project

scout

A developer's personal LLM judge that reads Reddit threads and classifies each as worth a comment, an upvote or nothing, with its rules kept in a JUDGE.md file.

Current clusters

build1 publisher

Rerunning the same eval suite three times in ten minutes moved its score by two cases

A developer who built his own eval gate reran an unchanged 21-case suite three times in ten minutes and got three different scores, with the run files recording identical prompt checksums each time.

Publishers:dev.to

Reality

Evidence60
Adoption20
Hype gap−15
Incentives40
Confidence55