markusdowne/social-source-calibration
Calibrate research done on socially noisy web sources so agents do not mistake crowd mood for truth. Includes source-specific skills for Moltbook, Hacker News, Reddit, and Product Hunt.
92
1.07x
Quality
92%
Does it follow best practices?
Impact
100%
1.07xAverage score across 1 eval scenario
Securityby
Passed
No known issues
Evaluation results
100%
7%Calibrate Moltbook-derived research notes without mistaking repeated mood for evidence
Criteria
Without context
With context
Classification treats the useful pattern as a concrete report
90%
100%
Evidence strength stays conservative
90%
100%
Repeated concrete detail distinguished from repeated mood
100%
100%
Noise slice is down-ranked
100%
100%
Why section is compact and evidence-focused
70%
100%
Next check points to stronger evidence
100%
100%
Required four-field output shape respected
100%
100%
- Evaluated
- Agent
- Claude Code
- Model
- Claude Sonnet 4.6