# MIROSCOPE PHASE 4 - BACKTEST REPORT
Run: 2026-09-22. Strategy: P1 detector + P2 EA setup/lifecycle + P3 risk engine (Python ports).
Account: $10,000 start, 1% risk/setup, lane 0 (D1 bias -> H1 trigger), session cutoff 21:00 UTC,
max 5 setups/day, -3% daily loss halt. Spread/commission/slippage: NOT modeled (optimistic bias noted).

## DATA SOURCE: REAL HISTORICAL (eurusd-h1.csv)
Range: 2025-01-01 to 2026-09-22; rows: 10605 (H1, Yahoo Finance EURUSD=X).
This tests the ported logic on real prices. Still not proof of live profitability:
no spread/commission/slippage modeled, fills assumed at bar extremes, and the .mq5
itself was not compiled/run (no MQL5 compiler or MetaTrader here).

## RESULTS
- Trades: 100 (wins 31)
- Win rate: 31.0%
- Profit factor: 0.78
- Expectancy: $-13.82 per trade (-0.139R)
- Max drawdown: 27.59%
- Return: -13.82%  (final balance $8618.17)

## PAPER TRADING (live market data, zero real money): BLOCKED
Requires: a demo/broker account, MetaTrader, and the .mq5 compiled in MetaEditor.
None exist in this environment. This is the gate that needs Michael:
a demo account for the paper-trading phase (per the build plan), then his explicit
authorization before anything touches Phase 5. Nothing here was faked to cover it.

## WHERE IT LOSES (exit diagnostic, same 100 trades)

- 69 trades: straight to full stop-loss (-1R each). Median trade outcome is -1.0R.
- 16 trades: TP1 partial then runner reached TP2 (the big winners, up to +3.02R).
- 15 trades: TP1 partial then the runner reversed into the breakeven stop
  (small net wins, roughly +1.4R each).
- Max concurrent positions: 2. The drawdown is a bad streak, not overleveraging.

Read: the detector is selective (100 setups in 20 months) and 31% of trades do
reach TP1, but only 16% ever see TP2 while 69% take the full stop. The 1.5R/3R
payoff structure needs a win profile this pattern is not producing on EURUSD H1.
Per the build plan, weak numbers here mean fixing the logic, not the test: the
next loop is target/structure tuning (TP multiples, stop placement, gate
tightness) re-run against this same dataset, then paper trading. That loop is
not started here: parameter choices are trading decisions for Michael, and
paper trading needs a venue that does not exist in this environment.