Auto-generated. This article is rebuilt from FastAPI route introspection by scripts/build_api_kb.py. Edit the route docstring in app/ and re-run the script — do not edit this file directly.

GET /api/v1/screener/trust-board

Handler: app.screener.routes.get_screener_trust_board
Methods: GET

Description

Per-archetype TRUST BOARD — evidence accumulating toward graduation.

One row per screener archetype showing how much realized evidence it holds and which sizing tier that evidence has earned. This is the surface behind every per-pick chip on the action board: a pick can never read stronger than its archetype’s row here.

How an archetype earns a tier. The evidence unit is the ENTRY-DAY COHORT — every name the archetype flagged on the same day is graded together as ONE observation, because same-day picks move together and are not independent bets (in the live ledger one archetype’s 485 graded episodes are 5 entry days). A cohort’s grade is the mean market-adjusted return of its episodes. Those cohort grades feed a skeptical posterior — each archetype starts at a zero-edge prior and shrinks toward the screener-wide average until its own evidence separates it — producing p_edge: the probability the archetype’s true entry-day expectancy is positive.

• NO-BET — below the probation threshold, or drift-degraded, gauntlet-falsified, or holding no evidence of its own. Watch only.
• PROBATION — the posterior cleared the probation threshold. Real money, minimum ticket, sleeve only.
• BETTABLE — the posterior cleared the higher threshold AND enough matured cohorts were earned after the ledger lock. Full sleeve size per the operator’s strategy rules.

BETTABLE must be earned live. Imported historical evidence lifts an archetype to PROBATION at most; the post-lock cohort floor is a hard rule. n_backfill / n_live split the provenance so this is visible, and graduation_forecast.post_lock_deficit names exactly what is still owed.

Gauntlet-verdict veto. An archetype the research program has FALSIFIED on a sealed-holdout forward grade carries a gauntlet_verdict sub-object ({verdict, date, lens, finding, note}; null when unconfigured) and is capped at NO-BET with gauntlet_falsified leading tier_reasons, checked AHEAD of the drift veto — a permanent verdict, not a recoverable window. The action board resolves the SAME predicate, so a chip here never disagrees with the pick chip; every other verdict is display-only.

Honest reads, not encouraging ones. p_edge is null when the shared pool is too thin to estimate a dispersion honestly — the board abstains rather than printing a confident number off a sliver. An archetype with no cohorts is listed with no_evidence: true rather than dropped.

graduation_forecast answers “how much more evidence, at the current average grade, until the next tier” — and it models every gate, so it never promises a graduation that would still be refused:

• cohorts_to_next: 0 — the probability threshold is ALREADY cleared and something else is holding the tier (for a PROBATION archetype, the live post-lock cohort floor named in post_lock_deficit). The note says which.
• cohorts_to_next: null — no count applies. Either the archetype’s current average never reaches the threshold at any cohort count, or its recent window has degraded below the reliability floor, in which case it recovers by performing better rather than by accumulating cohorts. The note distinguishes them.
• A positive integer — the honest estimate, floored by the minimum own-evidence any tier requires.

caveats carries the calm-regime and provenance notes that must be read with any posterior on this board. Tiers inform MANUAL sizing only — nothing here places, sizes, or closes an order.

PUBLIC — archetype-level evidence metadata only (no per-account data, no data-vendor names).

Degrade-honest per DOCTRINE P0: any internal failure returns 200 with an empty roster and a plain reason — never a 500, never a 404.

Curl

curl -s https://bigclawd.com/api/v1/screener/trust-board