYNNEX merely builds and delivers, and then recognizes only its margin as revenue (net-basis presentation). Per the filing, this effect depressed Hyve’s reported revenue growth by roughly 68 percentage points for the quarter and 70 percentage points for the half year.

For investors that means two things. First, the AI tailwind at Hyve is far stronger in gross terms than the 49 percent suggests. Second, revenue series and price-to-sales metrics at this company are becoming increasingly incomparable — the same business activity can appear in the revenue line at full merchandise value or at margin only, depending on the contract model. A side effect worth noting: in the second quarter of fiscal 2026, for the first time, no single customer crossed the 10 percent revenue disclosure threshold — not because the dependence shrank, but in part because the associated revenue is now reported on a smaller, net basis.

Whoever looks only at the "Borrowings" line at TD SYNNEX ($4.6 billion as of November 30, 2025) underestimates the actual financing apparatus. The annual report (10-K) for fiscal year 2025 lists two more pipes: under supplier finance programs, $3.7 billion of payment obligations that vendors had sold to banks sat on the books at the reporting date — tucked inconspicuously into the ordinary "Accounts payable" line. And through accounts receivable purchase agreements, the company had sold $1.8 billion of customer receivables to financial institutions without recourse; the discount fees for those programs cost a full $62.7 million in fiscal year 2025.

None of this is illegal or hidden — it is right there in the filing. But it shows how much the pass-through business of IT distribution hangs on a quiet financing web of banks: the inventory is stretched via vendor credit, the receivables are turned into cash upfront. If interest rates rise or the banks pull back, the working-capital model can turn into a bottleneck quickly.

The quarterly report (10-Q) as of May 31, 2026, records a deal you would not expect at an IT wholesaler: in May 2026, TD SYNNEX issued a warrant to Amazon.com NV Investment Holdings LLC — rights to up to 3,238,066 shares, roughly 4 percent of all shares outstanding. 215,871 of them vested immediately at an exercise price of $0.01 per share — with the stock trading around $246 on the grant date, effectively a gift worth a good $53 million. Amazon earns the remaining 3,022,195 warrant shares (exercise price $191.10) by hitting defined purchase thresholds at TD SYNNEX; the warrant runs through May 30, 2033.

The construction is a familiar AI-boom pattern — suppliers bind their biggest buyers by giving them a stake in their own share price. For existing shareholders it means: the better the hyperscaler business runs, the more discounted shares migrate to the customer, and vested warrant value is booked as a reduction of revenue. Growth paid for with your own shares is never entirely free.

Since acquiring the Chinese lift truck maker Maximal, Hyster-Yale has owned 90 percent — the remaining 10 percent sits with Y-C Hongkong Holding. The annual report (10-K) for 2025 documents a remarkable construction: Hyster-Yale may buy the remaining 10 percent at any time until June 8, 2056, for $16.8 million — but the partner mirrors that with a put option that can force Hyster-Yale to buy.

Because one of the triggers of that put option lies outside the company's control, the minority stake counts as "contingently redeemable" and is reported not within stockholders' equity but as temporary equity between debt and equity on the balance sheet. The amount is manageable for the valuation — but as a lesson in the fine print of China joint ventures reaching three decades into the future, the footnote is a genuine find.

At first glance, Hyster-Yale's pension setup is a model student: the defined benefit plans in the U.S. and U.K. are frozen, and both ended 2025 overfunded (U.S.: +$2.7 million, non-U.S.: +$14.8 million surplus). For the U.K. plan, the trustee signed a "buy-in" contract with an insurer in January 2025 — the preliminary stage of a full transfer.

The catch hides in accumulated other comprehensive income: actuarial losses of $22.4 million (U.S. plan) and $53.2 million (non-U.S. plans) sit there, never having passed through the income statement. The annual report (10-K) for 2025 says it plainly: if the buy-in becomes a "buy-out" and settlement accounting applies, the amounts relating to the U.K. plan — the majority of the non-U.S. position — are reclassified into earnings. A pure bookkeeping effect with no cash outflow, but one that could dent a future quarter by tens of millions without anything changing in the business.

Hyster-Yale puts the tariff-related costs of 2025 at roughly $100 million in the annual report (10-K) — more than the entire net loss of the year ($60.1 million). In February 2026, the U.S. Supreme Court ruled that tariffs imposed under the International Emergency Economic Powers Act (IEEPA) are not legally authorized, and in April 2026 the U.S. Customs and Border Protection agency even issued procedures for refunds.

The punchline sits in the contingencies footnote of the quarterly report (10-Q) as of March 31, 2026: Hyster-Yale has recorded no potential recovery whatsoever, "as the amounts and timing of refunds are uncertain." The full tariff bill therefore sits in the books — any partial refund would be pure tailwind that neither the balance sheet nor the guidance prices in. How much of the $100 million was IEEPA-related, however, the company does not break out — and the 2026 outlook explicitly assumes zero recovery.

On June 25, 2025, Wolfpack Research published its short report against CTO Realty Growth — the stock slid toward $17. You will search the SEC database in vain for a defensive press release. What you find instead: Form 4 filings dated June 26, 2025, the first trading day after the attack.

That day, per the mandatory filings, CEO John P. Albright bought 3,800 shares at about $17.05 each, CFO Philip Mays 1,000 shares at $17.29, general counsel Daniel E. Smith 1,000 shares at $17.00, and director George R. Brokaw 2,000 shares at $16.94 — all regular open-market purchases (transaction code "P"), roughly $133,000 in total. That does not prove the report wrong; insider buying after attacks is also a well-known signaling game. But it is the kind of answer that costs money instead of words — and a year later the stock traded about 24 percent above those purchase prices (data as of July 8, 2026).

Note 5 of the annual report (10-K) for 2025 hides a remarkable construction. CTO not only manages the external net-lease REIT Alpine Income Property Trust (PINE) for a base fee of 1.5 percent of equity — since December 2023 it has also managed the property portfolio of an unnamed third party (the "Portfolio Management Agreement").

The decisive sentence sits right next to it: "Although the Company has no direct relationship with the third party, PINE is a lender to the third-party pursuant to a mortgage note originated by PINE which is secured by the portfolio." In plain terms: CTO officially has nothing to do with the third party — but PINE, which is run by CTO's own executives and in which CTO holds 15.4 percent, has extended a mortgage loan to that third party, secured by exactly the portfolio CTO manages. Three roles, one leadership team: manager, co-owner of the lender, and steward of the collateral. Elsewhere, the 10-K itself warns that the conflicts of interest around PINE "could result in decisions that are not in the best interest of our stockholders."

Until May 2020, CTO Realty Growth was called Consolidated Tomoka Land Co — for over a century a classic Florida land company around Daytona Beach. Until recently, the company still carried an heirloom of that past on its books: subsurface mineral rights under roughly 352,000 acres of land in 19 Florida counties — the right to search for natural resources beneath other people's property, an area of about 550 square miles.

In fiscal year 2024 these "Subsurface Interests" were sold, along with the last mitigation credits of the formerly company-owned mitigation bank. The curious final twist: CTO parted with the ownership, not with the business — it signed a management agreement with the buyer (the "Subsurface Management Agreement"), so the shopping-center REIT still manages, for a fee, the very subsurface rights it just sold. A piece of corporate DNA hardly any investor would suspect behind the ticker "CTO".

Scroll to the executive biographies of the annual report (10-K) for fiscal 2026 and you find, under "Shawn Nelson, 49, Chief Executive Officer," a sentence rarely seen in an SEC filing: "In 2005, Mr. Nelson won Richard Branson's 'The Rebel Billionaire' on Fox and continues to participate in ongoing TV appearances." Nelson, who founded Lovesac in 1998 around an oversized beanbag he sewed himself, won the Virgin billionaire's casting show and still appears on television.

The rest of the bio fits the unconventional profile: Nelson is the lead designer of the company's patented products, runs sourcing, PR and company culture, taught as a graduate-level instructor at Parsons in New York — and, per the annual report, is fluent in Mandarin, not a bad tool for a supply chain centered on Asia. None of this is a buy argument, but it is a rare case of a founder's TV history being spelled out verbatim in a mandatory securities filing.

On June 11, 2026, Keith Siegner still signed the quarterly results filing as CFO — on June 12 he and the company agreed on his resignation effective June 15, as recorded in the 8-K dated June 15, 2026. The separation package: $576,800 (twelve months of base salary in installments), accelerated vesting of two equity tranches and subsidized health coverage. The filing stresses that the departure is "not related to any financial or accounting issues or any disagreement with the Company" — a formula that sits in the form precisely because investors look twice at this spot after the 2023 restatement.

The successor is the remarkable part: per the same 8-K, Andrew Farag already served as Lovesac's interim controller from August 2023 to January 2024 — in the middle of the restatement cleanup — as an outside consultant with Ankura. He knows the company's accounting construction sites first-hand. His package: $560,000 base salary, a $255,000 signing bonus, plus equity grants worth roughly $1.24 million.

On February 20, 2026, the U.S. Supreme Court ruled that the tariffs imposed under the IEEPA emergency-powers statute were unlawful. For Lovesac, whose couches are made in Vietnam, Malaysia, China, Indonesia, Taiwan and India, the biggest margin-eater turned into a claim against the government overnight: the company filed refund claims with U.S. Customs in April 2026 and additionally lodged a complaint and a protective action with the U.S. Court of International Trade seeking a full refund of all tariffs paid under IEEPA.

The first wires have landed: per the quarterly report (10-Q) as of May 3, 2026, roughly $3.6 million in tariff refunds had arrived by June 10, 2026. The accounting is conservative: under the gain-contingency model, refunds only reduce inventory or cost of merchandise sold once the cash is actually received — none of it is in the May 3, 2026 numbers yet. The catch: right after the ruling, the U.S. administration announced a new 15 percent global tariff under Section 122 of the Trade Act of 1974. The tariff storm is not over — it has merely changed its legal basis.

Buried in the definition of a key performance indicator sits a sentence that says more about customer concentration than any risk factor. Explaining how it counts downloads, Getty Images notes that the metric "excludes downloads related to an agreement signed with Amazon, as the magnitude of the potential download volume over the deal term could result in significant fluctuations in this metric without corresponding impact to revenue in the same period."

Read that plainly: a single customer can move so much content that Getty's own volume metric would become unreadable — so the customer is carved out of the statistic. Nothing about that is improper, and the carve-out is disclosed rather than hidden. But it quietly marks the shape of the new business: the growth of recent years came from a small number of very large, multi-year, bulk agreements — in the fourth quarter of 2025 alone, "two significant multi-year license agreements" with "meaningful accelerated revenue recognition" flattered every product category. Deals of that size are wonderful when they renew, and a cliff when they do not.

Getty Images' revenue grew 4.5 percent in 2025. Its customer base did the opposite. The key-performance-indicator section of the annual report discloses that total purchasing customers fell to 689,000 in the twelve months to December 31, 2025 — from 717,000 a year earlier and 799,000 in 2023. That is a loss of roughly every eighth paying customer in two years. Total active annual subscribers fell in parallel, from 314,000 to 278,000.

The revenue line holds up because the survivors pay more: annual revenue per purchasing customer rose 8.7 percent to $1,424. Management attributes the customer decline largely to iStock subscriptions and the discontinued free-trial acquisition programme in June 2025 — a deliberate shift toward "committed solutions". That is a defensible strategy. It is also exactly what the early stage of an eroding base looks like from the inside: fewer, larger, more locked-in customers carrying a revenue number that no longer grows on its own. One more number belongs beside it: the annual subscriber revenue retention rate was 89.9 percent — the cohort spent about a tenth less than the year before.

