[yaacovcorcos]

Problem

Scient's prior official boundary is d8e9fd267e26b7680f035edc0b40db09d2685dbb. This integrates all 84 official commits through frozen f34ba437a5dec23a205ef5f670f7b10439ae98eb with literal ancestry, without enabling upstream publication or changing Scient product policy.

Result

• Native V2 queue, Stop/Steer, delegated-child and provider updates composed with Scient's sole queue-release and result authorities.
• Current-execution Stop matching excludes historical interrupts. Failed delegated roots wait for surviving native children, including older untagged runs, and settle through the existing parent-locked finalizer.
• Incoming PDF/browser navigation retained. HTML-to-PDF export cannot publish the client-PDF fallback placeholder. Workspace/environment PDF.js readers, attachment previews, and browser tabs remain separate contexts.
• Adopted upstream palette/navigation, model-identity settings merge, large-tree virtualization, browser/terminal/remote/provider fixes, React catalog and native PDF dependencies.
• Standalone Scient image cards and inline assets now share workspace-edit refresh/coalescing. File-preview pixels survive capability refresh but cannot remain pinned across an explicit revision.
• Retained Scient identity, authorization, FULL SQLite durability, immutable migration history (incoming migration becomes 068), analytics allowlist/consent, cloud/mobile/release holds.
• Thin-host composition retains existing Scient modules and narrow mounts. Against the same official snapshot, added non-test host lines fall from 52,115/918 hosts to 50,794/857 hosts at the final source revision. The declared three policy hosts fall from 2,534 to 2,420 lines. These are structural measures, not behavioral proof or total product size.

History

• Owned base: 129f4af07bc3b78cab506086fcad0c5f12d7105f.
• Literal official merge: 9ef3f49f0de3d380389e2e26d0ff9753ef37ff11, exact official second parent.
• Owned-main catch-up: 667f48d7916020076ac97190ef84748566aaf06a of 7d6bd586a47220c44c880077fb64cf47e8fa9a78, carrying already-landed docs(release): add approved Scient 0.6.24 notes #599/test(server): keep server tests out of the source checkout #601/fix(providers): reconnecting no longer cancels a runtime install #604 without modifying those independent lanes.
• Fixture portability follow-up: 0ad0b7dde69723273861d00f8f20aa0d85c3a142.
• Reviewed source follow-up: 479dfd46d17cf22f8a33eadfdf2323b675e7803f.
• Receipt/cursor follow-up: f39e52c9a971646560ed275f21150081ce0b4bce, maintainer documentation only.
• Second owned-main catch-up: a0e936c50f2dfebe362a2fa3707aaecac1399797 of 17f7bfb2f47a4622f627b34d378bc33c1f43eb01, carrying already-landed fix(antigravity): no console window on Windows; say when the runtime check runs out of time #603/ci: let the merge queue reuse a PR run of the exact same content #605/test(web): steady five browser tests that fail on busy CI runners #606/fix(omp): keep a background job's late updates out of later turns #567/test: give two slow tests room on busy CI runners #608/fix: report an agent crash during prompt teardown, and stop false 'changed while exporting' #609/test(mobile): make diff grammar assertions independent of wall time #610. One conflict, in the MessagesTimeline anchor test: kept main's 5 s wait and the alignment's tolerance for repeated measurements. Every other path main changed matches main exactly.
• Queue/analytics test follow-up: d00f9e6d490b668ff73d16cf710b83ecca1eaf5c.
• Catch-up record: 68919d6ec33425f96d12cd93cc2d1995babd5e78, maintainer documentation only (state, UPSTREAM.md, receipt). integrationBase, owned base and lastRefreshMerge are unchanged.
• Upstream push URL remains DISABLED.
• Latest observed Scient main is 17f7bfb2f4, included. Official T3 main has moved beyond the deliberately frozen f34ba437a5; this PR does not claim that newer tip.

Verification

Second catch-up (head 68919d6ec3, local Windows 11):

• Hosted CI at f39e52c9a9 failed five server tests. All were incoming upstream expectations; none came from a catch-up interaction or a flake. Upstream #21473's "any client send releases a held queue" rule appears in SteeringCompletion (plain send after Stop) and ThreadLaunchService (direct send after failed worktree setup). Both are now marked Scient divergences under QueuedRunSend.ts's direct-send rule and noted in UPSTREAM.md. The OrchestratorMcpToolkit held-follow-up check called t3_thread_send instead of scient_thread_send. server.test.ts now expects upstream's feature.used call site, which the Scient analytics contract drops; a new conformance assertion covers that.
• All package typechecks passed (49 tasks, 0 syntax-gate or type errors). vp fmt --check passed on 9,040 files. Diff check, upstream provenance and all five seam checks passed against 17f7bfb2f4.
• Focused server runs: ThreadLaunchService, SteeringCompletion, QueuedRunSend, ThreadStop, QueuedStartRecovery and QueuedRunOrder passed completely. The MCP delegation test and the analytics test passed. Analytics conformance passed 4 tests. Pre-existing Windows-only failures, from fixtures and code unchanged from main: 16/50 NativeQueueHoldPolicy cases (: in temp-dir names), 13/249 server.test.ts cases (symlinks, POSIX MATLAB fixtures, directory fsync), and the Codex transcript fixture path. The MCP delegation test needs more than its 120 s budget on this host.
• Local browser tests could not start Chromium. Hosted CI must qualify the pushed head.

