[trissim]

Summary

3 commits, 5 files, +284/-36. Current head 6e5c16f720ecccc5e072ea7710648ba631a952c7, rebased onto main 03f32f270ec4ba52dc9ba559b43c4246fd5f945a. Four production owners and presentation-fencing controls are separated from #1393 so the usage repair can ship without the physical-run harness. The first commit mechanically preserved the paint blobs; the second repairs the acquisition/publication lifetime; the third corrects test ingress and establishes actual saved paint before freezing its source. The two reported serial failures remain unverified at this new head. No harness/runtime-package change is included.

What changes for the user

Following saved transcript output extends the already-painted prefix instead of replacing it. A known undrawn destination must actually publish before live output can retire. Failed append acquisition forces replacement publication, including when a rewritten file reuses its path and grows past the old byte offset. A held reader keeps its paint on a successful read. Saved-history height is not live/native prune pressure.

Owner and consumer change

Before:

• FollowTailCommit.extend acquired a replacement and republished preceding pages.
• Access-only success could retire live output before the known destination painted.
• Cursor-after acquisition gave ClosedHistory an incomplete replacement.
• Conversation prune pressure counted bound saved history as live/native output.
• The first version caught TranscriptChanged inside PreparedAppendCommit, read a complete snapshot, then independently selected append publication again. Filename/offset containment cannot recover that discarded acquisition failure.

After:

• PreparedAppendCommit inherits PreparedPageCommit's capture/source-cut/current/draw evidence and owns cursor-after acquisition. PreparedPageCommit owns complete acquisition/replacement; PreparedCommit owns capture without replacement paint.
• Each source's declared commit_preparation supplies acquisition capability. CommitPlan captures the corresponding publication member before that read. Its existing viewport family chooses retaining versus following behavior; a TranscriptChanged fallback fixes that same read's member to PreparedPageCommit. Later capture consumes the retained member rather than reselecting from the source. No read-mode flag, wrapper family, controller or type roster.
• Shared PreparedCommit.capture replaces the two viewport-specific extend implementations. Existing cursor containment still selects replacement for a changed filename or rewind; the retained failure decision additionally covers a same-path rewrite whose new cursor contains the old position.
• TranscriptSourcePreparation.publish_append shares bounded extension and requires current publication, no known gap and the actual requested endpoint before destructive retirement. Weaker access-only consumers retain their original contract.
• Live pressure derives from actual layout minus bound saved-history height; no mirrored counter.

The source-cut/current/ready/lock/commit/retirement guards remain. The genuine Textual self-wait correction landed separately in #1394 and is not duplicated.

Review53818: complete acquisition and held readers

ClosedHistory genuinely reaches CommitPublication, including while the viewport is held. Complete acquisition is deliberate: coverage for a retained-only prefix cannot depend on an absent saved reader. A successful held-viewport commit still captures PreparedCommit and keeps its original pages, fragments and reader position; complete acquisition does not mean replacement paint. The existing closed-source case now covers both viewport intents.

ProjectedTranscriptSource is different: the only production constructor is LineHistory.projected_source (widgets/transcript_history.py:862), consumed by SeekingFilter.advance (transcript_filter.py:197) to mount ProjectedTranscriptHistory as the filter overlay. It is not the BoundView's ConversationHistory source selected by CommitPublication. This is the bounded named-consumer trace, not a claim that arbitrary dynamic callers were proved absent. The default complete-read method is not a new projected-overlay commit path.

CommitPlan's stale viewport-only acquisition docstring is corrected in the same commit.

Source evidence

Existing NRA SourceModule/ModuleSyntaxIndex/CompactClassFamilyIndex map: 35modules,354original classes,353canonical classes,1unprojected OPEN,0parse omissions before and after; syntax sites17107→17173. All354 declaration/lexical-owner/base/decorator rows remain equal in traversal order. No classes or inheritance edges added. Raw declarations/writes/decisions/imports/consumers are retained in platform-source/paint-review-family-{before,after}.json in the owned Core checkout. Direct-name retrieval spans both repositories' src/tests/tools, with named Core/Textual dependencies. Aliases, dynamic binding, effects and arbitrary movement remain OPEN, not certified by this map.

Two changed files compile; three actual Toad modules import from this source using installed0402 Core in the read-only0402/3475 dependency prefix. Actual MRO/import origins and the declaration comparison passed rc0 in pi-ui-performance/paint-review-qualify-02.log. No local NRA gate, pytest or physical run is represented by this source check.

Executed app proof and outstanding counterexamples

The original five affected app cases collected5/passed5,71.91s,rc0 at26a application/fixture + installedceeed Core/Textual. Artifacts: /home/ts/.cache/agent-scratch/pi-ui-performance/dm-app-controls-03/{pytest.log,pytest.rc,origins.log}. They covered painted-page/fragment retention, the actually admitted known-gap destination, complete closed-source prefix and both live-pressure variants. Earlier setup5fail and4pass/1fail remain failed. No production method patch, guard weakening or restored Thread.project/Session details API.

Review54167 executed bf87092 against current-main dependencies: Claude reports 12 passes, two stale TranscriptChanged imports, two real serial failures, and six contention-only failures that pass serially. The serial failures were the held closed-prefix marker disappearing and a followed commit returning with drawn_end_more still True.

Latest test-only correction:

• Import TranscriptChanged from its owning coordination_errors module.
• Before stopping observations/suspending the source, require the saved history itself to paint the marker with source_ready and reports_coverage. A live replay showing identical text is not this precondition. The shared helper serves append, gap, closed-prefix and rewrite cases.
• Keep the actual captured gap publication for failure diagnostics: current publication/viewport, source readiness, follow intent, dirty/commit flags and actual cursors. The original gap and paint assertions remain intact.

No production guards or family declarations changed in this checkpoint. The file compiles; the latest setup correction is UNRUN and is not claimed to fix either serial failure. Commander owns the necessary combined selection, including the two original failures and both real rewrite/import cases. The original five passes are not current-head acceptance.

Headless app results do not prove configured-native read/edit arrivals, warm M-view paint/selection/draft/undo, hidden/reopen/tab-return, full physical performance, installation or original P867 coordinates. I retain implementation and verification ownership through the combined installation. Commander owns integration/install; auto-merge remains off pending exact-head review.

Separate work

#1393 retains the physical lifetime/observation overhaul and explicit module-entry fix; #3825 retains exact-input runtime/interpreter/archive evidence. The prepared0402/3475 prefix remains unchanged and borrowed read-only. Its preparation/readback and fresh stopped-history copy do not establish this revised paint head's live acceptance.

Assumed: the original five app results retain26a/ceeed scope; Review54167 retains its bf87092 dependency scope; the latest setup correction and two serial failures still need execution; the earlier NRA/import map is not proof of all current-base dynamics; current combined dependencies, installation and configured read/edit remain to be verified.

[Copilot]

Copilot was unable to review this pull request because the user who requested the review has insufficient remaining quota. Try requesting a review at a lower effort level.

An organization owner can choose the repository organization as the billing source for reviews requested by members with a paid Copilot license. Learn about Copilot code review billing.

[trissim]

Closed in Tristan's board clear (2026-11-02): superseded by #1393, which carries the retained followed paint. The branch is kept.

---

Generated by Claude Code