Skip to content

feat(evaluation): implement P-04 Room result repository - #245

Closed
daniele21 wants to merge 16 commits into
devfrom
agent/eval-p04-room-repository
Closed

daniele21 wants to merge 16 commits into
devfrom
agent/eval-p04-room-repository

Conversation

@daniele21

@daniele21 daniele21 commented Aug 15, 2026

Copy link
Copy Markdown
Owner

Superseded for integration by #260. The durable P-04 Room repository is being carried by the clean replay on the newer dev baseline. This PR remains only as historical implementation/validation context and must not receive further commits or be merged.

Original scope:
Implements EVAL-P-04 against the integrated P-03 privacy-safe schema and P-02 repository semantics.

  • adds a dedicated Evaluation Room database/DAO; telemetry database ownership remains untouched;
  • adds Room compiler/schema export wiring without destructive fallback;
  • implements EvaluationResultRepository create/update/case append/get/query/delete/retention through Room;
  • preserves immutable run config, lifecycle transition validation, sample-set case bounds, deterministic history order, active-run delete protection and terminal-only retention from the P-02 reference;
  • uses dynamic bounded SQL for filtered history rather than filtering after LIMIT;
  • maps the normalized entity graph back to domain contracts while recomputing and verifying sample-set, semantic-execution and run fingerprints so inconsistent stored identity fails closed;
  • keeps prompt, expected-answer and generated-answer content out of the database;
  • adds quality_present and category ordinal fields before Evaluation database v1 exists so optional-quality semantics and category ordering survive exact round trips;
  • adds mapper round-trip coverage including present-but-empty quality and privacy-safe case outcome fields.

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