Scan declaration: HTF-level proximity as a regime conditioner (mentor Q2)

Date declared: 2026-07-27 (written BEFORE implementation; run same day). Layer: 3 (no execution, no costs, research split only: BTCUSDT 2017-08-17→2022-12-31). Lineage: mentor-concept / our-parameterization (questionnaire Q2: range behavior happens at major HTF S/R ("Negotiation/Absorption"); between levels price moves through a "Vacuum" with velocity; MENTOR-ANSWERS-JULY-2026-v1.0.md §3). Charge: 12 hypotheses at registration.

Question

Does distance to the nearest major higher-timeframe level condition (a) the chop-vs-trend character of subsequent price action and (b) the continuation quality of our two established event families? A conditioner finding feeds gate-state/risk control or future prereg arms — no execution is licensed by this scan.

Definitions (frozen)

• HTF level book: R7 engine (levels.py) restricted to HTF kinds only: pivot_zone from D1 and H4 timeframes, prior_day_high/low, prior_week_high/low, round_major. (M15 pivots and round_minor excluded — Q2 is explicitly about H4/Daily/Weekly structure.) All causality rules unchanged.
• Distance d = min(support_dist, resistance_dist) to the live HTF book, in ATR14(M5) units, computed on completed M5 bars.
• Buckets (ours; Q2 gives no numbers): NEAR d ≤ 1.0 · FAR d ≥ 3.0 · the 1.0–3.0 band is an excluded buffer.
• Sampling for A/B: top-of-hour snapshots (M5-aligned) to limit overlap; block bootstrap by calendar day.
• Efficiency ratio (A): |close_{t+H} − close_t| / Σ|M5 close-to-close moves| over H. Lower = choppier.
• Prior direction (B): sign of the trailing 1h close-to-close move at snapshot.

Hypothesis cells (12)

• A. Negotiation-at-levels (4): efficiency ratio at H ∈ {4h, 24h} × {NEAR, FAR}, each vs its frequency-matched random-snapshot null. Q2 predicts NEAR < FAR.
• B. Vacuum drift (4): signed continuation of prior 1h direction over next {1h, 4h} × {NEAR, FAR}. Q2 predicts FAR > NEAR > 0.
• C. Event-family interaction (4): signed continuation of disp4 events at 1h and up-streak≥8 events at 4h, each split {NEAR, FAR} at event time.

Metrics in ATR14(M5) units (and bp), matching all prior scans. Null battery: Edge Lab standard four; noted in advance that label-shuffle is degenerate for the unsigned A cells (recorded, carried by the other nulls — same treatment as ledger 8/10/12). Per-year sign-stability blocks reported for every cell.

Pre-declared interpretation

• Q2 supported iff NEAR efficiency < FAR with CI-clear difference at ≥1 horizon AND ≥4/6 years sign-stable — then htf_level_dist_atr graduates into gate-state context fields (levels remain context, never signal — R7 posture unchanged).
• Vacuum supported iff FAR signed continuation CI-clear positive at both horizons.
• A new prereg arm is motivated (streak or displacement + NEAR/FAR filter) only if a C-cell shows |FAR − NEAR| ≥ 0.10 ATR with ≥4/6-year stability. Otherwise: the finding, whatever it is, stays a conditioner note.
• Null across the board ⇒ Q2 joins the retired list (level-reaction scan precedent: bounce/break were null in all 16 cells; this scan asks the regime question those cells did not).

What this scan cannot say

Nothing about Validation-era regimes; nothing about harvestability; nothing about his discretionary read of "major" levels (our book is a mechanization — run 18's lesson applies).

RESULT (run 2026-07-27, same night) — MIXED: Q2 weakly right, vacuum wrong, and one real lead

Ledger row 27 (+12 hypotheses → 273/2000); artifacts bot/reports/htf-proximity-scan/results-btcusd.json; features src/tradingbot/edgelab/htf_level_features.py + HTF parquet (rebuild ~67s); commit 1d4e3a8; suite 274 green. disp4/streak event counts reproduced exactly (517/498).

1. A (negotiation-at-levels): Q2 mechanically SUPPORTED at 24h only — NEAR is choppier than FAR with diff-CI clear of zero and 6/6 years stable, but the effect is tiny (~7% relative efficiency) and null at 4h. Per the pre-declared rule, htf_level_dist_atr graduates into gate-state context fields (context, never signal — R7 posture unchanged; GATE-STATE-v2 §3 already reserves the slot).
2. B (vacuum drift): REFUTED. Both NEAR and FAR arms are negative — trailing-1h direction mean-reverts at 1h (CI-clear), it does not "move with velocity through the vacuum." Second Q2 sub-claim dead.
3. C (event interaction): the night's real lead. disp4 displacement events firing FAR from HTF structure continue +0.53 ATR/1h; NEAR structure they reverse (−0.12) — gap 0.654 ATR, gap-sign stable 5/6 years. Clears the pre-declared ≥0.10-ATR + ≥4/6 bar ⇒ a successor prereg arm is formally motivated. Honest caveats attached: n=126/178 per arm, per-arm difference CI still crosses zero, and the displacement family's executed spec died on Validation (run 20 — final for that spec; any successor is a NEW spec on the research split first). Streak version is borderline (0.249 ATR, 4/6, year-flippy) — not pursued.

Mentor scorecard nuance of record: after six mechanization failures, Q2 is his first partially-confirmed structural claim — wrong mechanism (no vacuum), right that HTF-level proximity conditions behavior, strongest exactly where our own event family lives. Successor decision queued (Helm #57), to be declared in daylight, not run tonight.