[transitrix]

What changed

Release drift compared string-valued counts, so 8 commits could produce a reminder claiming to exceed a threshold of 50. Parse and validate numeric inputs before comparing, and reconcile the title, body and state of workflow-owned reminders. Below-threshold closure explicitly does not claim a release was prepared or published. Preserve manually maintained release issues and refuse ambiguous duplicate reminders.

PRs run the 22 reminder regression tests with read-only permissions; issue writes remain limited to scheduled/manual reconciliation after those tests pass. Declare the scripts in the existing publish manifest. No release is published.

Closes #639.

How it is verified

• 22 reminder tests pass, covering 8/27/49/50/51/100 counts, stale reminders, idempotence, unrelated issues, malformed inputs, pagination and denied API access.
• All 5 publish-manifest tests pass; the manifest and workflow-script checks pass.
• Notation documentation checks pass with four existing advisories; git diff --check passes.

Checklist

• Based on main, one concern per PR.
• Commits signed off for DCO.
• Public files contain only repository maintenance code and necessary tests.

[transitrix]

Independent read-only source review completed by a separate reviewer run at exact head 7431fcfdd191697acfe7bcbf85ff82a397003e19. No blocking findings in the full four-file diff.

The reviewer independently executed all 22 reminder regression tests, all 5 publish-manifest tests and workflow-script integrity. Verified numeric threshold boundaries, stale reminder closure without release claims, unchanged rerun behavior, exclusion of manually maintained issues, duplicate/malformed-input refusal and preserved 403 failures. PR execution has read-only permissions; the issue-write job is excluded on PRs and depends on passing tests. No release action was introduced.

This records independent source review; exact-head CI and supported merge remain separate checks.