trading / confluences / c6_room
Live gate · In the live P1 Master gate stack today.
Is there room to run? c6 is 1 when the nearest opposing obstacle sits at least 2× the stop distance away from entry.
Live confluence c6_room, read from the engine-stamped `conf_bits`. Engine meaning: `helpers.htf_and_room` calls `defs.room_to_target(entry, sl_dist, ..., mult=2.0)` — nearest opposing obstacle >= 2 × stop distance. Note this is ALREADY stop-relative, which is what killed the b_roomstop candidate as a duplicate.
Gate 2 of 7, EXPLORATORY. Conditional association stratified on the count of the other five live bits, against this bit's own prevalence-matched null. This report supersedes its own first version, which used a single pooled placebo p95 and published a false headline; the superseded line is left on the ledger unedited (PROTOCOL gate 7).
Family verdict (verbatim, 2026-08-05): EXPLORATORY, NO PROMOTION. Under a correct prevalence-matched null, 1 of the six live confluences clears its own noise floor: c4_sweep +0.1312 R, matched p=0.043 (95.7th percentile of its own null). Over the 6-bit family that is BH q=0.26, its 95% CI touches zero, and all seven years are selection-burned — so it earns a pre-registration (gates 3-4), not a promotion. The other five do not clear their matched nulls. The previous headline on this id ('no real bit clears the placebo p95', 'ALL SIX indistinguishable') was FALSE: it applied one mixed-cloud p95 to every bit, distorting percentiles by up to 22 points in both directions.
Paired replay on the certified conflab-livegate-v1 stream, $250-nominal basis (~2.6% high versus engine dollars), 31 cells examined with ZERO multiplicity correction and no p-values. Every cell below is post-hoc on data the bits were screened on.
| Cell | Net | Δ net | Trades | WR | expR | max EOD DD | Δ pre-2025 | Δ 2025+ |
|---|---|---|---|---|---|---|---|---|
| BASE — live A/B take rule (count >= 4) | $10,399 | — | 526 | 54.0% | +0.0791 | $10,826 | — | — |
| swap_b_actfloor__c6_room | $15,129 | +$4,730 | 1,178 | 53.4% | +0.0514 | $8,643 | +$3,408 | +$1,321 |
| CONFIG_6S_k4 | $11,254 | +$855 | 597 | 53.8% | +0.0754 | $7,923 | −$538 | +$1,393 |
| swap_b_onbreak__c6_room | $11,254 | +$855 | 597 | 53.8% | +0.0754 | $7,923 | −$538 | +$1,393 |
| swap_b_gapadr__c6_room | $7,296 | −$3,104 | 1,078 | 52.7% | +0.0271 | $13,967 | −$3,787 | +$683 |
| CONFIG_6S_k5 | $3,483 | −$6,917 | 121 | 54.5% | +0.1151 | $2,917 | +$633 | −$7,550 |
| CONFIG_6S_k6 n<100 | −$420 | −$10,819 | 8 | 37.5% | -0.2100 | $797 | −$1,179 | −$9,640 |
| CONFIG_6S_k3 | −$7,780 | −$18,179 | 1,505 | 51.2% | -0.0207 | $18,479 | −$13,791 | −$4,388 |
The grid attacked this slot two ways and the tribunal defended it both times. Replacing c6 with b_actfloor posted the grid's biggest headline (+$4,730) — and is a gate LOOSENING, not an edge: c6's 0.092 prevalence swapped for a 0.785 bit raises the mean confluence count by +0.688, so the take rule silently drops from >= 4 to an effective >= 3.31 and the book fills with grade-C rows this same study measures as negative everywhere. Replacing it with b_onbreak (CONFIG_6S_k4) adds only +$855. Standing order: do not touch c6. It is the rarest and therefore most selective live gate, and the 2026-07-21 deep-dive's c4+c6 interaction finding predicts exactly the damage observed when it is removed. A drop-c6 ablation is required before any actfloor-in-c6 claim is ever entertained again — as cell #1 of a NEW grid, never added to this one.
INDISTINGUISHABLE on its own numbers (+0.037 R, matched p=0.694, 30.6th percentile, q=0.694) — and yet it is the gate the 2026-08-08 tribunal ordered left alone. Two reasons. It is by far the RAREST live gate (prevalence 0.096), so it is the most selective thing in the stack; and the 2026-07-21 deep-dive found c4+c6 the best pair. When the config grid replaced it with a 78%-prevalent bit, the resulting +$4,730 was mechanically explained as gate loosening, not edge.
A drop-c6 ablation is a REQUIRED but STAGED experiment: it must be declared as cell #1 of a NEW pre-registered grid, never added to the burned one. If gate deletion alone reproduces ≈ +$4.7k, the real question becomes 'is c6 dead weight?' — a different study, argued against in advance by c4+c6.