Weekly Performance Digest reads closed theses from the trader memory state folder and aggregates the ones whose exit date falls in a chosen week. It computes headline metrics, then attributes wins and losses across six dimensions: source skill, exit reason, thesis type, sector, mechanism tag and screening grade.
It also lists the biggest winners, losers and process lessons. The script writes a JSON record and a Markdown report, and an empty week still produces a valid zeroed report. It is pure local calculation with no API key, and it describes past results without giving advice.
What You Get
- A JSON report and a Markdown report named by the end date of the week
- Headline metrics: win rate, expectancy, profit factor, averages, largest winner and loser, holding days, R-multiple average and spread, MAE and MFE
- Pattern analysis tables by source skill, exit reason, thesis type, sector, mechanism tag and screening grade
- Top winners, top losers and process improvement notes
Common Use Cases
- Reviewing a trading week in one page
- Combining four weekly digests for a month-end review
- Finding which screener or exit reason produced most of the losses
What Makes It Unique
R-multiples are calculated only when the recorded stop gives a positive risk, so gap fills and raised stops stay in P&L and win-loss counts and are left out of R averages. It is also built to feed the toolkit’s postmortem and coaching skills, with the double-counting rules for partial trims written into the method.
Under The Hood
The agent runs one Python script from the shell. The script reads thesis YAML files from the state folder that the trader memory skill writes, needs PyYAML, makes no network calls and uses no API key. It writes the JSON and Markdown reports into an output folder. A reference file explains the metric formulas.