[josjeon]

This temporary diagnostic compares the original PostgreSQL control lookup with shared Redis on a dedicated owned load-test stack. It caches only frozen control definitions; authorization runs on every request, and Redis misses, timeouts, corrupt values and failures fall back to PostgreSQL.

The feature is disabled by default and requires an exact fixture allowlist, digest, per-pair generation and absolute expiry. Configuration writes are frozen for the diagnostic. Metrics separate lookup stages, Redis outcomes and loader SQL counts. This is not production caching or write-event invalidation, and must not be merged as a production rollout.

The branch deliberately starts from deployed Agent Control 5084b5a to keep the OFF/ON image comparison bounded. It also includes an annotation-only HttpSpan union correction needed by the baseline Galileo evaluator type check, with no changed evaluator behavior.

Validation: full make check passed, including 1,016 server tests, remaining module suites, Ruff and strict mypy. One existing optional SDK import test skipped. Real local Redis 7.4.2 with redis-py 6.4.0 passed cross-instance reuse, generation isolation and TTL expiry. Cluster connectivity, deployment and load-test results remain pending.

Orbit launcher/harness integration and related PR links will follow. No Jira ticket was supplied.

[codecov]

Codecov Report

❌ Patch coverage is 92.40000% with 19 lines in your changes missing coverage. Please review.

Files with missing lines	Patch %	Lines
...rver/src/agent_control_server/diagnostic_lookup.py	90.07%	13 Missing ⚠️
...r/src/agent_control_server/endpoints/evaluation.py	90.16%	6 Missing ⚠️

📢 Thoughts on this report? Let us know!