Add review map and cross-file cursor foundation

.changeset/bright-review-maps-guide.md

@@ -0,0 +1,5 @@
+---
+'neondeck': patch
+---
+
+Show per-file PR review status in the changed-file tree and add a shared cross-file cursor foundation for files, hunks, threads, drafts, findings, and combined attention items.

.plans/DEVIATIONS.md

@@ -15,6 +15,13 @@ Use this format:
 - Follow-up: What remains, who/what should handle it, or `None`.
 ```
 
+## 2026-07-18 - Diff Review Performance Reconciliation
+
+- Roadmap item: Diff Improvements Plan / transition from Phase A to Phase B
+- Decision: Mark the specialized PR review performance plan complete for now and unpause Phase B while explicitly deferring the production tree median (642 ms versus <500 ms), one-time cold local object fetch (4,978 ms versus <3,000 ms), and uncached review-thread surface/read latency (1,511 ms surface and 655 ms initial GitHub-backed read versus the warm path's <500 ms target). Retain the specialized plan in place instead of archiving it while those measured misses remain.
+- Reason: Stable query identity, bounded immutable metadata reuse, active-patch prioritization, and bounded warm thread reuse delivered passing warm first-patch and thread medians and removed duplicate/abandoned work. The remaining misses are measured, isolated follow-ups that do not require overlapping changes in the Phase B review-map/cursor seam, but they must not be represented as passing or erased.
+- Follow-up: Reprofile the production tree boot/query/render boundary; separate cold network object-fetch time from local metadata before changing refspecs or the <3-second budget; and evaluate uncached GitHub thread latency without weakening cancellation or mutation invalidation. Remeasure the retained immutable real PR before changing any budget, then archive `.plans/PR_REVIEW_PERF_PLAN.md` only after these deferrals are reconciled.
+
 ## 2026-07-17 - Diff Review Phase A Sequencing
 
 - Roadmap item: Diff Improvements Plan / transition from Phase A to Phase B

.plans/DIFF_IMPROVEMENTS_PLAN.md

@@ -1,10 +1,10 @@
 # Diff Improvements Plan
 
-Status: active; Phase A implementation is complete in PR #143, Phase B is paused, and the specialized PR review performance workstream has resumed
+Status: active; Phase A implementation is complete in PR #143, Phase B is active, and the specialized PR review performance workstream is complete for now with measured misses explicitly deferred
 
-Progress note (2026-07-18): the specialized large-PR work now has real registered-PR measurements, stable review-thread identity, bounded local metadata reuse, active-patch priority, and a passing first-patch browser budget. Reconciliation after Phase A found that production tree visibility, review-thread visibility, and the one-time cold-object fetch still miss their retained budgets in `.plans/PR_REVIEW_PERF_PLAN.md`.
+Progress note (2026-07-18): the specialized large-PR work now has real registered-PR measurements, stable review-thread identity, bounded local metadata reuse, active-patch priority, and passing first-patch and warm review-thread browser budgets. The workstream is complete for now. Production tree visibility, the one-time cold-object fetch, and uncached review-thread latency still miss their retained budgets and remain explicit future follow-ups in `.plans/PR_REVIEW_PERF_PLAN.md`; those misses have not been reclassified as passes.
 
-Sequencing correction (2026-07-17): retain the completed Phase A foundation and PR #143, but do not advance into Phase B yet. Phase A was selected while the specialized performance plan still had partial acceptance; that ordering change was not an implicit deferral of the remaining measured misses. Resume the real registered-PR performance work first, beginning with review-thread latency, then reconcile tree visibility and the cold-fetch decision. Phase B remains paused until those items pass or are explicitly deferred with recorded rationale.
+Historical sequencing correction (2026-07-17): retain the completed Phase A foundation and PR #143, but do not advance into Phase B yet. Phase A was selected while the specialized performance plan still had partial acceptance; that ordering change was not an implicit deferral of the remaining measured misses. Resume the real registered-PR performance work first, beginning with review-thread latency, then reconcile tree visibility and the cold-fetch decision. This pause was lifted on 2026-07-18 after the remaining misses were explicitly deferred with recorded rationale; the original correction remains here for audit history.
 
 Contract note (2026-07-18): `shared/review-source.ts` now defines the versioned source, revision, repository, capability, ordered-file, and explicit patch-state vocabulary used by every current web diff surface. GitHub PRs use head SHA identity; prepared and Kilo/repo worktree views receive content-addressed changed-path fingerprints from metadata reads; skill patches and historical repo-edit events use retained content hashes. Missing identities remain explicitly unavailable rather than falling back to timestamps. The current viewers expose source/revision metadata on their mounted roots, ready for the Phase A registration and navigation event layer. On a synthetic 305-file changed worktree, metadata plus revision identity measured 335.5 ms median versus 243.0 ms for metadata alone (92.5 ms added), within the 500 ms warm-tree budget.
 
@@ -14,7 +14,7 @@ Fixture note (2026-07-18): `npm run bench:review-fixtures` now builds determinis
 
