STRATEGY LAB
Ten frameworks built in. Yours is number eleven.
Fork any built-in framework into your own copy, describe a screen in plain language, or paste a methodology you wrote elsewhere — then edit its screening criteria and run it through the same EDGAR-grounded pipeline. Your filters are your own editorial choices, applied uniformly across the whole universe of companies.
Fork it. The original stays intact.
Forking copies a built-in framework into your own editable version — you can adjust its screening criteria without changing the original.
Forking is one of three ways to start a screen. The other two begin from your own words or your own document, below. Either way the built-ins are untouched: a starting point, not a ceiling.
Fork a built-in, describe a screen, or paste a document
The Lab's New screen chooser offers three builders, in the app's own words, with the Describe a screen form beside it.
Every path ends in the same place: a screen you review, edit, and test before it runs.
Clone one of the house strategies and change its thresholds, universe, or scoring. The fastest route to something that runs.
Write what you look for in plain language — metrics, thresholds, qualities, sectors — and the Lab composes the filters, the rubric, and the filing list, ready to review.
Describe companies, not your own situation. Age, income, holdings, or risk tolerance cannot be used — RockTurners publishes impersonal research.
A markdown methodology, a "skill" file, or a JSON spec authored elsewhere. The importer maps it onto the catalog and shows what it kept and what it left out.
Write what you look for in plain language — metrics and thresholds, the qualities that matter, sectors to include or exclude, the tradition you follow. The Lab composes the Stage 1 filters, the Stage 2 rubric and the filing list, and shows you a checklist of what will run before anything is saved.
Describe companies, not your own situation. Age, income, holdings, cash, or risk tolerance cannot be used — RockTurners publishes impersonal research and does not give personalized advice.
Every path ends in a draft you read before anything is saved — each mapping decision and everything left out — and a readiness check that says what will run before a scan spends anything: Runnable, Stage 1 only, or Needs attention. A draft keeps composing if the page is reloaded (the sidebar lists it as Composing… or Ready to review), and a saved screen keeps a How this screen was built record. Shown: the composer's own example description; no built-in prompt or criteria set appears on this site.
Iterate like a regression suite
Shown: a derived framework with generic values. Built-in prompt bodies and complete criteria sets are never displayed on this site.
Iterate on this strategy: set up a test basket, pre-screen the universe to find candidates (Stage 1), then deep dive the ones worth a closer look (Stage 2). Results land below.
Filter the full universe down to securities worth a deeper look. Runs against the warm XBRL cache — no SEC fetches.
Passers are ranked and the top 20 advance to the deep dive — adjust with Edit screen.
Dry-run the full pipeline on your test basket or the pre-screen's advancers: reads EDGAR filings on each candidate to turn over the rocks that look valuable.
Dry runs take up to 100 candidates — the basket, the pre-screen's advancers, or both.
Engine choice affects speed and run cost, not the research methodology or what a verdict means.
Tickers you expect the strategy to pass or reject — your regression suite.
No test tickers yet — Add ticker to start the basket.
Each ticker can carry an expected outcome: Pass, Excel, Reject or Neutral.
Your criteria, the whole universe
Stage 1 filter criteria — thresholds, metrics from the filter catalog, how candidates are ranked and how many advance — plus universe scope. Every filter screens companies on your criteria; nothing about the screen refers to you.
A pre-screen also lists the near-misses — One tweak away, with the First breaking filter for each — so a threshold is tuned on evidence rather than guessed. Compare with parent runs your copy and the screen it was cloned from on the same candidates and shows the two results side by side.
The rigor isn't a built-in-only feature
Derived frameworks get the same EDGAR grounding, the same citation enforcement, the same honest unknowns, and the same Gemstone / Geode / Sediment vocabulary as the ten built-in frameworks.
Every run is on record: pre-screens in the History tab, and every deep dive with the same audit record as a scan — the model, the prompt, the checked response and its cost — stored privately and never shown to other members.