Earlier qualification on 479dfd46d1/f39e52c9a9:
Local source qualification is complete, with the failed aggregate and bounded rerun recorded separately:

• Independent queue/PDF follow-up review: no remaining findings in inspected repairs.
• Focused: 195 queue tests, 366 repaired mixed tests, 83 browser tests, 34 live Mermaid Chromium tests. Latest tagged/untagged settlement + SQLite run: 53 passed; Linux icon fixtures: 17 passed.
• Exact image-source review: no actionable findings. Focused follow-up: 1,892 tests across nine files passed, web types and scoped lint passed.
• Serial gates on 479dfd46d1: formatting, lint, knip, all package typechecks, build, desktop smoke, brand and both diff checks passed.
• The aggregate test run failed only in web: 14,696 passed, three dynamic-import timeouts. The affected files passed unchanged in isolation, 13 tests in 994 ms. The complete web workspace then passed with a synthetic home and pnpm --filter @t3tools/web test --maxWorkers=4: 14,699 tests / 1,087 files / 166.34s, without changing source, assertions or timeouts. Other workspace results are from the same-source aggregate run.
• Five alignment seams, mobile static checks, synthetic release smoke and ancestry/provenance check passed.
• Native Windows/Linux/mobile, real-provider and visual acceptance are not established by these checks.
• Final isolated candidate launched at f39e52c9a9 with worktree-local synthetic state. Managed runner/app/backend ownership and loopback listeners were checked; renderer and backend environment endpoints returned HTTP 200. The app remains available for owner review. Interactive UI verification is blocked by absent built-in Browser tools; no alternate automation was substituted. Startup checks do not establish visual or feature acceptance. Any tested in-app agent turn must use Codex with Luna 6.
• Hosted CI must qualify the final pushed revision separately. Earlier CI at 0ad0b7dde6 is not final-head evidence; its Windows Python 3.10 job failed during runner setup before checkout. No main merge or release is requested.

Documentation impact

Updated: docs/user/pdf-reader.md, docs/user/terminal.md, incoming user/provider guides, UPSTREAM.md, upstream-state.json, docs/internals/scient-fork-divergence.md and the exact alignment receipt.

Review and publication boundary

Draft only. No merge, merge queue, auto-merge, desktop release, publishing, cloud/mobile activation, telemetry-contract widening, or live-profile/credential operations authorized.

Harness: Factory Droid. Model identifier: not exposed in this session. Second owned-main catch-up and CI repairs: Claude Code, Claude Opus 5.5.

🤖 Generated with Claude Code

[github-actions]

Thread transfer impact

✅ Thread transfer remains within every enforced ceiling.

Provider	Metric	Main baseline	This PR	Impact	PR ceiling
Codex	Total thread wire	5.6 KiB	5.6 KiB	0 B (0.0%)	6.8 KiB	✅
Codex	Thread snapshot wire	4.3 KiB	4.3 KiB	0 B (0.0%)	4.9 KiB	✅
Codex	Live turn WebSocket wire	1.3 KiB	1.3 KiB	0 B (0.0%)	2.0 KiB	✅
Codex	Live turn WebSocket decoded	26.2 KiB	26.2 KiB	0 B (0.0%)	29.3 KiB	✅
Codex	Live turn messages	2	2	0 (0.0%)	8	✅
Claude	Total thread wire	5.6 KiB	5.6 KiB	0 B (0.0%)	6.8 KiB	✅
Claude	Thread snapshot wire	4.3 KiB	4.3 KiB	0 B (0.0%)	4.9 KiB	✅
Claude	Live turn WebSocket wire	1.3 KiB	1.3 KiB	0 B (0.0%)	2.0 KiB	✅
Claude	Live turn WebSocket decoded	26.6 KiB	26.6 KiB	0 B (0.0%)	29.3 KiB	✅
Claude	Live turn messages	2	2	0 (0.0%)	8	✅

Baseline: 17f7bfb · PR result: 68919d6 · Source CI: success

Scenario and decoded snapshot size

10 historical turns, 5 command tools per turn, 878.9 KiB retained MCP result per historical turn, and a 1.05 MiB retained result in the measured turn.
Full compact HTTP snapshot with live cursor.

• Codex decoded thread snapshot: 133.1 KiB
• Claude decoded thread snapshot: 133.4 KiB

Updated in place by a trusted workflow. PR artifacts are strictly validated and never executed.