 Related plans:
 
-- `.plans/PR_REVIEW_PERF_PLAN.md` — active large-PR data-path and performance workstream
+- `.plans/PR_REVIEW_PERF_PLAN.md` — complete-for-now large-PR data-path and performance workstream with explicit deferred misses
 - `.plans/OTHER_PEOPLE_PR_REVIEW.md` — current human PR review workflow
 - `.plans/archived/DIFF_UI_PLAN.md` — landed Pierre diff/tree adoption
 - `.plans/archived/DIFF_REVIEW.md` — earlier diff review research and interaction planning
@@ -545,7 +545,7 @@ before relaxing a gate.
 
 ### Phase B — Guided review
 
-1. Add review-map decorations and pure cross-file cursors.
+1. **Completed —** Add review-map decorations and pure cross-file cursors.
 2. Add visible navigation controls, scoped shortcuts, and help.
 3. Add typed Neon finding application and inline rendering with provenance.
 4. Add explicit promote-to-draft and promote-to-revision flows.
@@ -637,8 +637,9 @@ before relaxing a gate.
 - Add a changeset for each user-facing implementation phase.
 - Record deviations or deferrals in `.plans/DEVIATIONS.md` when implementation changes this priority
   order, trust boundary, performance gate, or surface coverage.
-- Keep `.plans/PR_REVIEW_PERF_PLAN.md` as the specialized performance implementation plan until that
-  work lands; then archive it and preserve this document as the broader diff product roadmap.
+- Retain `.plans/PR_REVIEW_PERF_PLAN.md` as the specialized performance implementation record while
+  its measured tree, cold-fetch, and uncached-thread misses remain deferred; archive it only after
+  those follow-ups are reconciled, and preserve this document as the broader diff product roadmap.
 
 ## Definition of Done
 

.plans/PR_REVIEW_PERF_PLAN.md

@@ -1,8 +1,24 @@
 # PR Review / File Tree Performance Plan
 
-Status: phases 1–5 implemented; real-PR verification, request-path remediation, active-patch prioritization, and warm review-thread remediation complete; tree and cold-fetch decisions remain
+Status: complete for now; phases 1–5, real-PR verification, request-path remediation, active-patch prioritization, and warm review-thread remediation are complete; the missed tree, cold-object-fetch, and cold-thread budgets are explicitly deferred below
 Prior art: `.plans/archived/DIFF_UI_PLAN.md`, `.plans/archived/DIFF_REVIEW.md`
 
+## 2026-07-18 completion-for-now decision
+
+This specialized workstream is complete for now, not complete against every
+retained budget. The production tree median remains 642 ms against the <500 ms
+target. The one-time cold local object fetch remains 4,978 ms against the
+<3,000 ms target. The uncached lean review-thread surface remained 1,511 ms,
+and its initial GitHub-backed read remained 655 ms, outside the warm UI path
+that reached a passing 459 ms median through bounded 15-second reuse. These
+misses are deferred future performance work; none is reclassified as passing.
+
+The implemented data path, immutable measurements, and completed remediation
+items remain active foundations for the broader diff plan. Phase B can proceed
+because the remaining misses are now explicit, measured deferrals rather than
+unreconciled acceptance gaps. Future work should remeasure the same immutable
+target before changing implementation or relaxing a retained budget.
+
 ## 2026-07-18 shared Phase A fixture baseline
 
 The broader diff-improvements Phase A now has deterministic 8-file, 90-file,
@@ -351,16 +367,26 @@ resources. The cold first read remains explicitly outside the warm-cache pass.
    now uses an 84.7% smaller query response plus a bounded 15-second cache with
    mutation invalidation and race protection. Production thread visibility is
    459 ms median, while the full-fidelity Flue action is unchanged.
-5. **Discuss later — revisit cold fetch.** The 4.98-second object fetch misses the target,
+5. **Deferred — revisit cold fetch.** The 4.98-second object fetch misses the target,
    but it is a one-time revision cost. Separate network fetch time from local
    metadata time before changing refspecs or the `<3s` budget.
-
-Acceptance is partial on the same real target: duplicate thread requests and
+6. **Deferred — revisit production tree visibility.** The final 642 ms median
+   remains 142 ms over the `<500 ms` budget even though the warm backend
+   diagnostic passes. Profile the production boot/query/render boundary before
+   changing Pierre or the tree budget.
+7. **Deferred — revisit uncached review-thread latency.** Bounded warm reuse
+   brings the median to 459 ms, but the uncached lean surface remained 1,511 ms
+   and the initial GitHub-backed read remained 655 ms. Preserve mutation
+   invalidation and cancellation while evaluating any more durable reuse or
+   GitHub query-path change.
+
+Acceptance remains partial on the same real target: duplicate thread requests and
 settlement-driven abandoned patch reads are eliminated, the first-patch browser
 budget now passes, warm thread visibility passes on the median, backend targets
 pass, and fallback code is unchanged. The tree and one-time cold-object budgets
-still miss and remain separate follow-ups; a cold GitHub thread read also
-remains slower than the warm UI budget. Raw baseline, remediation, and
+still miss and are explicitly deferred; a cold GitHub thread read also remains
+slower than the warm UI budget and is explicitly deferred. The workstream is
+complete for now on that basis. Raw baseline, remediation, and
 active-priority results are
 gitignored at
 `benchmarks/results/pr-12204-real-local.json` and