# Performance Debrief

> Judge a published piece against comparable ones, and take one lesson from it.

- Skill: `tinyhumansai/performance-debrief` (Agent Skill)
- Install (CLI): `npx skillmds@latest add tinyhumansai/performance-debrief`
- Raw SKILL.md: https://api.skillmd.com/api/skills/tinyhumansai/performance-debrief/raw
- Safety review: pending (external: skill-scanner PASS, skillspector PASS)
- Works with: Claude Code, Claude.ai, OpenAI Codex
- Category: Coding & Dev Tools
- Author: tinyhumansai (https://skillmd.com/u/tinyhumansai)
- Updated: 2026-09-22
- Page: https://skillmd.com/skills/tinyhumansai/performance-debrief

---


# Performance Debrief

Judge against something. A view count on its own is a number, not a result.

## When to use

- A piece has been out long enough for its numbers to settle.
- A format is being considered for repetition or retirement.

## Steps

1. **Compare against similar pieces,** not against hope or against the best
   piece ever made. Same format, similar length, similar topic.
2. **Read the retention curve before the comments.** Where people left is a
   fact; why they say they left is a story, and the two frequently disagree.
3. **Separate the hook from the piece.** A low click rate and a high retention
   is a thumbnail problem; the reverse is a content problem, and treating one as
   the other wastes the next month.
4. **Check the calendar effect.** Cadence, day, and whether the last piece
   underperformed — audiences respond to the sequence, not to the single piece.
5. **Take one lesson,** not five. Five lessons from one video is a superstition
   generator.
6. **Record it** — `performance` on the content row, and any real pattern on
   `audience-signals` or the baseline's `learnings`.

## Output

The content row updated with what it actually did against comparable pieces, and
one recorded lesson. "Indistinguishable from our usual" is a legitimate and
common result.