feat(judge): judge Actor - #8
Open
RobertCrupa wants to merge 3 commits into
Open
RobertCrupa wants to merge 3 commits into
RobertCrupa wants to merge 3 commits into
Conversation
RobertCrupa
added this pull request to stack #12
September 21, 2026 13:26
Reads the evidence artifact and full session log, asks the model for the rubric, merges deterministic checks and model output into one verdict, and writes scores, a trace comment and an evaluator observation. Builds on the trace contract from feat/eval-trace-contract.
The check stripped only the literal `mcp__apify__` prefix, so a tool call carrying any other server prefix missed the snapshot lookup and was silently skipped. With only such calls in the conversation `checked` fell to 0 and `check.schemaValidity` degraded to `not_applicable` instead of catching a bad input. Use the generic `^mcp__[^_]+__` pattern already used by `renderFacts` and `contract/src/checks.ts`, which operate on the same tool-call data.
The core.ts header, the README scope notes and the Console-visible input_schema description all still called full-log fetching "future work", contradicting both the shipped code (conversationFromFullLog runs on every item) and the README's own "What the model sees" section. The degraded-mode section also claimed schema validity is unconditionally not_applicable. The code only forces that when the trace is degraded AND has no evidence artifact; fetchEvidence keys off meta.evidenceUrl, independent of span validity, so a degraded trace with a reachable artifact runs the real check. Document the actual condition and the legacy-checks fallback.
RobertCrupa
force-pushed
the
feat/judge-actor
branch
from
September 21, 2026 13:28
b45bdea to
3fa30d0
Compare
artogahr
approved these changes
Sep 21, 2026
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
3 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Adds the Judge Actor (`actors/judge/`): reads the evidence artifact and full session log, asks the model for the rubric, merges deterministic checks and model output into one verdict, and writes scores, a trace comment and an evaluator observation.
Depends on the trace contract from #7.
🤖 Generated with Claude Code