trading / confluences / b_closeconf
Corpus lane — EN-30 · Killed before scoring, in the open, with a written reason.
Does the last closed 5-minute bar close beyond the level in the signal direction? Intended as a confirmation filter.
EN-30 close confirmation. 1 iff the last CLOSED 5m bar at the decision closes strictly beyond the candidate's `level` in the signal direction. No closed 5m bar, missing level, or a non-finite close -> 0. Correlation with `b_dwell` is expected and is reported by the screen's novelty diagnostic rather than hidden.
Every parameter below was fixed in the design document before any candidate value was computed. No sweep, no tuning: a bit that needs tuning to pass has failed.
| Field | Value |
|---|---|
| Source component | EN-30-close-confirmation |
| Evidence tier | CLAIMED |
| Backlog rank | #23 · 3 source threads |
| frame | "m5" |
| reference | "candidate level" |
| strict | true |
| Note | DROPPED pre-scoring as a structural tautology — see SCREEN_DROPPED['b_closeconf'] and design §A.4 |
| Design doc | backtests/conf_lab/SCREEN-2026-08-07-DESIGN.md |
| Doc sha256 | e2ba7ca77acce01c03ebcdb7afb51ffade7bb7a591303a2923329dba13a91d94 |
DROPPED PRE-SCORING (design §A.4) as a structural tautology: a P1 candidate's decision instant IS the close of the 5m bar that broke `level`, so 'last closed 5m bar closes beyond level' restates the emission condition — 1 on 400/400 sampled candidates. Same reduction class as b_roomstop, found one stage later. Only PREVALENCE was observed in reaching this decision (no outcome value was computed), so the drop is data-blind w.r.t. everything the screen measures. The function stays defined above as the record of the finding; it is not in SCREEN_BITS and consumes no multiplicity slot.
DROPPED PRE-SCORING as a structural tautology (design §A.4). A P1 candidate's decision instant IS the close of the 5m bar that broke `level`, so the bit restates the emission condition: it was 1 on 400 of 400 sampled candidates. A zero-variance column cannot be scored by the Gate-2 instrument at all. Same reduction class as b_roomstop, found one stage later. Crucially, ONLY prevalence was observed in reaching the decision — no outcome, R or association value was computed — so the drop is data-blind with respect to everything the screen measures, and the declared family shrank from 16 to 15. The function is left defined in bits.py as the record of the finding.
Never in this form. Its 1-minute sibling b_dwell was kept precisely because it is NOT implied by emission (sample prevalence 0.60) — and b_dwell was then scored and found negative, which answers the confirmation question anyway.