signal-postmortem

Automated post-trade analysis classifies realized trading outcomes into categories like true positive, false positive, missed opportunity, and regime mismatch. It compares predicted signal direction against realized return thresholds across fixed holding periods to distinguish skill errors from market regime shifts.

You run the postmortem recorder on an aggregated signals JSON file containing fields like signal_id, ticker, signal_date, and predicted_direction. The script processes batch outcomes and writes classified post-trade JSON records with realized returns and exit details into your reports directory.

Signal postmortem evaluation requires input JSON fields including signal_id, ticker, signal_date, predicted_direction, source_skill, and entry_price. These fields allow the recorder to compute realized 5-day and 20-day returns and classify outcomes against regime context for each matured signal.

Post-trade signal evaluation requires Python 3.9 or higher and uses the requests library for optional FMP price fetching. You can supply entry prices directly in the input signal JSON or enable automated price retrieval for computing realized holding period returns.

Post-trade signal classification generates weight-adjustment JSON for downstream edge-signal-aggregator calibration and produces YAML backlog entries for targeted skill improvements. This feedback loop turns realized outcome categories into actionable calibration data for refining future signal generation.