investment-journal-review

Review one completed investment/trade or a journal of many decisions while preserving the original plan, separating process quality from profit/loss, testing thesis and forecast accuracy, identifying

SKILL.md

GitHub

Investment Journal Review

Own learning after a decision. Support two modes: a single-decision postmortem and a multi-decision pattern review. Never judge process solely from P&L.

Read references/journal-protocol.md. Use scripts/journal_metrics.py for a normalized trade CSV when available. Use $probabilistic-market-forecast scoring for matured probability records and $investment-thesis-tracker for evidence deltas.

Workflow

  1. Freeze provenance. Preserve the original timestamped thesis, evidence cutoff, forecast, entry/exit plan, size, loss limit and any amendments. Never edit the old record; append review fields.
  2. Reconstruct execution. Record actual entry/exit, fees, spread/slippage, partial fills, holding time, deviations and whether the planned invalidation or review rule was observable.
  3. Separate four outcomes. Classify good process/good outcome, good process/bad outcome, bad process/good outcome and bad process/bad outcome. Luck does not repair a broken process; a valid process can lose.
  4. Score the forecast. When a probability or interval matured, compare the frozen forecast with the realized outcome and a declared naive/market benchmark. Preserve unresolved or early-closed cases separately.
  5. Diagnose cause. Distinguish research/evidence, thesis, timing, sizing, portfolio concentration, product mechanics, execution, discipline and unforeseeable shock. Avoid vague labels such as “emotion” without observable behavior.
  6. Analyze patterns. Normalize P&L by planned risk (R) and segment only when sample size permits. Compare repeated errors, regime, horizon, strategy and instrument; report missing data and selection bias.
  7. Change little. Recommend at most one or two falsifiable process changes for the next sample, with a measurement and review date. Do not optimize rules to explain the past perfectly.

Hard rules

  • Never use later information to claim it should have been known at the original cutoff.
  • Never celebrate a rule-breaking winner as a good decision.
  • Never infer skill from a handful of trades, one regime or selected successes.
  • Include costs, abandoned ideas and no-trade decisions when available; survivor-only journals overstate quality.
  • Do not respond to losses by automatically increasing risk, leverage or trading frequency.

Output

For one decision: original plan, evidence available then, execution delta, process/outcome quadrant, forecast score, cause, retained lesson and one next experiment.

For a journal: coverage/sample warnings, expectancy in R, win/loss distribution, drawdown, process adherence, forecast calibration when available, recurring strengths/errors, regime dependence and one or two measurable changes.