Turn every post into a calibrated experiment.
Score it, predict it blind, ship it, then let the real numbers correct your taste. The compounding asset is your judgment, not any single video.
npx skills add https://github.com/appleweiping/content-flywheel
Predict first. Grade honestly. Compound.
Five small steps. The only hard part is writing the prediction before you peek, and this enforces exactly that.
Score
Rate the draft on your rubric, before it ships.
Predict
Write a blind forecast, then lock it read-only.
Ship
Publish. Record the link. Touch nothing.
Retro
At T+3, real numbers meet your locked bet.
Evolve
Upgrade the rubric only if it beats history.
The discipline, in three lines.
Blind prediction
The forecast is written before any data exists, and frozen once locked. Hindsight is the enemy of calibration.
Upgrade equals full rescore
Changing a rule means re-scoring all past samples and beating the old rule on the data. No cherry-picking one lucky hit.
Rubric is a workbench
Beliefs the data disproves get deleted, not hoarded. Git history is the archive, not the document.
Drop it into your agent.
Works with Claude Code, Codex, and the skills ecosystem. Nothing to build, no service to run. It is format-agnostic: B站, YouTube, 抖音, X, 小红书.
npx skills add https://github.com/appleweiping/content-flywheel
Then say initialize, score this, predict, shipped, retro, evolve.
Standing on better-funded shoulders.
A lightweight, honest companion to work that came first. If the heavier version fits you better, use theirs.