Skip to content

feat(engine): log per-pass judgement frequency (judged vs cached)#29

Merged
thejefflarson merged 1 commit into
mainfrom
feat/jef-log-judgement-rate
Jun 21, 2026
Merged

feat(engine): log per-pass judgement frequency (judged vs cached)#29
thejefflarson merged 1 commit into
mainfrom
feat/jef-log-judgement-rate

Conversation

@thejefflarson

Copy link
Copy Markdown
Owner

Surfaces how often the engine calls the model: per pass it logs adjudication pass entries=N judged=K cached=M + a judged_this_pass gauge (judged = cache misses = model calls; cached = reused). Steady state should be judged≈0; sustained nonzero = fingerprint churn. Also fixes a stale comment — judging is once per entry (holistic), not 'once per path'. 🤖

"Feels like a lot" — make it visible. Each pass now logs `adjudication pass entries=N
judged=K cached=M` (judged = fresh model calls = cache misses; cached = reused verdicts)
+ a `verdicts{verdict="judged_this_pass"}` gauge. Steady state should be judged≈0; a
sustained nonzero means verdict-fingerprint churn driving model load.

Also fix a stale comment: judging is ONCE PER ENTRY (holistic over all objectives it
reaches), not "once per path" — the comment predated the per-entry refactor.

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01VtjoJttCvBY4dzCoE4f9vP
@thejefflarson thejefflarson merged commit efc81bb into main Jun 21, 2026
2 checks passed
@thejefflarson thejefflarson deleted the feat/jef-log-judgement-rate branch June 21, 2026 23:44
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant