conviction-funnellisted

# conviction-funnel Turn a market full of noise into a *small* set of researched, actionable names. The premise: any single scan can fire on a fluke, and a momentum name tends to be extended by the time it ranks, so picking off one screener tends to buy tops. The funnel runs the market gate, pulls the name lists, cross-references them, then spends real research effort on only a handful, ending in a side-by-side table of where to enter, where the stop goes, how big to size, and what kills the thesis. ⚠️ **How the 2026-05→07 backtests changed the middle step.** The sister scans carry outcome backtests, and the old "2–3 scans agree = conviction" premise did not survive them: overlap count ranked conviction **backwards** (3-scan names −4.5% excess at T+10 vs −1.5% for single-scan), with unusual-options co-flags marking *froth*, not smart money. The two skills built on that premise, cross-scan and unusual-options-scan, are **retired as of 2026-07** on those numbers. The funnel keeps the same shape but the middle step is now a pure file read: momentum's list plus each surviving scan's *validated pockets* form the candidate pool, and the per-scan validated filters (BaseWks ≥ 20; MR Score ≥ 40 on a fresh listing; volume-backed momentum entries) do the ranking that overlap counting used to do. It orchestrates skills the user already has rather than re-implementing anything: ``` regime-scan ── market gate: 🟢/🟡/🔴 + divergence flags (are we even adding risk?) momentum-scan ── p