E-15. Pair trading scan — 78 pairs × 6 parameters (2026-07-14)

E-15. Pair trading scan — 78 pairs × 6 parameters (2026-07-14)

• Run: run_id=pairs_full, all 468 runs, 0 failures. Preregistered under D-018. Cumulative ledger 5,344.
• Result: Stage 0 (trades ≥30) 7/468 → Stage 1 (DSR>0.5) 0. 0 survivors.
• The key diagnosis is not Sharpe but absence of activity: median trade count 8 per 2.4 years. 97% had 1~29 trades. Rolling EG pass periods are short and rare on real crypto pairs — the phase6-plan 5a warning about "unstable coin cointegration" was confirmed by measurement.
• Even the 7 eligible had annual Sharpe -0.65~+0.61 — even the best (DOGE-NEAR +0.61) reaches, against the null ceiling (3.64), only 1/6 of it. No economics even where there is activity.
• Pattern: the most active were all alt-alt (mostly DOT pairs) — pairs sharing sector beta trip cointegration more often [inference]. window 720 + z_in 1.5 was the most active (mean 15 trades).
• Mechanical integrity: 0 errors, 0 bankruptcies, EG test and all simulation tests pass — a clean case of "the mechanism works, the edge is absent".
• Recorded limits: (a) n_skipped_entries was not written to the ledger — the contribution of min_notional skips cannot be separated (consider adding the column in the next scan). (b) Assumes a taker + slippage cost structure — the conclusion might differ under maker fills, but that is a live-infrastructure issue (Phase 7+).
• Verdict: strategy 4 rejected (on 1h, taker costs and the registered search space). A retry only with a different mechanism definition (other resolution / measured maker slippage) — re-running the full protocol.