Anyone following the "Getty vs. Stability AI" headlines would assume the money flows toward Getty. In the 2025 accounts it flows the other way. After Getty dropped its primary copyright and database-rights claims during the June 2025 UK trial and prevailed only on trademark infringement, the High Court in December 2025 assessed an interim costs award in favor of Stability AI for the matters the claimants lost or abandoned — $5.8 million.

The accounting treatment is the tell: the amount sits in "Accrued expenses" on the balance sheet as of December 31, 2025 and runs through "Other operating expense - net" in the income statement — the same line that jumped from $15.8 million to $54.8 million in 2025. The plaintiff books its own lawsuit as an operating cost. Getty has been granted permission to appeal the secondary-infringement decision, and Stability AI is seeking permission to appeal the trademark ruling. Whatever the appeals bring: the first cash to move in this landmark AI copyright case moved from the rights holder to the AI developer.

The single largest liability on Getty Images' balance sheet has nothing to do with photographs. When the company went public in July 2022 by merging with the SPAC CC Neuberger Principal Holdings II, former holders of the public warrants sued — the Initial Warrant Litigation (Alta Partners, LLC v. Getty Images Holdings, Inc.) and the Follow-on Warrant Litigation. Four years later that dispute has grown into the biggest number in the accounts that investors rarely look at: litigation reserves of $205.3 million as of December 31, 2025 (up from $110.9 million a year earlier), which rose again to $208.4 million by March 31, 2026.

Two details make it uncomfortable. First, the insurance does not stretch: coverage runs to $60.0 million for these cases combined, of which a remaining recovery receivable of about $35.0 million was left at year-end 2025 — the rest is Getty's own money. Second, the appeal did not work: on January 15, 2026 the Second Circuit "affirmed the Court's opinion and judgment in all respects, with one judge dissenting"; Getty petitioned for a rehearing on February 19, 2026. The annual report notes drily that "to date, no portion of the judgments entered in the Initial Warrant Litigation or the Follow-on Warrant Litigation, has been paid." Put next to a market value of roughly $264 million (data as of July 16, 2026), a reserve of $205.3 million is not a footnote — it is most of the equity story.

Japan is, as of this writing, the one major market where Alpha DaRT may actually be sold: in February 2026 the Ministry of Health, Labour and Welfare granted shonin pre-market approval for unresectable locally advanced or locally recurrent head and neck cancer. What few investors have on their radar is who does the selling — and on what terms.

In March 2026 Alpha Tau signed a commercial agreement with the Japanese partner HekaBio K.K. covering distribution. The annual report notes, almost in passing, that the agreement "can be terminated with 90 days' notice". On top, Alpha Tau owes HekaBio shares for clinical, consulting and administrative services, milestone payments, and a running royalty of 3.5 percent of the reimbursement price of the products in Japan plus 10 percent of revenues received from distribution receipts. So the first commercial market comes with a partner who can walk in three months, and a cut off the top before the first yen reaches Jerusalem — while the approval itself obliges Alpha Tau to run a post-market surveillance study of 66 patients at five Japanese centers. First revenue, when it comes, will be narrower than the approval headline suggests.

A company without revenue lives on the shares it sells, and Alpha Tau's placements read like a price ladder. On April 24, 2025 it agreed to sell 14,110,121 shares to Oramed at $2.612 per share, closing four days later for net proceeds of roughly $36.7 million. On January 11, 2026 it sold another 1,443,002 shares at $6.93 — "the closing share price immediately preceding the agreement" — for $10.0 million gross.

The effect shows up in the share count, not in the headlines: the weighted average number of shares used to compute the loss per share rose from 70,450,897 (Q1 2025) to 89,705,391 (Q1 2026) — about 27 percent more shares carrying the same company. Issued and outstanding shares went from 88,009,737 (December 31, 2025) to 90,176,067 (March 31, 2026). And the queue is not empty: options on 15,985,500 ordinary shares plus 597,700 RSUs were outstanding per the annual report — roughly another 18 percent of today's share count. From inception through December 31, 2025, the company had raised $234.2 million in total, of which $225.3 million came from issuing shares. None of this is hidden, and for a clinical-stage company it is the normal way to breathe. But remember the arithmetic: your slice of the story shrinks even while the story gets better.

Alpha Tau's net loss in the first quarter of 2026 was $22.9 million — nearly triple the prior-year quarter ($8.7 million). But the operating loss grew far less dramatically, from $9.3 million to $13.3 million. The gap sits in a line most readers skip: financial expenses, net, of $9.6 million, against financial income of $0.7 million a year earlier. The interim report names the cause in one half-sentence: "primarily due to the remeasurement of warrants liability".

The mechanics are worth understanding, because they invert your intuition. Warrants from the 2022 SPAC merger (they trade separately as DRTSW) sit on the balance sheet as a liability and are revalued every quarter — and the offsetting entry runs through the income statement. When the share price rises, those warrants become more valuable, the liability swells, and the company books an expense: the warrants liability grew from $5.4 million (December 31, 2025) to $15.7 million (March 31, 2026). Translated: a good part of the "worsening" loss is not a worsening business but a rising share price. It works in reverse too — if the stock falls, Alpha Tau will report a gain it did not earn. Whoever reads only the net-loss headline is partly reading yesterday's share price.

Eos used the 2025 share-price rally with remarkable timing. On November 24, 2025, the company completed the sale of 35,855,647 shares at $12.78 per share in a registered direct offering — $458.2 million of proceeds in one transaction. On the same day it issued $600.0 million of convertible notes due 2030. That capital is the direct reason the going-concern doubt disappeared from the annual report: management points to "the significant amount of capital raised in 2025" as the basis for its conclusion.

The side-find is what it says about dilution and timing at once. Weighted-average shares outstanding went from 212.0 million (2024) to 260.8 million (2025); shares actually outstanding reached 339,434,259 by February 24, 2026. The company sold equity into strength — good treasury management, and the reason it now has $410.7 million of cash (March 31, 2026). Whoever bought that offering, however, paid $12.78 for a business whose cost of goods sold that year still ran to $2.26 per $1 of revenue.

When an auditor flags a "critical audit matter", it marks the spot where the audit was especially difficult, subjective or complex. Deloitte named two for the 2025 financial statements — and neither has anything to do with building batteries. The first: "Convertible Notes and Warrants Liability". The second: "Financial Instruments Associated with the Credit and Securities Purchase Transaction" — the Cerberus package.

The reasoning is worth reading, because it explains why this company's bottom line moves the way it does: the valuation of these instruments, the auditor writes, "is inherently subjective and involves the use of complex modeling tools", and auditing it "required a high degree of auditor judgment" and the involvement of the firm's own fair value specialists. A factory can be counted, weighed and walked through. The instruments that decide Eos's reported profit cannot — they are modelled. For a manufacturer whose plant is its entire story, it is a quietly remarkable finding that the hardest part of the audit was the financing.

The loan facility from the U.S. Department of Energy is the pillar of the Eos growth story — the first Title XVII battery loan ever closed, up to $303.5 million including capitalized interest, meant to finance the "Project AMAZE" production lines. What almost no summary mentions is how the money is cut up. The facility runs in up to four tranches, each tied to one specific production line, each with its own ceiling: Tranche 1: $101,979 thousand, Tranche 2: $117,326 thousand, Tranche 3: $71,836 thousand, Tranche 4: $12,309 thousand — and the annual report states plainly that "any amounts not withdrawn under a specified tranche cannot be allocated to another tranche".

Translated: this is not a $303.5 million credit line the company can draw as it needs. It is four separate pots, each of which only opens if the matching line actually gets built and the funding conditions are met — and every dollar left in a pot is gone for good, not available elsewhere. Each tranche funds only 80 percent of the eligible project costs; the remaining 20 percent Eos must fund itself. Through December 31, 2025 the company had drawn $90.9 million — the full commitment of Tranche 1, and less than a third of the headline number.

In May 2023, SandRidge's board approved a share repurchase program of $75.0 million. As of December 31, 2025 — two and a half years later — $68.3 million of it was still available. In other words: in thirty months the company bought back barely $6.7 million of its own stock, most of it in 2025 (595,635 shares for $6.4 million at an average price of $10.72).

That is a remarkable restraint for a company that trades at roughly book value, carries no debt and sat on $111.0 million of cash at year-end. It is also a tell about where management actually wants the money to go: into the Cherokee drilling program and, per the stated strategy, into "opportunistic, value-accretive acquisitions". Whoever expects a buyback to support the share price here should read the authorization for what it is — a permission, not a promise. The annual report says as much: "There is no guarantee of future dividends or stock repurchases."

Anyone who reads oil and gas annual reports knows two numbers that almost never match: PV-10 (the discounted value of the proved reserves before taxes) and the standardized measure (the same value after future income taxes). The second is normally the smaller one — that gap is the tax authority's share of the oil in the ground.

At SandRidge, the reconciliation table reads like a typo: PV-10 of $439,568 thousand, "present value of future income tax discounted at 10%" of zero — and a standardized measure of $439,568 thousand. Identical to the dollar. The reason is the loss carryforwards: on the reserves currently on the books, the company expects to pay no federal income tax at all over their entire producing life. It is the single cleanest illustration of what the tax shield is worth to this business — and a reminder that a comparison of "standardized measure" across E&P peers quietly compares companies that pay taxes with one that does not.

Most anti-takeover defenses exist to protect a board. SandRidge's exists to protect a tax number. The company runs a Tax Benefits Preservation Plan — a shareholder rights plan whose sole purpose is to stop anyone from buying so much stock that the $1.6 billion of loss carryforwards get cut down by Section 382 of the Internal Revenue Code. The annual report lists it under the risks, with unusual candour: "We have adopted a Tax Benefits Preservation Plan, which may discourage a corporate takeover."

The mechanism is a quirk of U.S. tax law: if the holdings of the "five-percent stockholders" rise by more than 50 percentage points within three years, an "ownership change" is triggered — and the NOLs are throttled. So the very thing that would normally please a shareholder (a bidder buying in) is the thing that could destroy the company's most valuable asset. The plan was approved at the annual meeting on May 25, 2021 and amended twice. On August 5, 2025, the board even had to grant a general waiver under the plan just so shareholders could reinvest their dividends into new shares under the newly adopted dividend reinvestment plan — without accidentally tripping the tax wire.

Fractyl's $30.1 million of notes payable (March 31, 2026) are not the polite, unsecured kind. The annual report spells out what stands behind them: the obligations under the Credit Agreement "are collateralized by substantially all of its assets, including its intellectual property, but excluding certain customary and agreed upon assets." For a company whose only real asset is the intellectual property — the Revita patents, the Rejuva gene therapy platform — that sentence describes the whole estate.

Two further details deserve daylight. First, the Credit Agreement carries a minimum liquidity covenant requiring a $10.0 million cash balance; the company was in compliance as of December 31, 2025, but the 10-K warns that "without additional financing, we may not be able to comply with the minimum liquidity covenant related to our 2023 Notes by the end of 2026" — that covenant is a named ingredient of the going-concern conclusion. Second, a second tranche was quietly forfeited: "Due to a shift in business strategy to include the weight maintenance study, we decided not to pursue the milestones required to access the second tranche. As a result, the second tranche was not extended." The pivot that produced the good REMAIN-1 data also closed a door on the money.

Fractyl is usually described as pre-revenue, which is almost true but misses something better. The company did sell Revita: a limited pilot commercial launch ran in Germany from the first quarter of 2023. The annual reports let you add up the whole adventure: revenue of $120 thousand in 2023, $93 thousand in 2024, and $0 in 2025. Total lifetime revenue: roughly $213,000. Gross profit was $43 thousand in each of 2023 and 2024 — the cost of goods sold ate more than half.

The stop was deliberate, not a market failure: the Strategic Reprioritization of January 31, 2025 paused the Revita programs for type 2 diabetes, including the Germany Real World Registry study, and the pilot went with them. Still, the number is worth keeping: a company valued in the region of $145 million (data as of July 15, 2026) has, across its entire existence, invoiced customers for about the price of a parking space in Manhattan — while spending $74.5 million on research and development in 2025 alone. That is not a scandal. It is the honest shape of clinical-stage medtech, and it is exactly what the word "pre-revenue" hides.

