feat(eval): dependency-aware eval ordering with DAG wave scheduler#1051
Merged
feat(eval): dependency-aware eval ordering with DAG wave scheduler#1051
Conversation
#331) Add `depends_on` and `on_dependency_failure` fields to EvalTest, enabling multi-agent swarm evaluation with dependency ordering between tests. The flat pLimit + Promise.allSettled dispatch is replaced with a DAG-aware wave scheduler that: - Validates the dependency graph (rejects cycles, missing IDs, self-deps) - Computes execution waves via topological sort - Dispatches independent tests in parallel within each wave - Tracks completed results for downstream context injection - Supports three failure policies: skip (default), fail, run - Injects `dependency_results` into evaluator context for dependent tests Tests without `depends_on` behave identically to before (single wave). Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
- Only treat execution_error (not quality_failure) as dependency failure A dependency test scoring 0.2 now correctly allows downstream to run. Only actual crashes/errors trigger skip/fail policies. - Add duplicate test ID validation in validateDependencyGraph - Add defensive assertion in computeWaves for unscheduled tests - Deduplicate skip/fail result construction into single branch - Add tests: transitive cascade (A->B->C), quality_failure distinction Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Deploying agentv with
|
| Latest commit: |
04686dc
|
| Status: | ✅ Deploy successful! |
| Preview URL: | https://4ea079b0.agentv.pages.dev |
| Branch Preview URL: | https://feat-331-dependency-aware-sc.agentv.pages.dev |
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
1 participant
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.
Summary
Closes #331
depends_onandon_dependency_failurefields toEvalTestfor declaring inter-test dependenciespLimit+Promise.allSettleddispatch with a DAG-aware wave scheduler that topologically sorts tests and dispatches independent tests in parallel per wavedependency_resultsinto evaluator context so downstream assertions can access prior test outputsskip(default),fail,runSchema additions
Test plan
depends_onbehave identically (no regression) - 1492 core tests passon_dependency_failure: skipskips downstreamon_dependency_failure: failmarks downstream as failedon_dependency_failure: runexecutes downstream regardlessdependency_resultsavailable in evaluator contextE2E red/green evidence
Red (main): All 4 tests dispatch simultaneously - integration and consistency-check run without waiting for dependencies. depends_on is completely ignored.
Green (feature branch): Wave scheduler respects dependencies. Wave 1 runs backend-api and frontend-ui in parallel. Wave 2 waits for wave 1 to complete, then applies failure policies: integration (skip policy) is skipped, consistency-check (fail policy) is marked failed. Error reason code is dependency_failed.
Generated with Claude Code