E-19. Master architecture v1.0 + E-17A directionality gate (2026-07-15)

E-19. Master architecture v1.0 + E-17A directionality gate (2026-07-15)

• Foundation: 36 trials registered in reports/experiment_registry.sqlite before execution, storing code commit/dirty, data SHA-256, period, costs/fill/latency, parameters, metrics and discard reasons. Every trial COMPLETED; 0 missing or left RUNNING.
• Fill foundation: F0 touch / F1 trade-through / F2 risk-averse queue; partial fills, cancel race, post-only, OCO, reduce-only and position reconcile separated into an event core. Same-bar ambiguity in the OHLCV engine is made explicit with counts and a worst-case policy.
• Direction experiment: HTF 1h/4h, LTF 1m/5m, 3 trend definitions, 3 neighbouring lookbacks = 36 runs. Next-bar market entry and a 12-bar hold keep execution alpha out of the measurement.
• Result: preliminary train pass 0. The best, 5m/1h structure/lookback20, traded 2.97 times per day, 2,611 signals, +0.00753%/signal before costs, but -0.19247% after baseline costs and -0.29247% after 1.5x costs. The 1m combinations mostly traded 17~24 times per day, exceeding even the target frequency.
• Verdict D-025: Stage 1 rejected, val/test not opened, E-17B~H not run. The system did not find a strategy; it blocked a path that would have dressed up weak directionality with limit-order assumptions.