In August 2022 — still flush, two years before the IPO — Fractyl signed a lease for 78,000 square feet of office and laboratory space at 3 Van de Graaff Drive in Burlington, Massachusetts. The term runs 128 months, expiring in June 2034, and the annual report puts the total bill in one line: total lease payments of $59.3 million. A five-year renewal option sits on top, not included in that figure.

Hold that against the rest of the balance sheet and the proportions turn strange. As of March 31, 2026 the company had $63.2 million of cash and 100 full-time employees — 82 of them at that headquarters. The operating lease liabilities on the books ($5.1 million current plus $25.5 million long-term = $30.7 million) are larger than the company's notes payable ($30.1 million). Put plainly: Fractyl has committed to paying rent through 2034 on a building it has funded through early 2027. Long leases are normal for lab space, and floor plans cannot be resized quarterly — but a fixed ten-year obligation is the one cost a going-concern company cannot cut by pausing a study.

Byline likes to tell a growth story, and the numbers back it: total assets of $9.65 billion at the end of 2025, five bank acquisitions since 2016. Which makes one sentence in the annual report read strangely: "Since our recapitalization in June 2013, our branch network has been reduced from 88 to 45, including 24 branches added through acquisition."

Do the arithmetic and it is starker than it sounds. The bank started with 88 branches, added 24 through takeovers — and still ended at 45. That means roughly 67 branch locations were closed in twelve years, while the balance sheet grew and grew. Deposits per branch now stand at $169.9 million (2023: $149.5 million). It is a reminder of what Byline actually was before it was a growth stock: the rebuilt remains of a distressed Chicago bank group, recapitalized in 2013 and pruned hard. The efficiency ratio of 51.83 percent that the market likes so much did not fall out of the sky — a good part of it was cut out with a pair of shears.

The single largest shareholder of Byline Bancorp is not BlackRock and not Vanguard. It is MBG Investors I, L.P., holding 11,875,953 shares, or 26.15 percent of the company (proxy statement, ownership as of April 8, 2026) — more than five times the stake of BlackRock (5.06 percent) or Dimensional (5.11 percent). The address on file is not in Chicago and not in Delaware: 365 Bay Street, Suite 800, Toronto, Ontario.

The construction underneath is the curious part. Voting and investment power over that quarter of the bank rests solely with Mr. Antonio del Valle Perochena as general partner — who, the proxy discloses, "owns 4.35% of the partnership interests of MBG Investors I, L.P." and "disclaims beneficial ownership of such shares except to the extent of his pecuniary interest therein". So one person controls 26.15 percent of a listed American bank through a vehicle in which he holds a 4.35 percent economic stake. None of this is hidden — it is printed in the proxy every year, and it traces back to the 2013 recapitalization that created today's Byline. But it is worth knowing before you assume the free float is what it looks like: our data shows about 28.8 percent of the shares in insider hands.

Byline Bank runs 45 branches, most of them in Chicago neighborhoods, and takes deposits from dry cleaners, dentists and diners. It also runs a business called sponsor finance — and as of December 31, 2025 it had $805.9 million outstanding there, more than a tenth of the entire loan book. Sponsor finance means: senior secured loans to companies that private equity firms have bought, to finance the buyout itself. The target size is spelled out in the annual report: portfolio companies with EBITDA "generally between $2.0 million and $10.0 million".

Read that again. These are leveraged loans to small companies whose owners bought them with debt — the classic lower-middle-market LBO. The bank does not hide it; it is proud of it, and writes that "we believe our expertise in this niche is unique for a bank our size". That is probably true, and it is also the point: a $9.7 billion bank is doing something most of its peers do not do, at a scale that matters. It helps explain the loan yield of 7.07 percent — and it is the part of the portfolio that would be tested first if small-company earnings turned. Whoever buys BY for the sleepy Chicago branch network is also buying an LBO lender.

ProPetro's new PROPWR business line is the reason the company is spending more than it earns. The order book and the contract book, however, are two different numbers, and the filing prints both a paragraph apart. As of March 31, 2026, PROPWR had "total committed capacity of approximately 240 megawatts and total delivered or on-order generation capacity of approximately 550 megawatts".

So for 310 megawatts — about 56 percent of the equipment on order — no customer contract existed at the reporting date. The company says it "continues to actively negotiate additional contracts amid increasing demand for power solutions", and in a market where data centers are hunting for electricity that is a plausible bet rather than a reckless one. But it is a bet, and it is being placed with borrowed and newly issued money: the same filing discloses a Caterpillar framework agreement for a further 1.5 gigawatts with a minimum purchase obligation of about $1,106.0 million. Building capacity ahead of demand is how you win a land grab — and how you end up owning very expensive idle iron if the demand shows up somewhere else.

Buried in Item 1 of the annual report is a transaction that reads like a handshake deal at a company of this size. On November 1, 2024, ProPetro sold its cementing business in Vernal, Utah "to a business owned by a former employee as part of a strategic repositioning". The consideration was not cash but a promissory note for $13.0 million — an IOU. ProPetro recorded an $8.2 million gain on the disposal.

The story ends well: the filing confirms the note "was fully repaid with interest in December 2025", and the buyer is no longer affiliated with the company. It is disclosed cleanly, under related-party transactions, exactly as it should be. Still, it is a useful window into how this industry actually clears its shelves: a business line the company no longer wants does not find a strategic buyer at a premium — it goes to someone who used to work there, financed by the seller. Keep it in mind when reading about the equipment ProPetro is buying: the resale market for used oilfield assets is thinner than the balance sheet suggests.

The 2024 write-down of ProPetro's conventional Tier II diesel-only fracturing pumps is usually quoted as one number: $188.6 million. The footnote holds the more instructive pair. On the September 30, 2024 measurement date, the carrying value of the Tier II Units was approximately $252.4 million — their estimated fair value was $63.8 million. About three quarters of the book value evaporated in one assessment.

How the appraisers got there is the part worth reading. Roughly 95 percent of the equipment was valued using the market approach, drawing on "research gathered from third party auctioneers" — and a key assumption was the "declining desirability for conventional diesel equipment due to emissions and fuel efficiency challenges". Translated: the auction houses were asked what diesel frac pumps still fetch, and the answer set the price. ProPetro also shortened the remaining useful lives of those units to no later than the end of 2027. In an industry where the customer decides which engine is acceptable, equipment does not wear out — it goes out of fashion, and the depreciation schedule finds out last.

Two sentences in the same filing, pointing in opposite directions. First: ProPetro's board extended a share repurchase program in May 2025 permitting the repurchase of up to $200 million of stock through December 31, 2026; as of March 31, 2026, $89.2 million remained authorized. Second: "During the three months ended March 31, 2026, the Company made no share repurchases under the share repurchase program as it prioritized the scaling of its PROPWR business line." The same sentence appears for the full year 2025 — not a single share was bought back in either period.

What the company did instead is the mirror image: in January 2026 it sold 17.25 million new shares at $10.00 apiece, raising about $163.4 million net — to fund, per the filing, "growth capital for additional power generation equipment". Share count rose by roughly 17 percent. A buyback authorization that stays untouched is not a scandal, and prioritizing cash over cosmetics is defensible. But it is worth seeing plainly: a company that formally still holds permission to shrink its share count spent the period expanding it — and every future per-share figure is divided by the larger number.

When Iovance acquired the worldwide rights to Proleukin in 2023, it also inherited the seller's manufacturing and supply agreement — including its purchase quantities. Those quantities are now larger than the business needs. The 2025 annual report books the consequence plainly: a $9.5 million increase in the excess and obsolescence reserve, "primarily related to excess Proleukin inventory resulting from a manufacturing and supply agreement inherited in the Acquisition for which we cannot yet fully utilize the required purchase quantities".

Translated: the company is contractually obliged to buy a drug in volumes it cannot currently turn into revenue, and writes down the surplus. It is a small line inside a $173.2 million cost of sales, and it has a benign reading — Proleukin is part of the Amtagvi regimen (roughly 15 vials per infusion), so faster Amtagvi uptake would consume the surplus. But it is a reminder that the 2023 acquisition did not only buy rights; it bought obligations, and the write-down is what a bet on a launch pace looks like when the launch runs slower than planned.

The company that makes the world's first approved TIL therapy for a solid tumor did not start in a laboratory. EDGAR keeps the trail under the same registrant number (CIK 0001425205): the entity was named Freight Management Corp until 2010, became Genesis Biopharma, Inc. in 2013, then Lion Biotechnologies, Inc. in 2017 — and only afterwards Iovance Biotherapeutics, Inc.

Three renamings and a complete change of purpose, all inside one filer identity. This is a common origin for U.S. small caps — a listed shell is a cheap way onto the exchange — and it says nothing about the quality of today's science, which is approved by the FDA and documented in five-year follow-up data. But it is worth knowing that the company whose accumulated deficit now reads $2.8 billion (December 31, 2025) began its listed life as a freight business, and that the corporate history in EDGAR is considerably longer than the therapy's.

Through the guidance cut, the 19 percent workforce reduction and the centralization of all manufacturing, Iovance has been led by an interim chief executive — and the same person is the company's General Counsel. Frederick G. Vogt, Ph.D., J.D., was already introduced as "Interim President and Chief Executive Officer" in the annual results release of February 27, 2025. He still signed the quarterly report as of March 31, 2026 — filed on May 7, 2026 — as "Interim Chief Executive Officer and President, and General Counsel (Principal Executive Officer)".

That is at least fifteen months of interim leadership at a company executing the first commercial launch of a therapy class that has never been launched before. The annual report 10-K for 2025 carries the same signature: the person certifying the numbers as principal executive officer holds the title General Counsel. None of this is improper, and a lawyer with a doctorate in chemistry is not an odd choice at a cell-therapy company. But when investors ask why a launch forecast was off by nearly half, "who is actually running this, and for how long" is a fair question to have on the list.

Iovance funds itself by selling shares, and in 2025 it sold a lot of them: 101,899,334 for $306.3 million net. In August 2025 it signed an amended "at the market" sale agreement with Jefferies for up to $350.0 million more. What almost nobody puts next to that number is the ceiling in the company's own balance sheet: the certificate of incorporation authorizes 500,000,000 shares — and as of April 15, 2026, 446,502,396 were already outstanding.

That leaves roughly 53.5 million shares of headroom, about 11 percent of the authorized total. At the price level of early July 2026 (about $4.20; data as of July 8, 2026), those remaining shares are worth on the order of $225 million — less than the $350 million the ATM program is sized for, and less than a single year of the company's recent operating cash burn ($302.4 million in 2025). Iovance is not out of room, and stockholders can authorize more shares at an annual meeting. But the arithmetic means the current financing path has a visible end date, and extending it is a vote, not a decision management makes alone.

Weibo breaks out who buys its advertising, and one line in that table quietly collapsed. Advertising and marketing revenue from SINA — the controlling shareholder, which holds 62.5 percent of the voting power — ran at US$45.3 million in 2023, then US$25.0 million in 2024, and just US$14.3 million in 2025. That is a 68 percent decline in two years, from the one advertiser that sits on the other side of the boardroom table.

Read alongside the second related-party line, it becomes a picture: Alibaba's spending rose from US$116.8 million to US$173.8 million over the same stretch, while third-party advertisers spent less each year. Weibo's ad book is quietly rearranging itself around a single large outside customer, even as its own parent pulls back.

Nothing here is improper, and related-party advertising is disclosed exactly as it should be. But it is worth noticing that the shareholder with the best view of the platform — and the least need for a sales pitch — has been buying steadily less of what it sells.

