Every public company publishes two versions of its year. One is the earnings deck: written by marketing, full of adjusted numbers and upward-sloping charts. The other is the 10-K: written by lawyers, reviewed by auditors, and filed under penalty of securities law. When the two disagree, the 10-K is the one with liability attached — which is exactly why the bad news lives there, phrased as quietly as legally possible.
Most people never open it because it’s 200 pages. But you don’t read a 10-K like a book. You run a route through it, the way the Earnings Teardown runs a route through a quarterly release. The route takes 45 minutes, and it’s looking for twelve specific things.
The 45-minute route
- Minutes 0–5: set up the diff. Open this year’s 10-K and last year’s side by side (both are free on SEC EDGAR — search the ticker, filter to “10-K”). Skip the CEO letter and the business overview entirely. They are the deck wearing a suit.
- Minutes 5–15: risk factors, new entries only. Don’t read all of Item 1A — it’s boilerplate by volume. Read what changed. A risk factor that appears for the first time is a sentence a lawyer fought to include. “We depend on a limited number of customers” showing up in year three of a customer relationship means the concentration got worse, or someone got nervous.
- Minutes 15–30: the three statements, five comparisons. Income statement, balance sheet, cash flow — but only five checks: revenue growth vs. receivables growth; net income vs. operating cash flow; deferred revenue direction; share count direction; inventory vs. sales. Each is a flag below. This is arithmetic, not accounting — a phone calculator is enough.
- Minutes 30–40: footnotes, four keywords. Search the document for “related party,” “revenue recognition,” “commitments and contingencies,” and “subsequent events.” Footnotes are where changes get disclosed at minimum legal volume.
- Minutes 40–45: the people check. Auditor: same firm as last year? Any “material weakness” language? Then the proxy statement (DEF 14A): what metrics is executive pay tied to? Management optimizes what it’s paid on — if bonuses key on adjusted EBITDA, expect the adjustments to grow.
The twelve flags
Revenue quality
- 1. Receivables growing faster than revenue. Sales up 9%, receivables up 30% means the company is booking revenue it hasn’t collected — often by pulling next quarter’s deals forward with loose terms. Days sales outstanding (receivables ÷ revenue × 365) creeping up year over year is the same flag in ratio form.
- 2. Deferred revenue shrinking while sales “grow.” For subscription businesses, deferred revenue is tomorrow’s income already paid for. If it falls while reported revenue rises, the company is eating its seed corn.
- 3. A metric that quietly changed definitions. “Active users” that now counts something new, “organic growth” that now includes a small acquisition. The tell is a footnote to a table nobody reads.
Profit quality
- 4. The adjusted-vs-GAAP gap widening every year. Every company adjusts. The flag is the trend: if the distance between adjusted earnings and GAAP earnings grows annually, the “one-time” items are the business.
- 5. Recurring “non-recurring” charges. A restructuring in year one is a decision. A restructuring in years one, two, and three is a cost center with a flattering name.
- 6. Capitalizing what peers expense. If a company capitalizes software development or customer-acquisition costs its competitors expense, its margins aren’t better — its accounting is friendlier. The depreciation footnote tells you.
Cash reality
- 7. Net income up, operating cash flow down. Profit is an opinion; cash is a fact. One divergent year can be timing. Two is a pattern that needs an explanation you can locate in the filing.
- 8. Buybacks funded by borrowing. Repurchases announced in the same year debt jumped means the EPS growth is financial engineering, and it stops working when rates don’t cooperate.
- 9. Working-capital springs. Stretching payables, factoring receivables, supplier-financing programs — each one borrows cash flow from next year to decorate this one. Look for “supplier finance” in the footnotes; disclosure is now required.
People and disclosure
- 10. CFO or auditor departures. CFOs leave for real reasons all the time. But a CFO exit plus an auditor change inside 18 months is the highest-signal pairing in this list.
- 11. Related-party transactions. The company leases its headquarters from the founder’s LLC, or buys services from a director’s firm. Small ones are common; growing ones are governance telling you who the business actually serves.
- 12. The vanished KPI. The metric management trumpeted for eight quarters — then silently stopped reporting. Companies retire metrics when the metric retires them.
Scoring: flags are questions, not verdicts
- 0–2 flags: normal corporate messiness. Every filing has some.
- 3–4 flags: the story and the statements disagree somewhere. Write down the specific questions and go find answers — earnings-call transcripts, the prior year’s filings, competitor disclosures — before you trust the narrative.
- 5+ flags: you’re in forensic territory. Whatever the stock does, the burden of proof has shifted to the company.
The rubric’s job is to stop you from doing the two dumb things: treating one flag as a fraud verdict, and treating twelve green boxes as a guarantee. A flag is a question the filing raised and didn’t answer.
Worked example: Meridian Fixtures (fictional, deliberately)
Say Meridian, a mid-cap kitchen-hardware maker, reports revenue up 9% — a nice year in a flat category. The route finds: receivables up 31%, with DSO moving from 61 to 74 days (flag 1). Adjusted EBITDA excludes a “transformation program” charge for the third consecutive year, cumulatively equal to about 40% of operating income over the period (flags 4 and 5). And a new footnote discloses a receivables-factoring facility opened in Q4 (flag 9).
No single item is damning. Together they sketch one coherent story: the 9% growth was bought — extended payment terms pulled sales forward, the factoring facility papered over the cash gap, and the recurring “transformation” charge keeps the adjusted numbers presentable while it happens. Maybe there’s an innocent explanation. That’s three specific questions for the next earnings call, and a growth number you now discount until someone answers them.
Where AI helps — and where it lies
The diff-and-summarize work is where a model earns its keep. Feed it both years’ risk-factor sections and ask: “List every risk factor that is new, removed, or materially reworded this year. Quote the changed language exactly.” Same trick for footnotes. This is the tedious 60% of the route, done in two minutes, and — because you demanded quotes — checkable.
What you never do is let a model extract the numbers unverified. Hallucinated figures are the signature failure: a confident table where one receivables number is from the wrong year, or invented outright. The rule from the verification trust ladder applies in full: AI drafts the map, the filing is the territory, and every number you’d act on gets checked against the PDF by your own eyes.
Our take: The market reads the press release in the first minute and the filing over the next six months — that lag is where most “suddenly” blowups come from. Nothing in this playbook requires a finance degree; it requires 45 minutes and the willingness to believe arithmetic over narrative. And to be clear: this is analysis hygiene for understanding businesses, not investment advice — a filing tells you what’s true, not what to buy.
Make it a habit
- Run the route once per year per company you own or track — filing season clusters in February–March for calendar-year companies.
- Keep a one-page log per company: flags found, questions raised, answers received. Next year’s read takes 20 minutes because you diff against your own notes.
- Pair it with the 15-minute weekly review for the market layer and the Earnings Teardown for the quarterly layer. Annual, quarterly, weekly — that’s the whole stack.