💡 Motivation and Context

Several tests could pass without exercising the behavior they claimed to protect. Some also depended on external HTTP services, scheduler timing, or assertions raised inside worker callbacks that catch exceptions.

This test-only change replaces those HTTP calls with controlled responses and strengthens checks for retry counts, event payloads, exception propagation, truncation, copy isolation, and lifecycle behavior. Consumer tests now wait for delivery and stop their workers. A controlled-clock test checks the remaining flush interval. Registry key checks catch missing or untested OpenAI resources, and placeholder checks retain both explicit expectations and coverage of every registered entry. LangChain error tests still exercise the real callback lifecycle through a mocked HTTP transport.

These are the validated repairs from a partial test audit. This PR does not claim that every repository test has been reviewed. No production code or public API changes are included.

💚 How did you test it?

• Root suite with external socket connections blocked: 4,143 passed, 16 skipped. The same guarded baseline had 4,137 passed and 6 failures involving external HTTP dependencies.
• Runtime branch coverage, excluding tests: combined coverage changed from 88.18059% to 88.18510%. Covered lines increased from 14,622 to 14,623 of 16,234. Covered branches stayed at 4,910 of 5,916. The main benefit is stronger assertions rather than broader execution coverage.
• OpenFeature: 68 passed. Django middleware and exception integration: 7 passed. Compliance adapter: 23 passed.
• Ruff lint and formatting, mypy baseline filtering, the public API checker tests, and import with warnings treated as errors passed.
• Changing the tracing default flush interval to 50 seconds in memory passed the old defaults test and failed the repaired test.
• Autoreview against origin/main reported no actionable findings for 4544aa5.

The full-suite and coverage results above were measured at 107b3fb. After review follow-ups at 4544aa5, 53 focused consumer and OpenAI resource tests, the API checker tests, and Ruff passed. An early-flush mutation passed the previous timing test and failed the new controlled-clock test. Extra and missing wrapper keys were both rejected.

Validation used Python 3.13.13. Live credential-dependent tests, MCP v2, and the full Python-version matrix were not run.

📝 Checklist

• I reviewed the submitted code.
• I added tests to verify the changes.
• I updated the docs if needed.
• No breaking change or entry added to the changelog.

If releasing new changes

• Ran sampo add to generate a changeset file

🤖 Agent context

Autonomy: Human-driven (agent-assisted)

Pi used read-only subagents for test inventories and review, then applied and validated focused test repairs in a dedicated worktree. Tools included Git, pytest, coverage.py, Ruff, mypy, and the autoreview helper. The audit remains incomplete, and unresolved review scope was not treated as approved. Local audit notes are kept outside the PR. No shared session link is available.