Instrumentation¶
Goal: Maintain measurable metrics on workflow friction so changes can be evaluated against real data, not vibes.
Three measurable metrics¶
| Metric | Definition | How captured |
|---|---|---|
| Lead time | Wall-clock time from "task captured" → "merged or accepted" | daily-log.md timestamps |
| Context switches per task | Count of distinct app/window/repo switches between task start and finish | daily-log.md tally marks |
| Rework rate | % of tasks requiring a second corrective pass after first "done" | daily-log.md outcome flags |
Secondary signals (lightweight): perceived friction (1-5), capability used (subagent / skill / MCP / none), and any blocked-on-tool moments.
Files in this phase¶
metrics.md— exact metric definitions and rules of thumb.daily-log-template.md— per-task line format. Append-only.baseline-report-template.md— the 3–5 day rollup format.weekly-retro-template.md— keep/kill decisions and one calibration change per week.
Workflow¶
- Day 0: read
metrics.mdonce, copydaily-log-template.mdintodaily-log.md(top of file is current day). - Each task: add one row when starting, fill outcome when finishing.
- Day 3-5: generate the first baseline report from
baseline-report-template.md. - End of each week: weekly retro using the template. Make exactly one calibration change before next week.
Success gate¶
A baseline report exists with at least 10 logged tasks across ≥3 days, and the next phase can reference concrete numbers instead of opinions.