Pre-registration: disp_follow_btc_v1 — displacement-follow execution test on BTCUSD

Date registered: 2026-07-23, before any execution-level test of this candidate. Lineage: hybrid — the event definition descends from the mentor's expansion concept (journals → expansion.py::detect_events); the entry timing (follow the displacement close immediately, no pullback wait) comes from our scan evidence and contradicts the taught first-pullback entry; stated per registry lineage rules. Evidence basis: engine/scans/2026-07-23-displacement-event-scan.md — continuation +0.38…+0.72 ATR at 15m–1h on BTC (5/6 years, full null battery) and ETHUSDT (near-identical magnitudes, 4-5/6 years); refuted on gold (crypto-class effect).

Scope: Research split ONLY (Binance BTCUSDT M1, 2017-08-17 → 2022-12-31). Validation (10/10 looks) and Vault untouched. This run answers §5.1 — does the effect survive execution geometry and drag — and, with the ETH scan, completes the evidence pack for Richard's §5.6/Validation decision. It is not itself validation evidence. Fresh strategy segment: run 1 of 3.

Frozen spec (zero free parameters; single declared value everywhere, no sweeps)

Event: detect_events-equivalent in scale-free bp form, exactly as the scan: M5 bar with range ≥ disp_mult × 24h-median baseline, body ≥ 60% of range, prior-hour compression, baseline ≥ 7 bp. Signal at event-bar close; enter at the open of the next M1 bar in the event direction (longs AND shorts — the scan effect is direction-signed both ways).

• Stop: 1.0 × ATR14(M5) at the event, from entry (the scan's own volatility unit; 1R ≡ 1 ATR so scan magnitudes translate directly to R).
• Exit: time exit at the open of the first M1 bar ≥ 60 minutes after entry (the measured harvest window), unless the stop fills first. No TP, no trail — the faithful harvest of exactly what the scan measured (close→close 1h continuation).
• One position at a time; an event arriving while in-position is skipped.
• Arms (all pre-declared): A1 disp4 (primary) · A2 disp5 · A3 disp4 with 30-min exit (the other measured peak) · C1 control: coin-flip direction on A1's exact events and geometry (seed 20260723).

Costs

Charged as round-trip % of entry, both variants reported; Vantage central 0.031% (BTCUSD_VANTAGE, VANTAGE-COST-UPDATE-v1.md) is the decision case; Dukascopy bound 0.0832% reported alongside. Drag in R varies per trade (stop = ATR): predicted mean drag ≈ 0.031% / mean(ATR%) — reported against measured (H4).

Registered hypotheses

• H1 — the scan effect survives execution: A1 gross expectancy > 0 AND > C1 gross.
• H2 — it survives realistic costs: A1 net (Vantage) > 0.
• H3 — dose-response persists: A2 gross > A1 gross (per-trade).
• H4 — drag law: measured mean drag within ±30% of cost/ATR prediction (by construction approximate here since costs are charged, not price-paid — reported, not counted as an independent law validation).

Pre-declared "worth a Validation conversation" bar (NOT Gate A; research-split only)

All four: A1 net (Vantage) ≥ +0.05R · n ≥ 200 · beats C1 net · gross positive in ≥ 4 of 6 year blocks. Meeting this bar does NOT trigger a Validation run — it triggers the decision memo to Richard (the 1-of-10 look is his call, as is the §5.6 within-class question).

Honest prior

The scan says +0.38 ATR mean forward move at disp4; but scan moves are close-to-close with no stop truncation — a 1-ATR stop will cut some winners' paths and convert some to losses, and costs take ~0.03-0.08R. Expected: H1 holds (the effect is large and the entry is immediate), H2 genuinely uncertain — this is the first candidate in the project where the net question is live rather than doomed by arithmetic. n expectation: several hundred (disp4 events 517 minus in-position skips).

Iteration budget

Run 1 of 3 on this segment. Any change after seeing results = new registration. Hypotheses logged: 4 (cumulative 44 of 2,000).

Result (filled 2026-07-23, after the single registered run)

ALL FOUR HYPOTHESES HELD; the pre-declared bar is MET — first candidate in the project to show positive net expectancy under realistic costs. Registry run 19 (disp_follow_btc_v1, lineage hybrid); artifacts reports/disp-follow-btc-v1/.

• H1 ✓ A1 gross +0.235R vs control +0.052R — the direction information survives execution (the +0.18R gap is signal, not geometry; the control's small positive gross reproduces the known event-vol artifact and costs eat it).
• H2 ✓ A1 net(Vantage) +0.101R. H3 ✓ dose-response persists (A2 ≫ A1 per trade).
• H4 drag measured 0.134R (predicted ≈ cost/mean-ATR ≈ 0.14R) — consistent; flagged charged-not-price-paid, so not an independent law validation.
• Year blocks (A1 gross): 2017 −1.11 (thin, 5-month partial year) · 2018 +0.45 · 2019 +0.40 · 2020 −0.03 · 2021 +0.74 · 2022 +0.06 → 4/6 positive, both bears positive.
• §5.1 note (important): the 3×-drag requirement is met by A2 only (0.410/0.130 = 3.15×; A1 is 1.75×). The §5-conformant candidate is therefore disp5 / 60m, which is also positive even under the pessimistic Dukascopy bound (+0.061R). A1 stands as the higher-n supporting arm.
• Low win rate (~28%) with time-exit tails: expect long losing streaks; drawdown profile and Monte Carlo (Helm #32) become relevant exactly now.

Per pre-commitment, nothing further runs without Richard: the Validation look (1 of 10) and the §5.6 within-class replication question are his; see engine/scans/DECISION-MEMO-2026-07-23-displacement.md.