fix(review): deterministic linked-issue gate never checks issue-PR scope match, only existence/openness

[JSONbored]

Context

Repro: JSONbored/metagraphed PR #3910 (real-venus, "feat(registry): add SN74 Gittensor per-repo commits subnet-api surface (#1275)"). Issue #1275 asks to enrich SN74 Gittensor with an sse surface ("Enrich SN74 Gittensor — add SSE stream"). The PR instead adds a subnet-api REST surface for a different endpoint. The AI reviewer correctly wrote, in the comment's prose "Blockers" section: "The PR claims to close #1275, but the linked issue is scoped to adding an SSE stream for SN74 Gittensor, not a subnet-api REST endpoint — the added surface does not satisfy that issue's scope."

Yet the same posted comment's structured "Linked issue" signal row read ✅ Linked · #1275, Gate result: ✅ Passing · No configured blocker found, and the top-line disposition was approve/merge recommended.

Root cause

addPullRequestFindings (src/rules/advisory.ts:697-698) derives the missing_linked_issue finding purely from existence/openness:

const noLinkedIssueCited = pr.linkedIssues.length === 0;
if ((noLinkedIssueCited || confirmedNoOpenLinkedIssue) && requireLinkedIssue) { ... }

There is no check anywhere in this deterministic path for whether the linked issue's scope actually matches the PR's delivered content. A PR that cites a real, open, but unrelated issue passes this exactly as if it cited a perfectly matching one.

This is a distinct gap from three already-closed issues in the same area — filing this fresh rather than reopening any of them:

• fix(review): disposition recommends approve/merge while blockers are active (+ approval-bar / confidence level may be too loose) #2002 / fix(review): decide whether AI-judgment blockers should gate the verdict, not just annotate it #2592 (closed, intentional-by-design): whether an AI-judgment blocker should ever override a gate decision: merge. Resolved as "the gate decision is authoritative, it already weighed advisory blockers" — see the documented rationale at src/review/unified-comment.ts:349-350.
• fix(review): reconcile AI verdicts against hard structured state before rendering #2996 (closed, completed): added a reconciliation pass cross-checking AI verdicts against gate blockers, duplicate/linked-PR state, and linked-issue open/closed/unlinked state. It deliberately doesn't cover a scope-match dimension, because the deterministic classifier has never computed one.
• [Bug]: gate predictor under-reports the missing-linked-issue blocker on mergeReadiness:block repos #1446 (closed, not-planned): the MCP pre-submission predictor (a different code path) under-reporting missing_linked_issue on mergeReadiness:block repos — about existence detection, not scope.

None of those cover "the linked issue exists, is open, but doesn't match the PR's content" — exactly the case metagraphed's own contributor policy calls out (a linked issue must genuinely match the PR, not just exist/be open), but which the deterministic gate today has no way to detect.

Requirements

• Add a scope-match dimension to linked-issue evaluation, gated per-repo via .gittensory.yml (e.g. gate.linkedIssue: "scope-strict"), default unchanged/off everywhere that doesn't opt in.
• When enabled, and the AI reviewer has flagged (with sufficient confidence, i.e. it placed the finding under "Blockers" not "Nits") that the cited issue's scope doesn't match the PR's delivered content, fire a linked_issue_scope_mismatch finding, demote the "Linked issue" signal row off ✅ Linked, and prevent decision: merge.
• Do not change default behavior for gittensory or any repo that hasn't opted in — this is additive/config-gated, not a revisit of the fix(review): decide whether AI-judgment blockers should gate the verdict, not just annotate it #2592 universal-AI-blocker decision.
• Regression test reproducing the fix(review): preserve gate verdict on auto-review skip #3910 shape (an open, cited issue whose scope doesn't match the PR body/diff): confirm the new signal fires when a repo opts in, and stays inert when it doesn't.

Deliverables

• New linked_issue_scope_mismatch finding code, config-gated per-repo
• .gittensory.yml schema + resolver + openapi coverage for the new gate knob (same PR, per the existing gate-setting-wiring template)
• Regression test reproducing metagraphed PR fix(review): preserve gate verdict on auto-review skip #3910's shape
• Docs: CONTRIBUTING/skill note that scope-strict repos require the linked issue to genuinely match, not just exist

Expected outcomes

• A repo like metagraphed can require genuine issue-scope match, not just existence, to satisfy the linked-issue gate — closing the loophole PR fix(review): preserve gate verdict on auto-review skip #3910 exposed.
• Zero behavior change for repos that don't opt in.

[linear-code]

JSO-2866

[JSONbored]

Companion issue: #3907 (same PR #3910 repro, different gap — that one is about the general AI-judgment-blocker-never-gates design; this one is specifically about the linked-issue check never verifying scope).

[RealDiligent]

Contributor fix ready on RealDiligent/gittensory branch \ix/critical-issue-mcp-linked-issue-extractor-parity\ (was #4054, auto-closed for missing tracking issue): MCP + predicted-gate \extractLinkedIssues\ still lack the #4039 inline-code-span strip, so the PR template (e.g. \\Closes #123\)\ line spuriously links #123 in local branch analysis. Fork tracking: RealDiligent#4. Could a maintainer please open a contributor-eligible \gittensor:bug\ upstream issue so we can reopen with \Closes?