回测基本一致

07-mt5-bridge.md

@@ -193,6 +193,31 @@ For a brand-new symbol you must first **let MT5 cache its history** (open a char
 `Tools → Options → Charts → Max bars: Unlimited`, scroll back) so the tester and the export have enough
 bars. Confirm the exact broker symbol code (it varies: indices and metals especially) before exporting.
 
+### 6a. Pull M1 even if your signal timeframe is higher — and pass it to the engine
+
+A common mistake: download only the signal timeframe (e.g. M5/M15/H1) and assume that's enough. It is
+not, **if your EA moves its SL during a trade** (break-even, trailing, basket trailing). The bar-level
+engine produces a −40% to −50% net gap on those EAs (doc 03 §7 measured failure mode) because the BE
+update and the SL trigger fall on the same bar's opposite extreme. The fix is tick-level exit
+simulation, which needs the **M1 bars covering the same window as your signal bars**:
+
+```python
+# Download BOTH timeframes. Same symbol, same window, exclusive end (match MT5 tester).
+m1_bars  = load_bars(DATA / f"{SYMBOL}_M1_{start}_{end}.parquet")
+m5_bars  = load_bars(DATA / f"{SYMBOL}_M5_{start}_{end}.parquet")   # the signal timeframe
+
+# Slice both to the exact same window (exclusive end matches MT5 tester's ToDate).
+m1_bars  = m1_bars[(m1_bars["timestamp"]  >= start) & (m1_bars["timestamp"]  < end)]
+m5_bars  = m5_bars[(m5_bars["timestamp"]  >= start) & (m5_bars["timestamp"]  < end)]
+
+# Pass m1_bars to the engine — it switches to tick-level exit simulation automatically.
+result = engine.run(m5_bars, signals_long, signals_short, sl, tp,
+                    instrument, sizing, deposit, m1_bars=m1_bars)
+```
+
+If your EA does **not** move its SL intra-trade (clean market entries with a fixed SL/TP), `m1_bars`
+can be omitted — the bar-level engine is exact for that class and faster.
+
 ---
 
 ## 7. Parsing the report
@@ -223,10 +248,15 @@ For every finalist, write an `auto-verification.md` that puts the two tiers side
 | Equity DD max | … | … | …% |
 | Total trades | … | … | …% |
 
-Then judge the delta **against the expected fidelity gap** (doc 03 §7): a clean-directional setup
-should show only a modest negative gap (MT5 a little below Python); a trailing/grid setup in volatile
-history can gap much wider, and that's *expected*, not a bug. The decision rule: if the **MT5** number still clears your bar after the gap,
-the finalist is real; if the edge only existed in the optimistic Python figure, discard it.
+Then judge the delta **against the expected fidelity gap** (doc 03 §7 + §8 target-gate table):
+a clean-directional setup (bar-level engine, SL not moved intra-trade) should show ≤ ~2% net gap;
+a trailing/BE setup on the **bar-level** engine is not trustworthy at all (−40% to −50% gap,
+doc 03 §7 measured failure mode) — **before** judging the gap "expected", confirm the Python run
+used M1 tick-level exit simulation (`m1_bars=` passed to the engine), which brings the gap into the
+≤ ~10% range. Only a gap inside the §8 target gate counts as "expected"; a wider gap is a
+missing-M1 / wrong-engine-mode bug, not fidelity noise. The decision rule: if the **MT5** number
+still clears your bar after the gap, the finalist is real; if the edge only existed in the
+optimistic Python figure, discard it.
 
 > A useful warm-up calibration: pick one known preset and run the full Python-vs-MT5 comparison on it
 > first. It tells you *your* stack's actual gap for *your* EA, so later finalists are judged against a