Weibo has two classes of ordinary shares: Class A carries one vote, Class B carries three. All Class B shares belong to SINA. The effect, as of March 31, 2026: SINA owned "approximately 35.7% of our total issued and outstanding ordinary shares and 62.5% of the voting power". Every ADS on the Nasdaq is a Class A share — the one-vote kind.

That majority has a formal consequence most ADS buyers never read. Because SINA controls more than half the votes, Weibo qualifies as a "controlled company" under The Nasdaq Stock Market Rules — and it says plainly that it uses the exemptions this brings: it is not required that "our director nominees must be selected or recommended solely by independent directors", nor that it maintain a nominating committee composed entirely of independent directors.

There is a second layer beneath it. The risk factors disclose that SINA's shares in Weibo are pledged as collateral for a loan facility: if SINA defaults, "the security agent may dispose of or cause SINA to dispose of the pledged shares" — and any acquirer "will be entitled to exercise the voting control", possibly "in a manner that could vary significantly from that of SINA". Control of Weibo can therefore change hands through a loan agreement its outside shareholders are not party to.

In December 2025, Weibo's board authorized a share repurchase program of up to US$200 million, running until December 31, 2026. It arrived at a convenient moment: the stock trades at about half of book value, and the company holds more cash than its entire market value. A buyback here would be, arithmetically, one of the cheapest available uses of money.

Item 16E of the annual report — the item that exists precisely to disclose issuer purchases — closes the topic in eight words: "We did not make any share repurchase in 2025." The program was authorized in December, so that leaves only about three weeks of the year; but the interim report (6-K) for the first quarter of 2026, published May 28, 2026, does not report any repurchases either.

The same annual report also warns, in its own risk factors, that "We cannot guarantee that any share repurchase program will be fully consummated." Authorizations are not purchases. For a stock whose bull case rests substantially on capital returning to shareholders, the difference between an announced $200 million and a spent $0 is the whole case.

Weibo cannot legally own its own China business: PRC law restricts foreign ownership of internet content licenses. So the licenses, the domain names and the operating permits sit inside two companies — Weimeng and Weimeng Chuangke — that the listed Cayman holding company does not own a single share of. Who does own them? The annual report names them, by name.

Weimeng belongs to four private individuals — "four PRC employees of us or SINA, namely, Yunli Liu, Wei Wang, Wei Zheng and Zenghui Cao", holding 29.70, 29.70, 19.80 and 19.80 percent — plus a third-party minority holder with 1 percent. Weimeng Chuangke belongs to two of the same people, Yunli Liu and Wei Wang, at 50 percent each. These entities produced 85.9 percent of Weibo's 2025 revenue.

What ties them to the shareholders in New York is not ownership but a stack of contracts — loan agreements, share transfer options, powers of attorney, even spousal consent letters, because a divorce could otherwise put a license stake in play. The annual report concedes the arrangement "may not be as effective as direct ownership in providing us with control over the VIEs" and notes the contracts "have not been tested in court to date". Whoever buys the ADS is trusting four named employees, their spouses, and a set of agreements no judge has ever reviewed.

A glance at the balance sheet looks alarming: cash fell from $190.7 million (December 31, 2024) to $8.3 million (December 31, 2025) and to $5.5 million as of March 31, 2026 — for a company the market valued at roughly $1.7 billion (July 16, 2026). Almost every distress screen would flag that.

The footnote turns it around. On March 25, 2025 the company used the money to redeem $181.4 million of its 10.75 percent senior secured notes at 101 percent of principal — "expected to generate an annual interest expense savings of approximately $19.5 million". By year-end 2025 the only debt left was $3.8 million of vehicle finance leases, with $0 drawn on a $175 million revolver and total liquidity of about $183.3 million. That is also why the scanner shows an Altman Z-score of 6.66, comfortably in the safe zone, right next to a fundamental grade of D: a company with no debt is hard to bankrupt; it is not thereby profitable.

The debt-free interlude lasted about nine months. In the first quarter of 2026 the company drew a net $30 million on the revolver "to fund growth of the WHS business segment", spending $45.5 million on capital expenditures in three months. And the biggest bill has not arrived: the AI Infrastructure Community alone requires $200 million to $210 million of capital investment net of customer advances, roughly 95 percent of it in 2026 — against $5.5 million of cash and $145 million of remaining revolver. Borrowings run at Term SOFR plus 4.25 to 4.75 percent, and the facility matures February 1, 2028. Whoever cheered the deleveraging in 2025 should note that the order book is about to re-lever the same balance sheet.

When the PCC contract ended on February 21, 2025, Target Hospitality did something unusual for a lessor: it kept the property. The communities that served the contract — Pecos (2,000 beds), Pecos Blue Lodge (1,000), Lodge 118 (1,402), Delaware Lodge (425), Pecos Trail Lodge (308) and Skillman Station Lodge — stayed on the books. The annual report framed it as an option: the company "retained ownership of these assets, enabling the Company to continue utilizing these modular solutions and real property to support customer demand across its existing operating segments". The quieter half of the same paragraph: "The Company is actively engaged in remarketing the remaining assets." Remarketing is the polite word for looking for a tenant — and meanwhile depreciation of specialty rental assets ran on almost unchanged at $57.2 million in 2025 (2024: $57.2 million), on buildings whose revenue had collapsed.

The follow-up, buried in the quarterly report, is the part almost nobody read — and it is the closest thing to a happy ending in this filing: "During the latter part of the current quarter, many of these assets were re-contracted or redeployed to support growth in the WHS segment." In March 2026 the company signed a Pecos Power Community agreement — 26 months from April 2026, a committed minimum of 400 rooms per night, about $23 million — to house workers building a natural gas power plant, in the same town where the idle beds sit. The mothballed migrant-housing community is being re-let to the power buildout. The remaining undeployed or uncontracted leased assets are to be demobilized over the next two quarters, at a cost the company flags but does not size. Modular really does mean modular: the same rooms, a different boom.

Target Hospitality looks like an ordinary Nasdaq small cap. It is not. Two entities named Arrow Holdings and MFA Global S.à r.l., both controlled by the London private equity firm TDR Capital, together held about 65 percent of the common stock as of December 31, 2025. The annual report is blunt about what that means: TDR "may have substantial control over matters requiring approval by our stockholders" — and adds the sentence that matters most: "TDR Capital may have interests that are different from those of other stockholders."

In April 2026 those interests became visible. On April 21, 2026 the two entities agreed to sell 7,000,000 shares in a registered secondary offering at $14.00 per share; the underwriters exercised their option in full, so 8,050,000 shares changed hands when the offering closed on April 23, 2026. The company itself sold nothing and received no proceeds — this was the owner cashing out, not the business raising money. On roughly 100.2 million shares outstanding, that trims TDR's stake to about 57 percent.

Now put the calendar next to it: the Data Center Hub (about $550 million) had been signed in March 2026, and in May 2026 — a few weeks after the sale closed — the company announced the AI Infrastructure Contract worth more than $750 million. The stock traded around $17.05 in mid-July 2026 (July 16, 2026). Nothing here suggests impropriety, and a private equity firm reducing a decade-old position is the most ordinary thing in finance. But it is worth knowing that the best-informed shareholder in the company chose to sell a chunk at $14.00 into the very rally the order book was fuelling.

Buried in the share capital note is a combination that does not obviously belong together. In fiscal year 2025 Gorilla repurchased 268,411 shares at a weighted average price of $14.55, for a total of $3,904,123 — treasury stock. In the same year, the company burned $28.7 million of cash in operating activities, and the number of shares outstanding rose from 18,058,135 to 26,188,972, largely through 6,019,162 shares issued on warrant exercises.

So the company was buying shares back with one hand at $14.55 while issuing far more of them with the other, and funding neither out of operations — financing activities brought in $101.2 million that year. The prior year had the same shape at a very different price: 1,103,618 shares repurchased at a weighted average of $3.29.

Buybacks are usually a signal that a company has spare cash and thinks its stock is cheap. Here the cash came from selling stock, and the repurchase price was more than four times what the company had paid a year earlier. Whatever the intent — treasury shares can also be warehoused for employee plans, which the note explicitly allows — it is not the signal a buyback normally sends.

In September 2025 Gorilla announced a three-year, $1.4 billion agreement with Freyr, a Singapore-based infrastructure platform company, to build a network of AI-powered data centers across Indonesia, Malaysia and Thailand. To put that number in perspective: it is roughly fourteen times Gorilla's entire fiscal year 2025 revenue, and about four times the company's market value of some $330 million (data as of July 15, 2026).

The annual report closes the paragraph describing it with one short sentence: "No revenue was recognized under this agreement in fiscal year 2025." The filing is also candid about the structure — the agreement "provides parameters for cooperation on multiple projects, each of which to be governed by a separate project plan, detailing scope and commitments including payment schedule, payment term and delivery schedule". In other words: it is a framework, and the actual work has to be contracted project by project.

None of this is improper, and framework agreements are ordinary in infrastructure. But a headline number fourteen times larger than annual revenue, with no revenue attached to it yet, is the kind of figure that travels much further than the sentence underneath it.

The segment note of the annual report contains a sentence that quietly undoes the map most investors draw in their heads. Gorilla reports approximately 99 percent of total revenue as coming from Taiwan — for 2025, 2024 and 2023 alike. In the very same report, the major-customer table shows that $77.5 million of the $101.4 million was paid by a customer located in Egypt. Both statements are true, and the note explains why:

"Revenues by geography are determined based on the region of the Group's contracting entity, which may be different than the region of the customer."

So the geography table measures where Gorilla's own signing entity sits, not where the money comes from. An investor who skims the regional breakdown to check for country risk would conclude this is a Taiwanese business with Taiwanese counterparties — and would miss the Egyptian government exposure entirely, along with the Egyptian pound it is billed in. The information is all there, in two tables three pages apart. It just never meets on the same line.

The human-capital section of the annual report contains a number that puts the valuation in perspective: as of December 31, 2025 — after the combination with Beckley Psytech — AtaiBeckley had 99 full-time employees and 18 contractors or consultants doing regular work. Of those 99, 66 work on research and development, and 23 of the employees belong to the Nualtis subsidiary, which does the licensing and research-services business that produced almost all of 2025's $4.1 million of revenue.

Set against a market value of roughly $1.83 billion (data as of July 16, 2026), that is about $18 million of market value per employee. This is not a criticism — it is how clinical-stage biotech works: the value sits in molecules, patents and trial data, and the expensive work is bought in from contract research organizations rather than staffed in-house. But it is worth knowing what you are buying. There is no factory, no sales force and no service organization here. There are 99 people, four drug candidates and a set of trial results.

Anyone who looks up ATAI at the SEC today lands on CIK 0002081043, AtaiBeckley Inc., a Delaware corporation whose filing history begins in 2025. The company's first two decades of paperwork are not there. They sit under a different number — CIK 0001840904, which carried the names ATAI Life Sciences B.V., ATAI Life Sciences N.V. and finally Atai Beckley N.V.

The reason is a three-step move the 10-K describes plainly: on December 30, 2025 the Dutch company merged into a Luxembourg entity, which converted the same day into a Delaware corporation named AtaiBeckley Inc., which "became the successor issuer to Atai Beckley N.V. pursuant to Rule 12g-3(a)". The old registrant then filed a Form 15-12G on December 31, 2025 and switched off its reporting duty. Nothing here is hidden or improper — redomiciliations are routine, and the successor picked up the reporting obligation without a gap. But it is a genuine research trap: a screener or a database that keys on the old CIK will show a company that stopped filing on New Year's Eve, and an investor comparing "the last five annual reports" has to cross two archives to find them.

