Decision memo: the displacement-continuation candidate — Richard's two calls

Date: 2026-07-23. Everything below is research-split evidence; Validation is untouched (10 of 10 looks). Chain of record: scan config + results (2026-07-23-displacement-event-scan.md) → execution prereg + result (bot/preregistrations/2026-07-23-disp-follow-btc-v1.md, registry runs 19; HQ Models & runs).

What we have, in one paragraph

After a compression→displacement M5 bar (the mentor's event concept, our timing), price continues in the bar's direction for ~1 hour. Confirmed on BTC and ETH with near-identical magnitudes (5.4 years each, full four-null battery, 4-5/6 years sign-stable including both bears); inverted on gold (crypto-class structure, not universal). Executed faithfully (enter at event close, 1×ATR stop, 60-min exit) on the BTC research split it nets +0.28R/trade at Vantage-class costs (disp5 arm, n=282, 3.15× drag) and stays positive even under the deliberately pessimistic Dukascopy cost bound (+0.06R). The coin-flip control loses money net. Every constant was declared before running; zero parameters were tuned; iteration budgets: 1 run spent of 3 on each registration.

Decision 1 — §5.6 replication interpretation

The frozen promotion bar requires replication "on a second instrument (XAUUSD or GBPUSD)". Gold actively refuted; ETH strongly confirmed. Question: does within-asset-class replication satisfy the requirement's intent (the effect is not a single-instrument fluke) for a candidate that will only ever trade crypto? Amending after seeing results is exactly what §14 forbids — which is why this is presented as an explicit, recorded amendment decision rather than assumed. Options: - 1a. Accept within-class replication for crypto-only candidates (record as amendment to §5.6, dated, with this memo as basis). Promotion path opens. - 1b. Hold the letter — candidate stays research-logged until a D4 quarterly review.

Decision 2 — spend Validation look 1 of 10

If 1a: the promotion step is one registered run of the frozen disp_follow_btc_v1 spec (A2 = disp5/1×ATR/60m primary, A1 supporting) against Validation (2023-01-01 → 2025-06-30) — one look, pre-registered first, both cost variants. Pass/fail criteria would be registered before the run (proposed: net(Vantage) > +0.05R, n ≥ 100, beats coin-flip control, drawdown within 2× control's). A fail is final for this spec (runs 2-3 of its research budget cannot be spent to "fix" a Validation fail — that is the no-re-splitting rule).

What I recommend

1a + 2, in that order, after you read the two caveats below. The candidate is the strongest evidence object the project has produced, the budget spend is exactly what the budget exists for, and waiting adds no information the Validation split wouldn't add better.

The two caveats to read first

• Costs are charged, not price-paid: net numbers deduct 0.031% (Vantage central, sourced) per round-trip rather than filling at real bid/ask. The Vantage demo (live spread) or the mentor's commission columns would harden this before capital — but the Validation look does not require that hardening; Gate B (paper) does.
• 28% win rate: the profile is long losing streaks punctuated by continuation tails. Expect ugly drawdown stretches; the Monte Carlo pass (Helm #32) is now genuinely due if this advances.

Nothing runs until you answer. Reply with e.g. "1a yes, 2 yes" (or ask for anything deeper first).