#81 — Composite 1 — Axis-1 D1 displacement fade × Axis-3 COT positioning vote
2026-07-07
Verdict: negative · ⛔ conjunction too rare, vote indistinguishable from a random gate — seal never opened
The book's own Appendix-E recommendation, executed to its end: gating a D1 swing-extreme fade with an independent COT crowding vote produced only 54 of the pre-registered 150-trade IS minimum, and its net mean (+23.3p/trade) sat inside a shuffled-positioning control's own null distribution (mean +21.3p, p95 +58.7p) rather than clear of it — the positioning vote filtered like a random gate, not an informative one.
Hypothesis — what it tests
Whether stacking Axis-1 price-displacement (D1 swing-extreme fade) with Axis-3 COT positioning (an independent, orthogonal directional vote) clears the retail spread with a working margin where neither axis clears it alone — and whether the COT vote's apparent gating power is genuine information or just trade-count-reducing selectivity, tested via a purpose-built shuffled-positioning null.
Description
Step 2 of the book's Appendix-E roadmap, built directly on step 1 (cot_positioning, Entry 77, which isolated Axis 3 alone and passed 4/4 IS gates but with a CI spanning zero). Composite 1 stacks two axes the book's own axis analysis (Appendix E) argues are genuinely orthogonal — Axis 1 (price-displacement: a fresh D1 rolling swing-extreme first touch, low tick-volume, structure-transferred from the deployed first-touch config, EPS=0.25xATR14, declared not fitted) and Axis 3 (macro/positioning: the COT crowding z-score vote, code reused verbatim from cot_positioning for parity) — on 7 direct-USD pairs, D1 bid/ask, 2002-2026-05-21 ceiling, inheriting cot_positioning's IS/OOS boundary so the held seal spends exactly once. The pre-registration's decisive fifth arm was a shuffled-positioning control: 200 block-preserving-by-currency permutations of the COT vote gating the identical Axis-1 trades, which the composite had to beat at the 95th percentile to prove the vote adds information rather than just thinning the trade count. Of 6 pre-OOS gates, 4 passed (RW self-test, Axis-3 parity vs cot_positioning exact to 0.0000p/wk, walk-forward 2/3 IS thirds positive, breadth 4/7 pairs gross-positive) and 2 failed: gate 3 (all three H1 money/vs-coin/vs-shuffled-null criteria) and gate 6 (trade count, 54 of the required 150, 36% power). The composite's net mean (+23.26p/trade, n=54) not only failed to clear the shuffled null's 95th percentile (+58.68p) — it barely cleared the null's own mean (+21.25p) and actually underperformed its own coin-flip control on the identical 54 timestamps (+41.90p). Per the pre-registration's decision rule, an IS gate failure closes Composite 1 with the seal intact and unspent, available for one future amended shot; the shuffled-positioning control failing while raw net theoretically passes is recorded as "selectivity in disguise" — the book's own Axis-5 (horizon/selectivity) trap, this time caught prospectively by a control purpose-built to test for it rather than argued from analogy.
Key result
IS COT: 5,830 weekly rows, 7 currencies, 2005-01-04→2021-01-26. D1 price load ceiling 2021-03-03. Axis-1 raw signals 268 → 256 after FIFO → 254 IS (entry<2021-02-01). Composite (gated) IS trades: 54 (21.3% of the 254 Axis-1-alone IS trades pass the +/-1.0 z-gate). Shuffled-positioning replicates: 200 (seed 20260709). Gates: 4/6 PASS, 2/6 FAIL (gate 3 composite-criteria, gate 6 trade-count). Gate 2 parity: fresh contrarian_mean = +1.0063196573p/wk EXACTLY vs cot_positioning's committed figure, diff=0.0000 (tolerance +/-0.1p/wk). 5-arm IS net means (pips/trade): composite +23.26 (n=54, std 231.5, WR 57.4%, maxDD 895.1p); Axis-1 alone +14.24 (n=254, std 172.6, WR 55.1%, maxDD 2128.7p); Axis-3 alone +1.01 p/WEEK (n=641 weeks, reference, not comparable unit-for-unit); coin-flip +41.90 (n=54, std 215.5, WR 57.4%); shuffled-positioning null mean +21.25 / p5 -18.17 / p50 +21.04 / p95 +58.68 (200 reps, avg n/rep=41.4). Gate-3 criteria: (1) money CI [-43.29p,+87.66p] spans zero; (2) composite worse than its own coin-flip null, diff -18.64p, CI [-100.41p,+58.94p] spans zero; (3) composite (+23.26p) below the null's p95 (+58.68p), barely above the null's own mean (+21.25p). Walk-forward thirds: +83.22p (n=18), +1.43p (n=20), -16.92p (n=16) — 2/3 positive. Breadth 4/7 pairs gross-positive (GBP_USD, NZD_USD, USD_JPY, USD_CHF+; EUR_USD, AUD_USD, USD_CAD-), per-pair n=3-14, dominated by GBP_USD (+232.29p, n=8) and USD_CHF (+97.65p, n=5). Composite exit mix: 28 timecap (52%), 21 TP (39%), 5 SL (9%).
Indicators
D1 rolling swing extreme (L=25 completed bars, highest high / lowest low)ATR(14, D1) touch tolerance and TP/SL multiplesD1 tick-volume vs 20-bar trailing mean (low-volume gate)COT 156-week rolling z-score of net non-commercial position / open interestcarry (broker-truth 2020+, FRED-differential pre-2020, R9 splice)
Algorithms
Axis-1 first-touch swing-extreme fade detector + FIFO trade calendar (swing_axis1.py, simulate.py)Axis-3 COT crowding vote at |z|>=1.0, imported verbatim from cot_positioning (axis3_vote.py)composite AND-gate (Axis-1 fade taken only when Axis-3 vote agrees)coin-flip control on identical composite timestamps (seed 20260709)shuffled-positioning control: 200 block-preserving-by-currency permutations of the COT vote (seed 20260709)day-block bootstrap 95% CI for all 3 H1 money/vs-coin/vs-shuffled-null criteria3-fold walk-forward on IS thirds
Interactive version (search, filters, figures): the experiment explorer.