Trade Review

When to Use

• After every closed (or cancelled, unfilled) entry. The order gate blocks the
  next entry until each entry has a review.
• End of session: distil 1–3 lessons.

How It Works

1. Gather facts. search_trades for the fills (P&L is on the closing fill;
   fees are separate), search_orders for what was placed and modified, and
   journal_read {kind:"plan"} / {kind:"order_placed"} for the plan and
   rationale.
2. Look back before grading (this is how the desk learns):
   • What you saw when you entered. Autonomous: the cycle that placed
     the entry in <FTH_HOME>/logs/cycles-<day>.jsonl (match its time and
     orders), then its record in logs/cycles/<day>/: the prompt (the last
     10 bars, what fired and its confluence, the trend rule, your last 10
     decisions) and what you did. Interactive: the plan entry. Then the bars
     after the entry (the harness tool bars_fetch with name: "review", or the runner's file): was the read wrong,
     or the execution?
   • Your last 10 trades. The harness tool lessons (it starts
     with the form of the last 10: record, expectancy, repeated mistakes)
     and journal_read {kind:"review", tag:"setup:<name>"} for this
     setup's last reviews. Is this a mistake you already made? If the same
     mistake:* tag appears twice in the last 10 trades, write the lesson
     now (with the trades as evidence), not at end of day.
3. Compute. Planned risk $ = |entry − stop| ÷ tickSize × tickValue × size.
   R = net P&L ÷ planned risk. MAE/MFE from get_bars over the hold, if useful.
4. Grade the process (A–F) independently of the result:
   • Was the trigger real (the plan's condition actually happened)?
   • Entry at the planned price ± slippage? Stop placed at the exchange at once?
   • Managed per the plan (no widened stop, no early panic exit, no moved target)?
   • Size per the position-sizing rules?
5. Write the review with journal_add {kind:"review", contractId, text, tags}:
   • text: plan vs. actual, R, grade, what to repeat or change.
   • tags: exactly one of result:win | result:loss | result:scratch (|R| < 0.2) | result:nofill; setup:<name>; regime:<primary> (from the plan); the
     symbol; r:<R net of fees> (e.g. r:-1.11); any mistake:<kind>
     (mistake:chased, mistake:no-stop, mistake:moved-stop,
     mistake:early-exit, mistake:oversize, mistake:rule-break).
     These tags are how the desk learns: the lessons tool
     turns them into instincts (each setup's record by regime, recurring
     mistakes) that every later cycle sees. A review without them teaches
     nothing.
6. Lessons (end of session) with journal_add {kind:"lesson", tags}:
   one rule, the evidence count, and the condition. Don't write a lesson from a
   single trade unless it's a rule break.
7. Also review blocked orders: the server's (order_blocked journal
   entries) and the harness gate's (<FTH_HOME>/logs/gate-log.jsonl, FTH_HOME defaulting to ~/.futures-trading-harness, with the
   checks that refused each). A block means the plan or sizing was wrong
   before a guardrail had to say so.

Examples

review: "ORB long MNQ. Plan: break of 21500 OR high, stop 21490, target 21520,
1 contract, $20 risk. Filled 21500.75 (3 ticks slip). Stopped at 21490 after a
sweep of the OR high and close back inside: 43 ticks = $21.50 + $0.74 fees = $22.24
net loss, R = -22.24 / 20 = -1.11. Process B: trigger valid,
but relative volume was 0.9x, which the strategy says to skip."
tags: ["result:loss", "setup:orb", "MNQ", "regime:trend-up", "r:-1.11", "mistake:rule-break"]

lesson: "ORB on MNQ with relative volume < 1.0x: 1W/5L over 6 trades. Skip."
tags: ["setup:orb", "MNQ"]