HOW IT WORKS
Two stages, from the whole universe to a cited verdict
Over 5,000 U.S.-listed companies file annual reports, and nobody can read them all. Here is how RockTurners narrows the field, reads the filings of the companies that rank highest on the screen, and shows its work so you can check it.
Every framework runs the same way: a fixed quantitative screen narrows the field, then an AI evaluator reads the filings of the top-ranked companies that pass. Nothing is personalized: who you are is never an input.
- 1 · SCREENThe framework's fixed filters run on over 5,000 U.S.-listed companies.
- 2 · READIts AI reads the filing sections the framework declares, for the top-ranked companies that pass.
- 3 · VERDICTEach company read gets Gemstone, Geode or Sediment, or Not scored when the filings can't support a verdict.
- 4 · YOU CHECKOpen the filings on SEC.gov, find the cited sections, and see what the filings don't say.
Filter the universe
Each framework declares a fixed set of quantitative filters — return on capital, balance-sheet strength, payout coverage, whatever its published tradition requires. They run against the SEC EDGAR operating universe, using figures from the filings and, for valuation and momentum filters, market prices, identically every time. Every metric is defined in the glossary.
Near-misses aren't discarded silently: the screen keeps the rationale for every company that dropped out, one filter from passing.
Stage 1 is the part RockTurners shares with any AI stock screener; Stage 2 is the part most screeners don't have.
Read the filings
The top-ranked companies that pass go to the framework's evaluation prompt — frozen, versioned, identical for everyone. The evaluator cross-references the pre-computed metrics against the filing text: the metrics say what, the filings say why and whether it's durable.
The output is the structured verdict you saw dissected on Anatomy of a verdict: scored dimensions, trap signals with sources, unknowns, and a tier.
Your derived frameworks from the Strategy Lab run through the same two stages.
One of three verdicts on framework fit
Framework fit means how well the company's filings satisfied the screening and evaluation criteria of the framework you selected — the strength of that match. It is not a recommendation to buy, sell, or hold anything, and not a prediction of returns.
General Analysis
Ten frameworks each express a published tradition, organized under the lenses. General Analysis is not one of them — it is the general-purpose strategy, for portfolio holdings and one-off company reads when you want the evidence organized before choosing a discipline.
A framework screens a universe of companies. General Analysis evaluates one named company, so there is nothing to screen out — it skips Stage 1 and goes straight to the filings read. Same pipeline, same filing sections, same citation rules, same honesty about gaps. Instead of a named framework's rubric, it applies a general, methodology-neutral one: business quality, financial strength, earnings quality and risk profile. It belongs to no lens, because it is not a framework.
- ASSESSMENT Prose grounded in the filings it was given — figures are drawn from the 10-K sections provided, and the countervailing items get named alongside the strengths.
- FRAMEWORK SCORES Four dimensions — risk profile, business quality, earnings quality, financial strength — each scored 0–5 on how well the provided filing sections support it. The citations behind a score are checked against the filing sections provided; where the filings are silent, the dimension is marked unknown instead of scored on inference. A framework score describes how well the analysis is supported by the cited filings — it is not a rating of the company or a level of conviction in any action.
- CONFIDENCE A roll-up of the same idea: how well-supported the overall analysis is by the cited filings. Missing sections and unverifiable claims lower it. It is not a rating of the company or a level of conviction in any action.
- INSUFFICIENT DATA Sections the evaluator wasn't given become named unknowns, not silent guesses — the honesty-about-gaps rule made visible.
- AI TRANSPARENCY Every card names the model that wrote it and carries the AI and NOT ADVICE marks.
Sample, Inc presents as a mid-sized established operator with a total framework score of 16 — clearing many, but not all, of the conditions for the top tier of this general analysis. Per the 10-K filed February 12, 2026, revenue grew from $3.8 billion in 2023 to $5.1 billion in 2025, and operating cash flow of $810 million exceeds net income of $505 million — evidence of sustained profitability across the full three-year filing record. The most material countervailing item is capital intensity: $760 million of property and equipment purchases in 2025, roughly double the 2023 level. No restatement, going-concern, or auditor-change signal was found in the provided sections.
Recreation of the app's results card. "Sample, Inc" is a fictional company — illustrative data only.
See a whole verdict, taken apart
The workhorse page: one sample output, annotated part by part.