Under "Sources of Liquidity", between the pipeline and the at-the-market program, the quarterly report as of March 31, 2026 lists something you do not expect in a drug developer's balance sheet: Bitcoin. The wording is the company's own — "A potential source of non-dilutive funding resides in our investment in digital assets, subject to market conditions, volatility, and price fluctuations." Market value as of March 31, 2026: $6.8 million.

The position is small next to $209.9 million of cash and short-term securities — but it is not free. In 2025 the "change in fair value of digital assets, net" cost $1.2 million, booked straight through the income statement. Listed right beside it is a second non-dilutive reserve of a more conventional kind: the company's stake in the American depositary shares of COMPASS Pathways, worth $20.9 million on the same date. Together, roughly $27.7 million of the liquidity story rests on two prices AtaiBeckley does not control — one of them the most volatile asset class there is. Whoever counts the runway should know which parts of it can move on their own.

Amprius sources its anode material exclusively from the Chinese partner Berzelius (Nanjing), which the company itself describes in its annual report (10-K) as a former related party. Per the 2025 annual report, Dr. Kang Sun — Amprius' former chief executive, today an executive advisor and a member of the company's own board of directors — also sits on the board of Berzelius. That puts an Amprius director on both sides of a supply agreement that contains no fixed prices.

Inside the insurer's income statement sits a chapter from the green investment wave: as part of its ESG initiatives, Citizens had invested in BlackRock's "Global Renewable Power Fund III" — a $4.8 billion flagship renewables fund. In December 2024, BlackRock announced a substantial write-down, triggered by the collapse of its two key holdings, Northvolt and SolarZero.

At Citizens this landed as a $3.3 million write-down in the fourth quarter of 2024 and another $5.2 million in 2025 — $8.5 million combined, more than half a year's net income. The insurer did not sell the position; in 2025, fair value gains in other fund investments offset the effect. The find is remarkable all the same: a conservative life insurer whose portfolio is 89 percent bonds picked up its largest single loss of recent years through an ESG prestige project.

In 2018 Citizens took former employees and independent consultants to court, alleging they had taken confidential information to compete unfairly. Six years later the jury turned the tables: Citizens is to pay former consultants Alexis Delgado and Carlos Nalsen Landa about $1.3 million ("money had and received" — withheld commissions), and the trial court additionally awarded defendants Michael P. Buchweitz and Randall Riley roughly $3.5 million of their legal fees. The judge signed the Final Judgment on August 16, 2024.

Citizens has posted an appeal bond and appealed (pending before the Court of Appeals, Third District of Texas) and considers both awards wrong. The $3.5 million in legal fees was accrued through the income statement in 2024. For investors, the punchline remains: a lawsuit the company itself initiated could end up costing it almost $5 million — no footnote for a business earning about $15 million a year.

Citizens has two share classes with a power split straight out of a founder-control textbook: the Class B shares carry, per the articles of incorporation, the exclusive right to elect a simple majority of the board of directors; in exchange, Class A shares receive twice the cash dividend per share — should one ever be paid. For decades this construction was the key to controlling the company.

Today all 1,001,714 Class B shares sit in treasury — not a single one is outstanding, and the annual report notes that since 2021 Citizens has been a non-controlled company "for the first time in over 20 years." The super-voting class still exists, disarmed in the company's own vault. Whoever buys the stock should know: the charter that allows a takeover of the board via Class B has not been abolished — merely mothballed.

Citizens runs a mechanism you will find at almost no other listed company: policyholders can assign their policy dividends through the "Citizens, Inc. Stock Investment Plan" (SIP) to Computershare Trust Company, which uses them to buy Class A shares of the very insurer that wrote the policy. The result sits in the risk section of the annual report: almost 81,000 shareholders, roughly 40 percent of them holding fewer than 100 shares — many in Latin America and the Pacific Rim, where the policies are sold.

The annual report names the side effects itself: because of the many overseas micro-shareholders, turnout at annual meetings is so low that resolutions requiring a majority of all outstanding Class A shares — a merger, for instance — "may be difficult to get approved." And: the Class A stock is not registered as a security in any foreign country; a foreign securities regulator could bar its sale through the SIP in its jurisdiction. An insurer whose customers double as its free float — that is remarkable, and it only appears in the fine print.

Fiscal year 2025 ended with a net loss of $8.1 million for Clearfield — caused by the impairment on the Nestor Cables business. The board’s answer: on November 20, 2025, nine days after the Nestor sale and five days before the annual report was filed, the share repurchase program was increased from $65 million to $85 million.

The pace behind it is remarkable: in fiscal year 2025, Clearfield bought back 550,766 of its own shares for roughly $16.5 million, and another $12.6 million followed in the first six months of fiscal year 2026 — against a market value of roughly $550 million most recently (data as of July 10, 2026), one of the more aggressive buyback ratios in small-cap land. The debt-free balance sheet with $147 million in cash and investments makes it possible. You can read that as management conviction — or as an admission that the company currently has no better use for its money than its own stock.

The BEAD program Clearfield’s customers are waiting for comes with a condition spelled out in the risk factors of the annual report (10-K) for fiscal year 2025: products must meet the Build America, Buy America (BABA) domestic content requirements for customers to qualify for grant funding. If products fail those requirements, the report warns of "lost sales, lost business opportunity, breach of warranty claims, and damage to our reputation and customer relationships."

The piquant detail: per the same annual report, substantially all final build and assembly happens at the plants in Brooklyn Park, Minnesota, and Tijuana, Mexico. So the company must prove, product line by product line, that enough America is inside when America pays — a compliance hurdle hardly any investor has on the radar when hearing "BEAD winner."

The subsequent-events footnote of the annual report (10-K) for fiscal year 2025 records the end of an acquisition in a single dry sentence: on November 11, 2025, Clearfield sold its Finnish subsidiary Clearfield Finland Oy — the parent of cable maker Nestor Cables Oy — for $1 in cash. On top, the company contributed $5,785,000 of intercompany receivables to the divested business — money Nestor Cables actually owed Clearfield.

Clearfield had bought the company in July 2022, at the peak of the fiber boom: 7.9 million euros ($8.0 million) for the shares plus $7.8 million to repay Nestor debt, funded by a $16.7 million draw on the credit line. In the fourth quarter of fiscal year 2025 came a total impairment of $16,589,000 — and then the one-dollar deal. A complete European adventure that shrank from strategic beachhead to footnote in barely three years.

Open Turtle Beach's SEC file and you find a different name under the same registration number (CIK 0001493761): Parametric Sound Corp — a company founded in 2010 around directed ultrasound audio that merged with the headset maker in a 2014 reverse takeover and adopted its name. Today's company is incorporated in Nevada, headquartered in San Diego and employs just 290 people.

The ticker has a history too: for more than a decade the stock traded under the telling symbol "HEAR"; on January 7, 2025 it switched to "TBCH" — according to the press release, because after the PDP acquisition the company wants to be more than audio: a supplier of controllers, keyboards and simulation gear. Whoever digs for old price series or forum threads has to search under two symbols — and whoever "knows" this stock often knows only one of its moults.

On June 9, 2025, the board adopted a Rights Agreement — a "poison pill" in market jargon: whoever crosses 10 percent of the shares without board approval lets all other shareholders buy in at a steep discount, drastically diluting the acquirer. The trigger is remarkably low — 15 to 20 percent is more common — and the annual report itself warns that the construction could hurt the share price and dilute shareholders if it were ever triggered.

The context makes it interesting: Turtle Beach has a long activist history, the risk section states explicitly that activist stockholders have attempted to assert influence and may do so again — and only two months after the pill, of all people the activist on the board, William Wyatt (Donerail), privately added $10 million to his position. Whoever buys into this small cap should know: control here is actively guarded.

On August 14, 2025, DC VGA LLC ("Diversis") — the private equity firm that sold Turtle Beach the controller maker PDP in 2024 and received stock in return — sold a block of almost 1.4 million Turtle Beach shares. Buyer number one: the company itself (694,926 shares for $10.0 million). Buyer number two: TDG CP LLC ("Donerail") — an entity of William Wyatt, a member of the board of directors (693,962 shares, likewise $10.0 million). The price for both: the 30-day volume-weighted average of $14.41 per share.

The piquant detail: the selling side sat on the board as well — Dave Muscatel, a director at the time, is affiliated with Diversis. The transaction was approved by the audit committee, staffed solely with independent directors; everything is disclosed. What remains is a rare constellation: the activist who once applied pressure from the outside now adds to his stake from the inside — via the same agreement through which the PDP seller exits.

The annual report for 2025 contains a sentence you do not expect from a headset maker: control weaknesses in the supply chain meant that the misappropriation of $3.4 million of inventory that was in transit to customers in 2024 was not detected and prevented in time — including insufficient oversight of the third-party service provider running the ordering process. The weaknesses are considered remediated as of the end of 2025.

The counter-entry is remarkable: in 2025 Turtle Beach booked a $9.4 million insurance recovery for the transit inventory lost primarily in the fourth quarter of 2024 — noticeably more than the quantified misappropriation, and a good third of the 2025 operating profit. Without this one-off, the year's earnings would look considerably paler.

Pediatrix retired the Mednax name in June 2022 — in its marketing. In the technology it lives on: the machine-readable version of the annual report (10-K) for 2025 still declares its company-specific accounting tags under the XBRL namespace "www.mednax.com" — the data address under which market software files the Pediatrix numbers still carries the old corporate name.

The practical consequences are minor, but it illustrates how stubbornly a retired identity clings to a company: our data provider and the Reddit tracker ApeWisdom also still listed the stock as "Mednax" in July 2026. Whoever searches only for the old name ends up vetting the wrong company — or none at all.

Pediatrix operates in 37 states — but the revenue is anything but evenly spread: Texas alone accounted for about 32 percent of net revenue in 2025, and the five largest states (Texas, Florida, Georgia, California, Washington) together for 64 percent. The annual report names the risk itself: adverse developments in these states — healthcare reforms, reduced Medicaid reimbursements, tighter eligibility, but also weather events and natural disasters — could materially hurt the business.

The payor mix makes the concentration spicier: in Texas especially, Medicaid is a central payer for childbirth and neonatal care. Whoever buys Pediatrix is also betting on the healthcare politics of a single state.

The payor-mix table in the annual report (10-K) for 2025 contains a gap you have to read twice: government programs — mostly Medicaid — accounted for about 53 percent of Pediatrix's gross billings but only 24 percent of net revenue. Translated: for more than half of the care delivered, the paying party is one whose rates are, per the report, "substantially less" than those of commercial insurers.

The same gap explains a balance-sheet position hardly any investor recalculates: $1.15 billion of gross accounts receivable stood against only $229.7 million of net receivables at the end of 2025 — the rest is expected contractual adjustments and write-offs. Per the report, a shift of just 0.5 to 1.5 percentage points in the estimated collection rate moves $5.5 to $16.5 million of revenue.

Read the footnotes and the Mednax era is still sitting in the middle of today's balance sheet: per the annual report (10-K) for 2025 the company continues to carry retained professional liabilities for its anesthesiology and radiology medical groups divested in 2020 — malpractice exposure from physicians who have been working for other owners for five years. At the same time, Pediatrix had to write down a retained interest in the divested anesthesiology group by $7.9 million in 2025.

But the same past also paid out: in the third quarter of 2025 a company in which Pediatrix still held an investment from an earlier sale was acquired — the company received $30 million in cash and booked a net gain of $20.9 million on "investments in divested businesses". Whoever buys the stock also buys a small portfolio of leftovers and liabilities from businesses sold long ago.

W&T Offshore trades on the NYSE under the symbol WTI — the same symbol under which the entire financial world knows the U.S. benchmark crude West Texas Intermediate, whose price runs through the news every single day. The punchline: in the glossary of its own annual report, the company defines the term itself — "WTI. West Texas Intermediate grade crude oil." — as a crude oil grade, not as the company.

