回测

pv 0.7.1 = pv 0.7.0 + the portfolio cap inputs (engine build 0.7.2): 4 open positions / 1.0% open risk / 2.0% daily account loss / 3 per currency, evaluated by every instance on the account before an order (Helpers/KurosawaPortfolio.mqh). No trading value changed; the tester runs one instance so the caps never bind and the result is identical to 0.7.0 on this window. Tier unchanged (regime edge). Headless run.

Tier set: REGIME edge (2023-2026 passed, 2019-2022 PF 0.86-0.93). Status: live at minimum lot with the kill switch.

Regime stress test: proven pv 0.6.0 on a third window, 2019-2022, never used for any decision. Generated ticks. Covers low-vol 2019, COVID 2020, calm 2021, the 2022 rate shock. RESULT: below 1.0 - status reduced from proven to candidate (regime edge, 2023-2026); kept at minimum lot as a forward test.

Engine 0.6.0 identical-check of the proven pv 0.6.0 (D1 gate off, direction switches on). IDENTICAL to 0.5.1.

Engine 0.5.1 re-run of the proven pv 0.6.0 (clock moved to broker server time). Result identical to the 0.5.0 run: the tester's TimeGMT was already server time. WALK-BACKWARD out-of-sample. Generated ticks.

Engine 0.5.1 re-run of the proven pv 0.6.0 (clock moved to broker server time). Result identical to the 0.5.0 run: the tester's TimeGMT was already server time. IN-SAMPLE.

pv 0.6.0 (MaxAdx 34, long-only) WALK-BACKWARD out-of-sample; tune chosen on 2025-01-01..2026-09-05. Generated ticks (no tick history before 2025).

pv 0.6.0 (MaxAdx 34, long-only) IN-SAMPLE baseline: both values chosen on this window.

pv 0.5.0 (MaxAdx 31, long-only) WALK-BACKWARD out-of-sample; tune chosen on 2025-01-01..2026-09-05. Generated ticks (no tick history before 2025).

pv 0.5.0 (MaxAdx 31, long-only) IN-SAMPLE baseline: both changes were chosen on this window.

WALK-BACKWARD out-of-sample: pv 0.4.0 was tuned on 2025-01-01..2026-09-05; this window was never seen. Generated ticks (OANDA has no tick history before 2025).

pv 0.4.0 baseline (MaxAdx 22->34). IN-SAMPLE: the gate optimisation that chose 34 ran on this same window. Cross-pair validation follows. | Corrected 2026-09-09: configured for real ticks, but OANDA tick history begins 2026-01-02; 40.2% of the window is real ticks, the rest generated from M1. Derived from the tester log; the report was overwritten.

20-month real-tick run of the 09-04 M15 config. NOTE: tester ran the leftover RangeRevertEA inputs, not RangeRevert_Multi_M15 - risk sizing ON (engine default), magic 2026090302. | Corrected 2026-09-09: configured for real ticks, but OANDA tick history begins 2026-01-02; 40.2% of the window is real ticks, the rest generated from M1. Derived from the tester log; the report was overwritten.