[zhongkechen]

Requirement 10-1 intermittently reports the greeting before invocation-start even though the execution succeeds and the SDK calls the start hook synchronously. The hook writes to stdout while the greeting uses Python logging, so Lambda can assign their CloudWatch timestamps through different paths. Examples: failed run and earlier identical failure.

Send the lifecycle hook records through the same root logger as the SDK step logger, preserving the top-level plugin fields and execution ARN. Enable JSON/INFO logging explicitly on this test function. Regression tests cover the initial invocation and a replay that emits the hooks again without repeating the completed greeting step.

The upstream assertions remain unchanged. Conformance PR #143 clarifies that runtime structured logging is permitted with the same top-level JSON fields; merge that wording clarification first.

Validation:

• 13 conformance-package tests and Hatch formatting/lint checks passed after cleanup.
• The unchanged logging fix passed all 11 cloud conformance suites, including 23/23 plugin requirements and a one-off 40/40 lifecycle stress run. Those samples had no reversed log ordering; their raw records remain in that run's conformance-report-plugin artifact.
• Python 3.11, 3.12, 3.13, and 3.14 build jobs passed before the diagnostic tooling was removed. Fresh CI is running on the cleanup commit.

[github-actions]

Codex AI review

No actionable findings. Review was static: tests were not executed, and Lambda/CloudWatch log ordering was not independently verified.

Reviewed commit 71a9fa69ac5cee4477a99956131fc8b6c0d1c1b7. Workflow run

[hln33]

Is this issue present in other SDK conformance test handlers or just Python?

[zhongkechen]

Is this issue present in other SDK conformance test handlers or just Python?

I created this PR because I found a flaky test in Python. I didn't check the other SDKs.