docs: invalidate truncated 5m replay horizon

Author: dharmanan

research/regime-trend-v1/intrabar-replay-errata-001.md

@@ -0,0 +1,57 @@
+# Regime Trend v1 — Intrabar Replay Errata 001
+
+Status: **Accepted before corrected replay results are viewed**
+
+## Invalidated report
+
+The first `intrabar-replay-report.json` produced by the original 5-minute replay implementation is invalid and must not be used for research conclusions.
+
+The invalid output was identifiable because every target reported zero `STOP_FIRST` trades despite the baseline ledger containing initial-stop and trailing-stop exits.
+
+## Root cause
+
+The baseline 4-hour engine records a stop exit timestamp as the opening timestamp of the 4-hour candle whose open or low triggers the stop.
+
+The original replay iterated only while:
+
+`five_minute_timestamp < baseline_exit_timestamp`
+
+This excluded the entire 4-hour stop-exit candle, including the five-minute candle in which the stop actually occurred.
+
+The original replay also excluded a trailing-stop update whose activation timestamp equaled the baseline stop-exit candle open. Such an update is active for that candle and must be included.
+
+## Corrected replay horizon
+
+### Baseline trend exit
+
+A baseline `trend_exit` executes at the recorded 4-hour open. Replay remains strictly before the baseline exit timestamp:
+
+`[entry_timestamp, baseline_exit_timestamp)`
+
+The baseline trend exit has precedence at its recorded open.
+
+### Baseline initial or trailing stop exit
+
+A baseline `initial_stop` or `trailing_stop` timestamp identifies the opening of the 4-hour candle containing the stop event. Replay must include the full candle:
+
+`[entry_timestamp, baseline_exit_timestamp + 4 hours)`
+
+Within that final 4-hour candle, five-minute candles determine whether the frozen target or active stop occurs first.
+
+## Stop-update activation
+
+A stop update calculated at 4-hour timestamp `T` becomes active at `T + 4 hours`.
+
+An update with activation timestamp equal to the baseline stop-exit candle open must be included.
+
+## Consistency failure
+
+If a baseline stop-exit candle has complete five-minute coverage but replay finds neither the target nor active stop during that candle, classify the trade as `DATA_MISMATCH` rather than silently retaining `BASELINE_EXIT`.
+
+A missing five-minute candle before resolution remains `DATA_GAP`.
+
+## Research consequence
+
+All net-PnL, profit-factor and event-count results from the invalid first replay are discarded. Only the corrected report may be interpreted.
+
+The final holdout remains closed.