For investors this is more than an anecdote: mentions of "WTI" in forums, news tickers and social media scanners may refer to the oil benchmark, not the stock — attention metrics for this ticker are systematically noisy. If you mean the stock, search for "W&T Offshore"; if you just google "buy WTI," you may end up with something in your portfolio quite different from what you had in mind.

W&T Offshore has been paying a dividend again since November 2023: $0.01 per share per quarter — four cents a year, roughly $6.4 million in total (2025). The curious part: the balance sheet this penny is paid from shows stockholders’ equity of minus $199.8 million as of December 31, 2025 and an accumulated deficit of $780.3 million; the 2025 net loss was $150.1 million. The annual report itself cautions that there is no assurance dividends will continue.

Economically the penny is meaningless — at a price around $3.20 (data as of July 8, 2026), it yields about one percent. As a signal it is interesting: it keeps the stock inside dividend screeners and broadcasts normality where the balance sheet has the features of a workout case. A penny as a sedative.

Another footnote from the related-parties chapter: an entity owned by CEO Tracy Krohn holds interests in certain wells in which W&T Offshore itself has no ownership stake at all — and those wells are nonetheless covered under the company’s insurance policy. The CEO’s entity reimburses the company for its share of the premiums and certain administrative costs; $0.3 million flowed back that way in 2025.

Again: disclosed, reimbursed, none of it forbidden. But it shows how tightly the founder’s private assets and the corporate sphere are interwoven — the same report devotes a dedicated risk factor to the CEO’s position of power: he owns a significant portion of the shares, and his interests may diverge from those of the other shareholders.

The related-party footnotes hold a deal you would not expect at a company with 370 employees: in May 2023, W&T Offshore bought a corporate aircraft from a company affiliated with and controlled by its own Chairman, CEO and President, Tracy W. Krohn. Purchase price: $19.1 million — $9.0 million in cash, the rest through the assumption of a loan (the "TVPX Loan") that carries a stated 2.49 percent interest rate but an effective 9.0 percent (2025), and comes due in September 2026 with a balloon payment of $8.0 million.

The audit committee reviewed and approved the transaction, and everything is disclosed. The constellation remains remarkable nonetheless: a company that is simultaneously fighting surety insurers over collateral for its decommissioning obligations, and that posted a $150.1 million net loss in 2025, services an aircraft loan from a deal with its own founder-CEO.

The metrics definition in the annual report contains a sentence hardly any investor has on their radar: reported gross merchandise value (GMV) is not reduced for returns and order cancellations — and those added up to 24.2 percent of GMV in 2025 (2024: 24.4 percent; 2023: 26.4 percent). Of $2.13 billion in goods sold, merchandise worth roughly half a billion dollars went back to the sender.

Part of that is industry logic — high-priced fashion gets tried on and sent back, and The RealReal also refunds when authenticity is questioned — but it explains why the company maintains a second metric alongside GMV: net merchandise value (NMV) of $1.61 billion (2025), after returns and cancellations. And it means real costs: every returned item passes through the company's logistics and warehouses a second time. Whoever celebrates GMV growth should put the return rate next to it.

In November 2018, Chanel sued the marketplace in federal court in New York: trademark infringement, unfair competition, false advertising — at its core the allegation that The RealReal sold counterfeit Chanel goods as "authenticated". The case is in its eighth year and has taken a remarkable turn: since 2021, The RealReal has fought back with counterclaims under the Sherman Act — U.S. antitrust law — accusing Chanel of trying to obstruct the secondhand market for its own products.

Two attempts at settlement failed; after two years of fruitless mediation, the case resumed in October 2025, and a court settlement conference was scheduled for March 5, 2026. The annual report explicitly calls the outcome "uncertain". For a company whose business model rests on the word "real", a perennial lawsuit about exactly that word is more than a footnote.

The RealReal's secured 2029 notes (13.00 percent interest: 8.75 percent cash plus 4.25 percent paid in kind) officially mature on March 1, 2029. But the debt footnote contains a condition that can pull the date forward: the notes become due as early as December 1, 2027 if, at any point from then on, more than $20 million of the old 2028 convertibles are still outstanding and unrestricted cash minus that residual 2028 principal falls below $75 million.

Translated: the company has to clear the remaining $48.2 million of 2028 notes in time or hold enough cash — otherwise the largest maturity on the balance sheet moves 15 months closer. As of March 31, 2026, cash stood at $124.0 million; the cushion exists, but it is not a given. "Springing" maturities of this kind are the price perennial loss-makers pay for their debt exchanges.

In the February 2024 debt exchange, The RealReal handed its creditors warrants on 7,894,737 shares at an exercise price of $1.71 — the stock traded around $2 at the time, and lately around $12 (data as of July 8, 2026). These warrants sit on the balance sheet as a liability and are remeasured to fair value every quarter — with the offsetting entry running straight through the income statement.

The result is a seesaw investors should know: when the share price falls, the warrants lose value — and the company books a gain (Q1 2026: +$47.3 million, turning a $2.3 million operating loss into $38.9 million of net income). When the price rises, the opposite happens: the remeasurement cost $35.8 million in 2025 and as much as $68.2 million in 2024 — in the middle of the best operating year in company history. Whoever reads only the "quarterly profit" headline is mostly reading yesterday's share price, not the business.

Sweetgreen has a dual-class share structure: Class A carries one vote per share, Class B ten votes — and all Class B shares sit with the three founders, Jonathan Neman, Nicolas Jammet, and Nathaniel Ru, who per the annual report thereby hold the majority of the voting power, including control over director elections. The super-voting rights end only upon sale, by majority vote of the Class B holders, ten years after the IPO — or nine months after the death or permanent disability of the last founder.

Add a compensation package from the IPO era: performance stock units (PSUs) for the founders with a total expense of $103.0 million, recognized as stock-based compensation over the vesting period — at a company that has never reported an annual profit. Whoever buys the stock should know: having a say is structurally not part of the deal.

Since its IPO, Sweetgreen has told the story of a technology-driven restaurant company: its own app, 61.8 percent digital revenue, a robot kitchen. The footnotes of the annual report then put the research and development expense at $1.0 million for fiscal year 2025 — the same as in 2024, and less than in 2023 ($1.2 million). That is roughly 0.15 percent of revenue.

Per the report, these costs went "primarily" into developing the Infinite Kitchen — precisely the technology that was sold to Wonder at the end of December 2025. Software development for the ordering platform is partly capitalized rather than expensed, which softens the tiny number somewhat. Still, the finding stands: the tech story was mostly bought-in — and now sold-off — technology, not a lab of its own.

The related-party footnote contains a constellation hardly any investor has on their radar: Sweetgreen's three founders and its former Chief Financial Officer hold indirect minority stakes in Luzzatto Opportunity Fund II, LLC — which in turn holds indirect equity interests in "Welcome to the Dairy, LLC," the owner of the properties housing Sweetgreen's principal corporate headquarters.

For its headquarters, the company paid this entity $5.0 million in fiscal year 2025 (2024: $3.9 million; 2023: $4.2 million). All disclosed, none of it forbidden — but it remains remarkable: part of the rent paid by a loss-making company ends up, through several layers, with the founders who sit on the board and control shareholder votes through ten-to-one super-voting shares.

The sale of the "Infinite Kitchen" technology to Wonder brought Sweetgreen $186.4 million on paper — but only $100 million of it in cash. The rest came as 10,803,620 Series C preferred shares of Wonder Group with an implied value of $86.4 million. Wonder is not publicly listed, and the quarterly report says it with unusual clarity: the shares are "illiquid and fair value is not readily determinable."

The position is carried at cost, adjusted only for impairments or observable price changes. Translated: almost half the sale proceeds for the centerpiece of the automation story is a bet on a private start-up that also remains the sole supplier of the robot kitchens. Whether the $86.4 million ever turns into money is not Sweetgreen's call — it depends on where Wonder goes from here.

In January 2024, the U.S. telecom regulator FCC published a rule that would have hit the lead-generation industry's phone and text marketing hard: "one-to-one consent" — consumers would have had to consent to each individual advertiser separately instead of approving an entire partner list with one click. For a business model that resells completed comparison forms to insurers and agencies, that would have been a body blow.

Then the punch line: on January 24, 2025 — three days before the rule was scheduled to take effect on January 27 — the U.S. Court of Appeals for the Eleventh Circuit vacated the regulations. EverQuote records the episode in its annual report as an example of how close regulatory luck and business model sit here: the danger is averted, but the legal framework around automated calls (TCPA) remains a permanent construction site for the industry.

In 2021, EverQuote acquired the insurance distributor PolicyFuel and built its own direct-to-consumer agency on top of it. Four years later the chapter ended in court: in 2025 the company booked $7.9 million in legal settlement costs from litigation with the former owners of the entities acquired in 2021 — and as part of the settlement, on May 1, 2025, EverQuote sold the remaining carrier contracts of its agency, along with related software, back to those very former owners.

With that, the second diversification within two years has been unwound: 2023 brought the exit from the health insurance vertical, 2025 the end of the in-house agency business. What remains is the pure marketplace — more focused, but also more dependent on the ad budgets of a few large carriers.

Whoever buys EverQuote stock buys Class A shares with one vote — the unlisted Class B shares carry ten votes apiece. The consequence sits soberly in the annual report: directors, executive officers and large holders together held roughly 58 percent of the voting power as of January 31, 2026; Link Ventures — directly or through a voting agreement, together with Cogo Labs — controlled about 57 percent.

That makes EverQuote a "controlled company" under Nasdaq rules, exempt from core governance requirements — such as the rule that a majority of the board must be independent. The report itself warns that this concentration of voting power will "limit or preclude" other shareholders' ability to influence corporate matters for the foreseeable future. Translated: as a Class A holder you ride in the passenger seat — someone else is steering.

In August 2025, EverQuote put its freshly authorized $50 million buyback program to a rather particular use: the company repurchased 900,000 Class A shares in a negotiated transaction from Link Ventures — per the quarterly report an entity affiliated with funds advised by David Blundin, EverQuote's chairman of the board and co-founder. Price: $23.33 per share, $21.0 million in total — a discount of about 1.8 percent to the August 8, 2025 closing price. The shares were immediately retired.

None of this is forbidden, all of it is disclosed, and the discount argues for rather than against the remaining shareholders. The constellation remains remarkable nonetheless: nearly half of the first buyback program in company history went in a single block to the chairman's camp — the same camp that controls a majority of the company's voting power through voting arrangements.

In the risk chapter of Castellum's annual report for 2025 — filed on March 9, 2026 — it says, verbatim: "We had an accumulated deficit of $56,588,218 as of December 31, 2025, and we expect to continue to generate a net loss in the year ending December 31, 2025." Translated: the company expects a loss for a fiscal year that had been closed for more than two months when the report was filed — and whose audited result (a loss of $2.40 million) the very same report presents a few pages later.

Obviously, the year in the sentence is a leftover text block from the prior-year report. A slip like this usually has no consequences, and it is not even wrong in substance — a loss did in fact come out. But it is a neat exhibit of how much boilerplate sits even in audited SEC filings — and a reminder that "honest under penalty of law" does not mean "flawlessly proofread." Whoever reads filings should be able to tell text blocks from findings.

In the quarterly report (10-Q) as of March 31, 2026, Castellum lists — among the cautionary notes on forward-looking statements — an item you rarely read this bluntly from a government contractor: the "potential impact of the U.S. DOGE Service Temporary Organization on government spending and terminating contracts for convenience." That is the U.S. government's cost-cutting organization that grew out of the Department of Government Efficiency.

The mention has bite because Castellum, per its annual report, generates "substantially all" of its revenue from U.S. government entities, and three customers accounted for 73 percent of 2025 revenue. "Termination for convenience" is the government's right under U.S. procurement law to end a contract without any fault of the contractor — if it hits one of the three key customers, a fifth to a quarter of the business wobbles. The clause itself is standard; a small contractor naming the efficiency office as a risk by name is not.

