reel-scorecard
Skill14 starsChanged 35 days agoMITView on GitHub
Reads trial-reel and TikTok results at 72 hours and says which hook won each idea, which reel to graduate to the main feed (the creator taps it), which hooks to kill and which ideas to retire, using matched windows and one data source so numbers are comparable. Pulls Zernio analytics read-only (or a CSV export), joins them to the trial-post ledger, and writes SCORECARD.md plus a weekly learning log that changes at most one rule a week. Use for 'scorecard', 'which hook won', 'what should I graduate', '72 hour results', 'how did the trials do'.
In this repository63 other files it gives its agents.
- CLAUDE.mdCLAUDE.md
- board-advisorSkill
- bottleneck-removalSkill
- business-auditSkill
- business-starterSkill
- competitor-researchSkill
- founder-independenceSkill
- hand-it-offSkill
- happy-customerSkill
- hiring-replacementSkill
- meeting-follow-upSkill
- monday-numbersSkill
- new-hireSkill
- no-meeting-mondaySkill
- obsidian-brainSkill
- ops-deskSkill
- safe-resaleSkill
- spreadsheet-dashboardSkill
- warm-leadSkill
- downloadable-asset-engineeringSkill
- html-to-pdfSkill
- linux-server-opsSkill
- precision-pdf-editingSkill
- repo-radarSkill
- secure-credentialsSkill
- security-auditorSkill
- best-timeSkill
- brand-deal-filterSkill
- comment-replySkill
- content-monthSkill
- daily-socialSkill
- dm-closerSkill
- idea-bankSkill
- marketing-teamSkill
- reel-ideationSkill
- trial-postSkill
- viral-breakdownSkill
- youtube-seoSkill
- caption-engineSkill
- content-multiplierSkill
- edit-styleSkill
- footage-ingestSkill
- longform-editSkill
- podcast-repurposeSkill
- raw-to-reelSkill
- ready-to-filmSkill
- reel-captionsSkill
- script-polishSkill
- thumbnail-generatorSkill
- use-local-whisperSkill
- conversion-funnel-auditingSkill
- llms-txtSkill
- seo-agentSkill
- website-designerSkill
- website-devSkill
- be-conciseSkill
- humanizerSkill
- launch-weekSkill
- linkedin-writerSkill
- newsletterSkill
- The first 60 shown.
The file
We split this skill into the rules it gives an agent
1 rule. About 0 are common, 1 is distinctive, and 0 are very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈1
Distinctive · estimate
This author's own choices. Worth a look.
0
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Distinctive: line 21
Same source, same window: Only Zernio numbers against Zernio numbers, only 72 to 96 h old against each other.
≈ “Common” is an estimate: at least 1 in 5 of the 364,980 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.
Files
Reel scorecard: the 72-hour read
Last step of the pipeline: idea -> script -> record -> "the card is in" -> edit -> trial-post -> reel-scorecard,
and it feeds the next reel-ideation run (winning hooks and formats get copied again, losers retire).
Requirements
Claude Code (or Claude.ai with a CSV export pasted). The Zernio MCP connector for read-only analytics, and the
trial-post ledger ($HOME/.trial-post/ledger.jsonl), which carries each post's idea and hook ids.
cp config.example.json config.json to change the thresholds.
Rules that keep the numbers honest
- Same source, same window. Only Zernio numbers against Zernio numbers, only 72 to 96 h old against each other. Instagram insights lag up to 48 h, TikTok fills in 24 to 48 h: a 24 h look is a peek, never a decision.
- Views are shown, never used alone. Instagram: skip rate and hold (average watch / length) decide. TikTok: completion rate first, For You share next to it.
- No peak without its decay. Quote a number with its window and date, or do not quote it.
- Follows are not in the API for reels. The creator checks follows for the finalists in the app before graduating.
Steps
- Pull (read-only). For every ledger post 72+ h old:
mcp__zernio__analytics_get_analyticswithpost_id(or withplatform+from_datefor a batch, paginated). Save each result toscorecard/raw/<date>-<n>.txt. - Score.
python3 scripts/scorecard.py --ledger $HOME/.trial-post/ledger.jsonl --analytics scorecard/raw/*.txt --out scorecard/<date>.md - Read it out to the creator, short: per idea, the winning hook and why (skip, hold), graduate candidates (max 2 a day so the main feed is not flooded), kills, retirements, and what is not in the window yet.
- Graduation is the creator's tap: Instagram app > the trial reel > "Share to everyone". There is no API for it. One graduation per idea. Never re-upload the trial file through another scheduler (duplicate content); a main-feed repost of a winner is a new cut or a new take.
- Weekly learning log (once a week): for every idea with 2+ hooks scored, which hook TYPE won (callout, belief
flip, proof...), which format, which opening visual. Append to
scorecard/LEARNING-LOG.md. Change at most ONE rule per week, and only after 3+ tests agree. Feed the winners to the nextreel-ideationrun.
Default thresholds (config.example.json; tune with your own numbers after two weeks)
| decision | Instagram rule |
| graduate candidate | lowest skip AND hold >= 28% AND skip < 55% AND >= 1,000 views |
| split | best skip and best hold on different hooks: the creator picks |
| tie | skip within 2 points (skip rate is an estimate) |
| kill | skip >= 65%: that hook never graduates and is not reused on that idea |
| retire the idea | every hook >= 60% skip and under ~1,300 views |
These come from one creator's account; they are a starting point, not a law.
Output
scorecard/
raw/<date>-<n>.txt saved analytics results (never edited)
<date>.md the scorecard
LEARNING-LOG.md one dated block per week, the one rule changed (if any) and the tests behind itFailure handling
- A post missing from analytics: it may still be processing; list it under "not in window yet".
- Ledger has no idea/hook ids (old posts): add them by hand to the ledger lines, or score those posts by eye.
- Numbers look impossible (skip 0, views 0 after days): say so, do not decide on them, check the post in the app.
Does it work?
Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.
No reviews yet
People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.
- It worked
- Partly
- It didn't
Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.
Your trail
Files you open will collect here, so you can find them again.
Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.
Now do this for your own.
Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.

