Search verified records
Candidate or malformed records are not eligible. If relevance is too low, AEG abstains.
60-second product moment
Retrieve verified experience before the next agent starts from scratch.
This lightweight simulation mirrors the repository’s verified-only retrieval workflow. It demonstrates the interaction—not a performance benefit.
Ready to search verified recordsAEG may match, rank, or abstain.
VERIFIED MATCH
objective repair
TR-04 · thin proxy delegation
WHY THIS MATCHED
Overlap in wrapper delegation, stale resource ownership,
protocol layer, and set_nodelay.
RECOVERY PRINCIPLE
Trace current resource ownership through the protocol abstraction; repair the public wrapper rather than a stale client-side object.
FAILURE WARNING
A green patch can be a false positive when it fixes the wrong public API surface.
LIMITATION
This prior collection run is not causal evidence that retrieval improves repairs. Inspect local code and run the full relevant test suite.
What just happened
Evidence traveled with the recommendation.
The match is not an answer guarantee. It packages a prior verified outcome, explainable retrieval signal, failed path, recovery principle, constraints, and provenance.
See why it matched
Query terms, contributing fields, outcome, limitations, and public provenance stay visible.
Treat guidance as a hypothesis
The next agent inspects local code, reproduces the failure, and validates focused plus regression tests.
Evidence boundary
The workflow works. Its benefit is still under test.
The prior comparison for this synthetic challenge found the same successful repair path and identical patch in baseline and assisted arms. The assisted arm used 13,842 more non-cached tokens and took 4,288 ms longer.
That is useful negative evidence: this demo proves discoverability and auditability, not improved correctness or efficiency.
Try the real surface