Description
Verified live: three response ids taken from a current codex-personal session all resolved into the same-hour codex-lb conversation archive. The join adds engine, proxy, cost and account context that the transcript cannot carry.
It is optional because codex-lb is being retired, because model_provider already varies between codex-lb and openai within one namespace, and because the archive keeps only recent hours while transcripts go back 90 days hot and forever cold.
Acceptance Criteria
- #1 The enricher is off by default and every signal is useful with it disabled; a test asserts no field is required from it
- #2 Lookup is by response_id against the gzipped archive with a bounded cache
- #3 enrich_lookups records cache_hit, hit, miss, error and disabled
- #4 A miss is normal and is never an error; a session whose model_provider is not codex-lb is not looked up at all
- #5 Enrichment fields are record and span attributes, never metric dimensions
Definition of Done
- #1 just check passes: fmt-check, lint, build, test-short, probe-ci, dashboard-check and docs-links all clean