Hidden in the footnotes of Castellum's annual report for 2025 is a warrant of a special kind: one covering 1,080,717 shares of common stock at a total aggregate exercise price of one single U.S. dollar — not one dollar per share, one dollar for all of them together. It was granted in connection with a $5,600,000 note payable maturing August 31, 2026; the cash consideration was, per the report, "nonsubstantive" — economically meaningless — which is why Castellum counted the shares into its loss-per-share calculation even before exercise.

As of December 31, 2025, the warrant had been fully exercised. For investors, the footnote is a case study in how nano-caps pay for credit when banks are expensive: with equity that is effectively given away. The extra million-plus shares are a small but typical building block of the dilution that keeps the stock's per-share price in the pocket-change range.

On the balance sheet in Castellum's annual report for 2025 sits a number that is easy to skim past: 3,000,000,000 authorized shares of common stock — against 94,612,750 actually outstanding (December 31, 2025). The company could thus more than thirtyfold its share count without asking shareholders again. For scale: Castellum has used up barely 3 percent of that authorization so far.

In theory, a large authorization is just a reserve resolution. In practice, Castellum has used it: from the end of 2023 to the end of 2025, shares outstanding rose from 47.7 to 94.6 million — up 98 percent in two years, through equity offerings (December 2024: $3.7 million; March 2025: $4.5 million; June 2025: $5.0 million), warrant exercises and shares used as acquisition currency. The report itself warns that future equity financings may involve "substantial dilution." Whoever reads the stock's pocket-change price as a bargain should know about this pantry.

The last annual report of the Mercurity era (20-F for 2024) tells a growth story that no longer exists a year later: the Hong Kong subsidiary Aifinity was to manufacture liquid cooling panels for AI infrastructure and high-performance computing, and the Shenzhen subsidiary Yingke Precision was to build a factory for it. On September 1, 2024, the holding company even signed a purchase contract for $2 million of Inspur AI servers — $1 million was paid up front, and per the report the servers had not been delivered as of December 31, 2024.

In the first 10-K under the new name Chaince Digital, none of it remains: Yingke Precision was deregistered in 2025, Aifinity is in the process of deregistration ("did not contribute materially to the Company’s operations in 2025"), the Inspur servers are not mentioned with a single word, and AI manufacturing survives only as a statement of intent in the management plan ("pursuing selective growth opportunities in … AI-enabled intelligent manufacturing"). For investors, this is a pattern worth knowing at this company: grand narratives from one annual report can vanish without a trace in the next.

In November 2025, Mercurity Fintech gave itself the new name Chaince Digital — and in December 2025, a few weeks later, the board decided to discontinue the last remaining digital asset business: Filecoin mining, which has been wound down as a discontinued operation ever since. What remains is a scattering: as of March 31, 2026, the company held $843,968 in crypto assets per the 10-Q (Bitcoin $681,819, Solana $83,360, Filecoin $78,789) — and in the first quarter of 2026 alone, those holdings lost $278,661 in market value.

The actual business of the "Digital" holding today is classic capital markets advisory through the broker-dealer Chaince Securities; the cash ($36.7 million) sits in dollars, not in coins. Anyone who hears the name Chaince Digital and pictures a crypto treasury will find the opposite in the balance sheet: an advisory shop currently winding down its digital legacy.

On January 23, 2026, Chaince Digital dismissed its auditor, Singapore-based OneStop Assurance PAC — with a justification you rarely read: per the 8-K, OneStop itself concluded it could not continue auditing the company because its principal executive offices are located in the United States. One day later, Tang Qian & Associates PLLC was appointed, a PCAOB-registered firm from Dallas, Texas.

Eight weeks and two days after the appointment, on March 26, 2026, Tang Qian signed the opinion under the first 10-K in the company's history — including the assessment of the going-concern doubt that was first raised and then declared alleviated ("substantial doubt … alleviated"). All of this is formally permissible. But an auditor working through a company with four names, a discontinued mining business and $686 million of accumulated losses within two months is one of those details the annual report itself does not comment on — alongside the material weaknesses in internal controls it concedes.

Through the end of 2025, Chaince Digital was registered as a foreign private issuer — insiders did not have to report their purchases and sales to the SEC. Since March 18, 2026, U.S. reporting duties (Section 16) apply, and the first Forms 4 in the company's history arrived promptly: Wilfred Daye, chief strategy officer and director, sold a total of 12,000 shares in five tranches between May 19 and June 2, 2026, at prices between $8.53 and $9.66 — having held 81,765 shares per the 10-K.

On June 3, 2026, one day after the last tranche, the company reported his resignation as director and chief strategy officer in an 8-K — "voluntary and not the result of any disagreement," as the standard phrase goes. The amounts are small (a good $110,000). What is remarkable is the pattern: the first insider ever required to report used a price level more than twice as high as the stock traded in mid-July 2026 — and was gone afterwards.

The principal-shareholder section of Chaince Digital's first 10-K contains a name that will make many readers think of the U.S. giant Apollo Global Management: Apollo Multi-Asset Growth Fund. The address behind it, however, is not in New York but on the 16th floor of the Tung Ning Building in Hong Kong — and the position is substantial: 14,251,781 shares plus warrants on another 42,755,344 shares at $1.00 each, exercisable through November 30, 2026. The fund received the package in December 2023 for $6 million — back when the company was still called Mercurity Fintech.

Counting the warrants in full, the fund would hold about 46.7 percent of all shares then outstanding, based on the March 20, 2026 figures (57.0 of 122.2 million) — more than all of today's directors and officers combined (who, per the ownership table, come to less than 0.2 percent). The company's biggest power factor thus sits only partly in the share register — the larger part lies in a drawer as subscription rights, and they expire at the end of November 2026.

Rackspace resided in San Antonio in a converted shopping mall — internally called "the Castle". In March 2024 the company sold its own headquarters; in the annual report the transaction surfaces as a $9.0 million "MEIA early termination fee", a charge for ending a site agreement ahead of schedule. Of the roughly 1,800 North American employees, only about 500 still work at the headquarters, per the annual report; group-wide, roughly 70 percent of all "Rackers" are classified as work-from-home.

In business terms this is consistent — whoever operates data centers does not need a castle. But as a footnote to the company's history it is remarkable: the building that once stood as a symbol of Texan tech culture was monetized in the same month the company restructured its debt. If you want to know how a company is really doing, sometimes the best place to look is not the slide deck but the real-estate footnotes.

On September 3, 2025, a new chief executive took the wheel — and the compensation footnote documents a remarkably timed package: 4.0 million restricted stock units and 6.0 million stock options, granted outside the regular 2020 incentive plan as "Employment Inducement Awards", registered on September 4, 2025. The grant date fair value: $1.30 per RSU and $0.93 per option — the stock was flirting with penny-stock territory at the time.

A few months later the shares traded around $6.40 (data as of July 8, 2026) — on paper, the value of the RSU tranche has roughly quintupled before the new chief has been accountable for a single full fiscal year. The awards vest over four years, nothing about them breaks any rule, and inducement awards outside the plan are common in the U.S. But the order of magnitude — 10 million share instruments against roughly 245 million shares outstanding — shows how cheap dilution is when the price sits on the floor, and how strongly management profits from the rally scenario.

The debt footnote of the annual report contains a sentence that says more about the situation than any price rally: the fair value of the 5.375% Senior Notes (due 2028) was $26.4 million as of December 31, 2025 — against $125.4 million of outstanding face value. The market thus valued the unsecured notes at about 21 cents on the dollar.

These notes sit at the very back of the creditor queue: in the March 2024 debt restructuring, the majority of creditors exchanged into new, secured instruments of a new intermediate entity ("Rackspace Finance"); whoever did not exchange was left holding unsecured paper. A price of 21 percent on a bond that would have to be repaid at 100 in 2028 is the kind of price bond professionals set when they consider full repayment unlikely. The same company's stock multiplied over the same period — one of the starkest disagreements between the equity and the credit market you can currently find.

That Apollo Global Management controls Rackspace is on every fact sheet: about 53 percent of the voting power as of December 31, 2025, plus "controlled company" status, which exempts Rackspace from several Nasdaq governance requirements — such as the rule that a majority of the board must be independent.

Less known is the second role: the debt footnote discloses that Apollo affiliates are simultaneously lenders in the company's most important loan. As of December 31, 2025, $79.7 million — 4.9 percent — of the FLSO Term Loan Facility was owed to Apollo vehicles. The majority owner thus earns not only on the equity but also on the interest of the very debt its own buyout left behind. All disclosed, none of it forbidden — but whoever buys the stock should know that the most important shareholder still holds claims even if the equity ends up worthless.

The stock-compensation footnotes contain a contract you would not expect at a marketplace company: on March 11, 2025, Groupon signed a three-year agreement with the marketing firm Major Rocket LLC. Major Rocket sources enterprise merchant offerings for the platform in North America — and earns incentive compensation if those offerings reach financial benchmarks ranging from $10.0 million to $25.0 million.

Payment is not necessarily in cash: the incentives may be satisfied through the issuance of up to 954,000 Groupon shares (grant-date fair value: roughly $9.3 million), or cash equal to the then-current value of those shares at the company's election. As of December 31, 2025, achievement of the benchmarks was deemed "not probable" per the annual report. An external sales partner paid in the shareholders' own currency — disclosed, legal, and a remarkable window into what Groupon is willing to spend to buy growth.

For more than a decade Groupon dragged two Italian tax assessments behind it: the authorities demanded a combined roughly $170 million for transfer-pricing matters from 2011 and transactions from 2017 — $134.9 million (the "Italy 2012 Assessment") plus $35.1 million (the "Italy 2017 Assessment"), each including penalties and interest. For scale: that is more than a third of today's annual revenue.

On December 29, 2025 came the closing line: a binding framework agreement with the Italian tax authorities settles all outstanding disputes for a total of approximately $25.3 million (21.5 million euros) — of which $10.4 million had already been paid through installment plans. Measured against the claim, a settlement at roughly 15 cents on the dollar. A piquant detail on the side: Groupon had long since given up its Italian operating business — the exit was approved in July 2024.

On paper Groupon is a Chicago company — in practice, a Prague investment group has set the tone since 2023: per the 2026 proxy statement, Pale Fire Capital is the largest stockholder with 10,181,070 shares (26.24 percent). CEO Dusan Senkypl is simultaneously chairman of Pale Fire's board, COO Jiri Ponrt was a partner and group CFO there — and Pale Fire had already backstopped the company's $80 million rights offering in 2023.

How far the shift goes shows in the human-capital footnote: of 1,734 employees at the end of 2025, only 421 work in North America — 1,313 sit in the international business, even though North America delivers 77 percent of revenue. The chief technology officer leads his team of roughly 350 people from Munich, per the annual report. All disclosed, none of it forbidden — but whoever buys GRPN should know: the majority of the workforce and the defining minds of this "U.S. turnaround" work in Europe.

Deep in the investments footnote of the annual report sits an asset hardly anyone associates with the deal merchant: Groupon holds 1.79 percent of the privately-held payments company SumUp (SumUp Holdings S.a.r.l.), one of the larger European fintechs. The balance-sheet line "Investments" stood at $74.8 million as of December 31, 2025 — and per the footnote, SumUp is the only equity investment with a positive carrying value. Against a market value of roughly $0.9 billion (data as of July 8, 2026), this single position equals almost one-twelfth of the entire company.

The heirloom dates back to 2012, when Groupon was still acquiring in every direction. In 2023, in the liquidity squeeze before the new leadership took over, the company already sold about 21.1 percent of its SumUp stake for $19.0 million in cash and wrote the remainder down by $25.8 million. Whoever buys Groupon stock buys a small fintech portfolio with it — a hidden reserve or an emergency till, depending on how you read it.

