docs: freeze hierarchical intrabar replay plan

Author: dharmanan

research/regime-trend-v1/intrabar-replay-plan-v1.md

@@ -0,0 +1,128 @@
+# Regime Trend v1 — Frozen Hierarchical Intrabar Replay Plan
+
+Status: **Frozen before lower-timeframe replay results are viewed**
+
+Purpose: resolve the OHLC path ambiguity inside each 4-hour strategy candle and test whether an immediately executable profit target is reached before the active stop.
+
+The existing 4-hour strategy remains the signal generator. Lower-timeframe candles are used only to reconstruct event order and execution.
+
+## 1. Core correction
+
+A 4-hour candle containing both a favorable high and an adverse low does not reveal which event occurred first.
+
+Therefore 4-hour OHLC data alone may not be used to conclude that:
+
+- a target was reached before a stop,
+- a stop was reached before a target,
+- a favorable excursion was realizable,
+- waiting for the 4-hour close was necessary.
+
+## 2. Data boundary
+
+Use only:
+
+- Development: 2019-01-01 through 2022-12-31 UTC
+- Validation/research: 2023-01-01 through 2024-12-31 UTC
+
+The final holdout beginning 2025-01-01 remains closed.
+
+## 3. Hierarchical resolution
+
+### Stage 1: 1-hour replay
+
+Download and hash-complete Binance Spot 1-hour candles for BTCUSDT, ETHUSDT and BNBUSDT through 2024-12-31.
+
+Replay every frozen 4-hour baseline trade using the 1-hour candles contained inside its active lifetime.
+
+### Stage 2: selective 15-minute replay
+
+When the same 1-hour candle contains both the active stop and the tested target, classify the event as `AMBIGUOUS_SAME_1H`.
+
+Only these ambiguous 1-hour windows may be refined with 15-minute candles.
+
+If the same 15-minute candle still contains both levels, retain `AMBIGUOUS_SAME_15M`. Do not invent an intrabar path.
+
+## 4. Frozen immediate-exit targets
+
+Test only full-position immediate target exits at:
+
+- +0.50 frozen entry ATR,
+- +1.00 frozen entry ATR,
+- +1.50 frozen entry ATR,
+- +2.00 frozen entry ATR.
+
+The target is measured from the baseline entry fill.
+
+A target exit occurs immediately when the lower-timeframe price first reaches the target. It does not wait for a 4-hour close.
+
+No additional target may be introduced after viewing the replay results.
+
+## 5. Baseline stop schedule
+
+The original initial stop and 4-hour trailing-stop updates remain unchanged.
+
+- The initial stop is active immediately from entry.
+- A trailing-stop update calculated from a completed 4-hour candle becomes active only at the next 4-hour candle open.
+- A pending trend exit remains an exit at the next 4-hour open.
+
+Lower-timeframe replay must not use future 4-hour closes to update a stop early.
+
+## 6. One-hour event ordering
+
+For each active 1-hour candle:
+
+1. Check opening gaps.
+2. If open is at or below the active stop, the stop occurs first.
+3. Else if open is at or above the target, the target occurs first.
+4. If the candle range reaches both stop and target, classify `AMBIGUOUS_SAME_1H`.
+5. If only the target is reached, classify `TARGET_FIRST`.
+6. If only the stop is reached, classify `STOP_FIRST`.
+7. Otherwise continue to the next 1-hour candle.
+
+If no lower-timeframe target or stop occurs before the original baseline exit, retain `BASELINE_EXIT`.
+
+Any missing lower-timeframe interval inside the required replay window produces `DATA_GAP` rather than a guessed result.
+
+## 7. Execution costs
+
+Use the frozen baseline costs:
+
+- commission: 0.10% per side,
+- adverse slippage: 0.05% per side.
+
+A target touch uses the target reference with adverse sell slippage. A stop touch uses the active stop reference with adverse sell slippage. A gap through a stop uses the lower-timeframe candle open with adverse slippage.
+
+## 8. Required reports
+
+For each partition, symbol and target report:
+
+- baseline trade count,
+- `TARGET_FIRST` count,
+- `STOP_FIRST` count,
+- `AMBIGUOUS_SAME_1H` count,
+- `BASELINE_EXIT` count,
+- `DATA_GAP` count,
+- target-first count among baseline winners,
+- target-first count among baseline losers,
+- stop-first count among baseline winners,
+- stop-first count among baseline losers,
+- lower-bound total net PnL, treating same-hour ambiguity as stop-first,
+- upper-bound total net PnL, treating same-hour ambiguity as target-first,
+- profit-factor bounds,
+- win-rate bounds,
+- results by symbol and by calendar quarter.
+
+Also report the exact 1-hour timestamps requiring selective 15-minute refinement.
+
+## 9. Interpretation discipline
+
+The completed-bar profit-protection matrix is a separate test and must not be used as evidence against immediate intrabar targets.
+
+An immediate target candidate may advance only when:
+
+- its lower-bound result improves both development and validation/research,
+- the improvement is not carried by one symbol or one quarter,
+- ambiguity is low enough or resolved by selective 15-minute replay,
+- doubled execution costs do not remove the improvement.
+
+This stage does not authorize live trading, Pine implementation or final-holdout access.