fix(validator): pick issue-bonus tier from live mirror maintainers, not stale per-issue snapshot

[plind-junior]

Summary

_calculate_issue_multiplier decides whether a linked issue grants the maintainer tier (maintainer_issue_multiplier, default 1.66) or the standard tier (standard_issue_multiplier, default 1.33) by reading li.author_association off the mirror's stored copy of the linked issue.

That field is snapshotted at mirror-ingest time and never refreshed. If an issue's author becomes a maintainer (or stops being one) after the issue was filed, every PR that links to that issue keeps using the stored snapshot's role — even though the mirror's /repos/:repo/maintainers endpoint already knows the current role and the maintainer_cut carve-out already uses it.

Concrete miss on phase-rs/phase: matthewevans (id 1388610) and plind-junior (id 59729252) are both currently in the mirror's maintainer list (MEMBER / COLLABORATOR). But all of their stored historical PR/issue rows have author_association = CONTRIBUTOR — they were private collaborators when the rows were ingested. So any solver PR linking to one of their issues gets the standard 1.33× tier instead of the maintainer 1.66×, indefinitely. Only a brand-new issue ingested after the role flip would have the right tier — and even then only for new linkers.

Fix

Route _calculate_issue_multiplier's tier check through the mirror's /repos/:repo/maintainers endpoint — the same source build_maintainer_uids_by_repo already trusts for the carve-out. Identity treatment is now unified across both bonus channels.

# before
is_maintainer = issue.author_association in MAINTAINER_ASSOCIATIONS
# after
is_maintainer = bool(issue.author_github_id and issue.author_github_id in maintainer_github_ids)

maintainer_github_ids is built once per scoring round by a new MirrorClient.get_maintainer_github_ids(repo) -> frozenset[str] that wraps get_repo_maintainers with a per-instance cache. First PR for a repo in a round hits the mirror; subsequent PRs in the same repo are O(1) dict lookups. On MirrorRequestError the method returns an empty frozenset (conservative — the tier silently defaults to standard rather than upgrading on stale data).

Related Issues

N/A. The bug surfaces in production scoring; see the description above for the matthewevans/plind-junior reproduction on phase-rs/phase. No prior issue or PR targets this specific stale-snapshot path.

Type of Change

• Bug fix
• New feature
• Refactor
• Documentation
• Other

What this does NOT change

• The author_association field is still stored on linked issues and still written to the DB. Other consumers (analytics, audit) are unaffected.
• The PR-author drop gate at _should_skip_merged_mirror_pr still uses MAINTAINER_ASSOCIATIONS on the PR itself (not on a linked issue), and is intentionally left as-is.
• The mirror's underlying staleness is not fixed here — that's a separate problem on entrius/das-github-mirror. This PR makes the validator resilient to that staleness for the issue-bonus tier, the same way the maintainer carve-out is already resilient.

Testing

• uv run pytest — 945 passed.
• uv run pyright — 0 errors, 0 warnings.
• uv run ruff check / ruff format --check — clean.
• uv run vulture — clean.

New tests:

• TestGetMaintainerGithubIds (6 tests): cache hits, per-repo isolation, mirror-failure fallback to empty frozenset, failure result also cached, drops entries with empty github_id.
• test_stale_contributor_association_still_gets_maintainer_tier: a stored CONTRIBUTOR snapshot whose author is currently in the maintainer set now correctly returns the maintainer multiplier.
• test_stale_owner_association_drops_to_standard_when_not_currently_maintainer: a stored OWNER snapshot whose author has since departed correctly drops to the standard multiplier.

Existing _calculate_issue_multiplier / _calculate_pr_multipliers tests were updated to pass the maintainer set explicitly; the legacy preference-when-multiple-valid test (legacy parity for #673) still asserts the same behavior, just sourcing maintainer identity from the set rather than the stored field.

Checklist

• Code follows project style guidelines
• Self-review completed
• Changes are documented (if applicable)

[anderdc]

Closing — this is now fixed at the mirror layer instead.

Your diagnosis is correct: li.author_association was a stale ingest snapshot, so a PR linking to an issue whose author became a maintainer after filing kept the standard tier indefinitely.

das-github-mirror#192 (deployed) resolves the linked-issue author's association at serve time against a live maintainers table — the pr_linked_issues view now emits COALESCE(maintainers.association, stored) for issue_author_association. So the li.author_association the validator already reads is the current role, with no validator change needed. The concrete matthewevans (1388610) / plind-junior (59729252) CONTRIBUTOR → MEMBER/COLLABORATOR case you flagged now resolves to the maintainer tier — verified live on phase-rs/phase.

Closing.