For years, O-I Glass told the story of a revolution in glassmaking: "MAGMA" — modular, smaller melting furnaces that could be switched on and off with the seasons and were meant to slash the capital intensity of the business. The annual report for 2025 records the ending soberly: the program "had not met the operational and financial thresholds required," development was halted in the second quarter of 2025 — including roughly $104 million of restructuring, asset impairment and other charges.

What stands out is less the farewell than the consequence: spending on research and development is expected to "significantly decline," the company writes. A manufacturer whose core problem is falling demand for glass containers is cutting precisely the technology bet that was meant to make its business model more flexible — and is going all in on the "Fit to Win" cost program. That may be the right business call. But it also means: there is no plan B anymore.

Tucked between restructuring tables, the annual report hides an episode straight out of a history book: from late 1956 through June 1967, a subsidiary operated a paper mill on the shore of the Cuyahoga River in Ohio — on a site that is now part of Cuyahoga Valley National Park, managed by the National Park Service. The U.S. government sued O-I for $50 million to remedy contaminated soils at the site plus its past and future costs.

In 2024 the company recorded charges of $11 million as its best estimate, and in the first quarter of 2025, after a tentative settlement with the Park Service, another charge of roughly $4 million followed. Measured against the balance sheet these are rounding errors — the lesson is what matters: an industrial company with 120 years of history still drags obligations from factories that have stood still for almost 60 years. That is exactly why reading the footnotes is mandatory with companies like this.

The outlook section of the annual report for 2025 contains a paragraph hardly anyone expects: O-I Glass is monitoring "recent developments in Venezuela" and recognizes "that the situation remains fluid." The backstory reaches 16 years into the past: in 2010, Venezuela expropriated the majority interest of subsidiary OI European Group B.V. in two glass plants — without fair compensation. The international arbitration body ICSID subsequently issued an award in O-I’s favor.

The actual side find: O-I long ago sold the rights to that award — to an investment fund domiciled in Ireland, as disclosed in the annual report for 2020. Should money ever flow from Caracas, the terms of the sale limit any additional payments to O-I, and the company itself writes there is no assurance it will receive anything. An expropriated multinational, a sold claim on billions and a fund betting on Venezuela’s ability to pay — you do not read that in every balance sheet.

Hidden in the fourth-quarter 2025 results release (6-K of April 10, 2026) is a footnote that makes you sit up: Lotus Tech admits a "mathematical error" in the previously published third-quarter 2025 figures. The reported total other comprehensive loss for Q3 2025 was simply added up wrong; corrected, it comes to $25.2 million, with total comprehensive loss for the quarter at $90.6 million. The company classifies the error as immaterial.

Taken alone, a footnote-sized slip — in context, not quite: the annual report lists five "material weaknesses" in internal controls over financial reporting at the UK and EU subsidiaries for 2023 and 2024 (from revenue recognition to inventory counts to unauthorized journal entries), reported as remediated since, and as of December 31, 2025 — in the middle of the going-concern environment — the company switched auditors from KPMG to Grant Thornton. An addition error in the quarterly release is then less a stumble than a reminder of how young and strained this reporting apparatus still is.

On June 12, 2026, Lotus Tech announced in a mandatory filing that it would temporarily suspend the release of financial results for the first and third quarters of 2026 — the stated reason: prioritizing the "acquisition-related compliance work" around the takeover of its sports car sister Lotus UK, expected to close in 2026. Investors in a company with an audited going-concern qualification thus go two of four quarters without learning how cash and losses are developing.

What makes this possible is the foreign private issuer status: for foreign issuers on U.S. exchanges, quarterly reports are voluntary attachments to 6-K filings, not an obligation like the 10-Q for U.S. companies. Formally, the suspension is allowed. It just happens to hit a company whose auditors have documented "substantial doubt" about its ability to continue — precisely the constellation in which an investor needs more interim updates, not fewer.

On November 7, 2024, Lotus Tech announced a subscription agreement with Kershaw Health Limited for a senior bond due 2029 with a principal amount of $500 million — at 100 percent of face value. For a company that recorded $333.9 million of operating cash outflow in 2025, that would be the single largest financing building block of all.

Except: the annual report 20-F for 2025, filed on April 28, 2026 — almost a year and a half after the signature — states laconically: "As of the date of this annual report, the transaction has not been closed yet." Half a billion dollars announced 18 months ago and never funded says more about the financing situation than many a ratio — and explains why Geely convertible notes and RMB credit lines fill the till instead.

The liquidity chapter of Lotus Tech's annual report for 2025 contains a sentence you have to read twice: of the related-party loans outstanding as of December 31, 2025, $374.5 million was secured by the company's intellectual property — "with carrying amount of nil", a book value of zero, because the research and development costs were historically expensed. The lender (Geely affiliates) thus holds a pledge that is officially worth nothing on the borrower's balance sheet — while its real-world value to a car group that wants to keep developing the Lotus platforms is obviously substantial.

A second clause fits the pattern: $231.3 million of the related-party loans is structured as "stock-settled debt" — in an event of default, a Geely affiliate may subscribe for Lotus Tech shares at market price in the amount of the outstanding debt. Translated: if Lotus Tech cannot pay, the creditor becomes a shareholder, and the dilution lands on everyone else. Whoever holds the stock should know that the crown jewels — the platform technology and, if needed, fresh shares — are already posted as collateral inside the system.

On June 1, 2026, Agora announced a "Management Share Purchase Plan" in a mandatory filing: founder, chairman and CEO Tony Zhao intends to put up to $20 million of his personal funds into Agora ADSs or Class A ordinary shares within twelve months — in the open market, in block trades or in privately negotiated transactions, within the bounds of insider trading rules. That comes on top of the company's buyback program, of which $156.2 million of the authorized $200 million had already been used by March 31, 2026.

The constellation is what makes it remarkable: Zhao already holds all Class B shares carrying 20 votes each, and with them 86.2 percent of the voting power on 27.0 percent of the capital (March 31, 2026) — control is not what he lacks. A personal purchase of low-vote Class A paper is therefore above all a signal to the market: the insider with the best view considers the price too low. The plan is not binding, though — it is a statement of intent, not a contract.

With China-linked stocks, one standard worry goes: the money sits behind the People's Republic's capital-control wall and never gets out. The credit-risk concentration footnote in Agora's annual report for 2025 paints a different picture: of the $379.7 million in cash, bank deposits and bank-issued financial products (December 31, 2025), 25 percent sat with financial institutions in the United Kingdom, 24 percent in the United States, 22 percent in Hong Kong, 19 percent in Singapore — and only 10 percent in mainland China.

For weighing the "more cash than market cap" argument, that is an important side note: the bulk of the liquidity is not locked up in RMB on mainland accounts but internationally distributed. The usual caveats — dividend restrictions on Chinese subsidiaries, statutory reserves, capital controls — remain in the risk chapter; per this breakdown, however, they touch the smallest part of the money pile.

Deep in the risk chapter of the annual report for 2025 sits a sentence hardly any buyer of the Agora stock has on the radar: "We believe we likely were a passive foreign investment company, or PFIC, for 2025" — Agora considers itself likely a passive foreign investment company for 2025, with a significant risk of remaining one. The reason is, of all things, the balance sheet's greatest strength: a foreign corporation counts as a PFIC when at least 50 percent of its assets produce passive income — and cash is a passive asset. With $366.1 million of liquidity (March 31, 2026) against a market value of about $357 million (July 15, 2026), the money pile dominates the balance sheet.

The consequence hits U.S. investors first: per the report, they face "adverse U.S. federal income tax consequences" — gains can be taxed under a punitive regime instead of the friendlier capital gains rate, plus reporting and election duties. The finding remains remarkable either way: the very cash pile that makes the stock look cheap has written a tax warning label into the company's own annual report.

In the "Property, Plants and Equipment" chapter of Agora's annual report for 2025 sits a construction project you would not expect from an API vendor with $141 million in revenue: in June 2022, Agora agreed with the local government to acquire the land use rights for roughly 42,000 square meters in the riverside area of Shanghai's Yangpu District — total consideration per the report: "approximately RMB2.5 billion", in the order of magnitude of Agora's entire market value of July 2026. The project is held through a joint venture with two independent third parties in which Agora owns 46.39 percent of the equity but, per the report, 100 percent of the economic interest.

On the balance sheet as of December 31, 2025, it shows up like this: a $161.6 million land use right, $84.2 million of construction in progress — and $80.4 million of long-term construction borrowings, the group's only sizable financial debt. Completion is estimated for 2026; Agora plans to use part of the building itself as the future headquarters of its China business (Shengwang). For investors, the takeaway is simple: a noticeable slice of the famous cash pile has already been converted into Shanghai concrete — and a software company has become a construction developer on the side.

The Buckreef license SML04/92 (16.04 square kilometers, renewed through June 2032) contains more gold than the official resource estimate shows: the two deposits Bingwa and Tembo are expressly not included in the current mineral resources, because the surface rights there were never secured. The reason, per the annual report: TRX has not paid land compensation on these areas — due to the artisanal-scale miners active there, whose activities the state co-manages, including by charging royalties.

For investors that is a double footnote: on the one hand, additional, historically drilled gold potential slumbers on the company's own license area and appears in no valuation. On the other, it shows the reality of mining in the Geita region — where a listed company has to plan around the very ore veins on which informal miners work, and where the annual report openly lists artisanal mining as a risk ranging from environmental damage to theft to project delays.

Whoever takes TRX Gold for a young success story should look at the company register: in 2006 the company traded as Tanzanian Royalty Exploration Corporation, in April 2019 it became Tanzanian Gold Corporation, and in May 2022 finally TRX Gold Corporation. Three names, the same company, the same mine — and a balance-sheet line that preserves the past: as of February 28, 2026, an accumulated deficit of $145.8 million sits on the books — more than half of the entire market value as of July 15, 2026 (about $264 million).

The number tells you what today's record quarters were paid with: nearly two decades of exploration and development costs, financed through ever new shares. Only since fiscal 2023 has the company earned money operationally. For context that means: the production records of 2026 are real — but they are the first chapter in which shareholders are not merely paying in, and part of the first-half 2026 profit was promptly consumed again by the warrant revaluation.

Since the fourth quarter of fiscal 2025, Buckreef Gold has had a "Gold Sale Service Agreement" with the Bank of Tanzania: as required by law for all mining companies in the country (Section 59 of the Tanzanian Mining Act), at least 20 percent of local gold production must be set aside for sale to the central bank and local refineries. So the state does not merely co-decide as a 45 percent shareholder — it is also a mandatory customer of its own joint venture.

The curious part: the obligation comes with a built-in incentive. For domestic sales through the central bank, the royalty drops from 7.35 percent to 4.35 percent — three percentage points of discount that are worth real money at record gold prices. In Q3 2025, Buckreef set aside 646 ounces for this and sold them to the Bank of Tanzania at market prices in Q4 2025 — with a positive effect on revenue, cash flow and working capital, per the annual report. A state that turns its own royalty dial to keep the gold in the country: for investors that is opportunity (lower royalties) and warning (the same state can change the rules at any time) in one.

Hidden in the interim report for the second quarter of 2026 is one of the more curious capital measures you will find in SEC documents: TRX Gold maintains an at-the-market program (ATM) under which the company may, at its discretion, sell shares directly into the market for up to $25 million. In the first half of fiscal 2026 it used the program for exactly 19,408 shares with net proceeds of approximately $15,255 — not millions, dollars. The report's verbatim justification: "to gain market intelligence around trading activity."

A sale worth roughly a used car, drawn from a $25 million program, is not a financing event — it is a trial balloon, and at the same time a quiet reminder that the dilution tool sits ready on the table at all times. Whoever holds the stock should know: the pipeline through which new shares can flow into the market is laid, and it demonstrably works.