signal-postmortem

---
name: signal-postmortem
description: Use when recording or analyzing post-trade outcomes for signals from the edge pipeline or other skills — tracking false positives, missed opportunities, and regime mismatches for weight feedback.
category: Performance
version: 1.0.0
tools: []
---

# Signal Postmortem

Record and analyze the outcomes of trading signals generated by the edge pipeline, screeners, and other skills, to close the feedback loop back into signal weighting and skill quality.

## When to Use

Use after a trade has closed and the user wants the outcome recorded; when reviewing a batch of signals that have reached their 5-day or 20-day holding period; to identify systematic false-positive patterns from specific skills; to generate feedback for `edge-signal-aggregator` weight calibration; to build a skill-improvement backlog from decision-quality metrics; or for periodic (weekly/monthly) signal-quality audits.

## Inputs

Gather closed or matured signal records, each ideally including: a unique signal ID, ticker, the date the signal was generated, the predicted direction (LONG or SHORT), which skill generated it, and, optionally, the entry price. If the realized-return price data isn't already known, ask the user for the exit price and exit date, or use web_search to find historical prices for the ticker and dates in question.

## Workflow

**Compare predicted vs. realized.** For each signal, compare the predicted direction against the realized 5-day and 20-day returns.

**Classify the outcome** into one of four categories:

| Category | Definition |
|---|---|
| TRUE_POSITIVE | Predicted direction matched realized return sign |
| FALSE_POSITIVE | Predicted direction opposite to realized return |
| MISSED_OPPORTUNITY | Signal not taken but would have been profitable |
| REGIME_MISMATCH | Signal failed because of a market regime change |

**Generate feedback.** Roll individual postmortems up into: (a) suggested weight adjustments for `edge-signal-aggregator` — e.g. a skill with a 15% false-positive rate concentrated in one regime gets its weight suggested downward, with the sample size and reason attached — and (b) skill-improvement backlog entries (issue type, severity, evidence, suggested action) for skills showing a systematic problem pattern.

**Summarize.** Produce aggregate statistics grouped by skill, by ticker, and by time period (e.g. by month) so patterns are visible across the whole postmortem history, not just one signal.

## Output

Return, in chat:

- **Per-signal postmortem record** — signal ID, ticker, signal date, source skill, predicted direction, entry price, realized 5-day/20-day returns, exit price/date, holding days, outcome category, regime at signal vs. at exit, and any notes.
- **Weight feedback** — per-skill current weight, suggested weight, reason, and sample size, plus an overall confidence level.
- **Skill improvement backlog entries** — skill, issue type, severity, evidence (e.g. false-positive rate, sample size, regime correlation), and suggested action.
- **Summary report** — grouped statistics by skill, ticker, and period.

## Key Principles

1. **Honest attribution** — every outcome is attributed to its source skill for accountability.
2. **Regime awareness** — record regime context to distinguish genuine skill failure from a market regime shift.
3. **Minimum sample size** — require 20+ signals before suggesting a weight adjustment; treat anything below that as directional only.
4. **Feedback loop closure** — findings should flow back into both signal-aggregation weights and the skill-improvement backlog, not sit unused.