Release 1.0.3 with documented Pluto limitation on macOS Intel

[Azzaare]

Released 1.0.3 — documented macOS Intel Pluto limitation

PerfChecker 1.0.3 is public on GitHub. PR19 was merged normally as 06b6d36, with parents fa5634e and 0649f92, tree 5121ee16d7ddcc9839c9416d13cefa9d5f988604.

Publication was explicitly authorized with 34/35 native groups physically qualified and the macOS Intel Pluto first-notebook limitation accepted for this version. Both general CI campaigns passed4/4 (push38826272023, PR38826278390). Of the previously required28 groups,27 passed physically; all seven non-blocking Windows1.96 groups also passed physically. No failed or missing qualification is represented as PASS.

macOS Intel Pluto: known limitation for the next version

Original native job116734466127/attempt1 failed FRESH fresh-pluto-published-install: the first native notebook was absent after the one-hour harness availability wait. The cause is unknown. Its later configured/prepared passes do not compensate for that failure. Artifact upload then failed separately with ENOTFOUND, leaving no complete receipt for this lane. The original FAIL and absence of qualification remain recorded. Targeted retry116823640672/attempt2 was accepted on the same V/H3; cancellation was requested when the user authorized publication with NO NEW CI. No retry outcome is counted as PASS and no further CI was requested.

H4 packaging correction after the release tag

Marketplace rejected H3 because its icon was1254×1254 pixels. The same transparent logo is now1024×1024; source commit a91d2b7/tree788420ef60394923ff3377ef144bd293db21ef9a changes only media/perfchecker.png. The v1.0.3 tag stays at06b6d36281bd208270595c95ddbd2ea8f0657e30. The current download is H4, a deterministic icon-only derivation of H3:50 members, only the icon changed,49 other members AND compressed data byte-identical, including manifest and all compiled modules. No product rebuild, compile, test or CI was run.

H4 VSIX826236B/SHA256c1eca4caf5a18a1884079cac60cece6d5bb753701bbc0ebb820e15447032d68a; checksum81B/SHA2564d8ca1311b789a4abb234bea03c9051245262fff5b42c6c9d95e3e3810ded169; provenance1813B/SHA2567240b757ac64346a23399cd5d0673e343596b9f42c93f6bdaf70b0cc535f2284. Provenance identifies the after-tag icon correction and original H3 source/run. Native evidence and the accepted Intel Pluto limitation remain unchanged.

Final behavior, original H3 archive and evidence scope

PerfChecker1.0.3 uses Core1.1.1. Explicit Refresh after a saved edit invalidates old TestItem inventory and queues one follow-up scan after physical child close; Resolve joins the correct pending scan. Deactivation rejects queued discovery before controller shutdown. Both Copilot Studio idle-snapshot and Pluto Feature suite fixes remain preserved. Reviewed qualification fixes retain validated idle snapshots, fail-closed process ownership and the narrowly scoped wait_task wall-profile reuse fixture; the CPU fixture and correctness oracles are unchanged.

The initial GitHub release assets were the exact H3 bytes, packaged once on Linux by source4a27ae0d76a297c58d141c8396808e5d86e61ef3/treea20c19dc66670bf39bb3e203ede0081f6048ddcf from origin38769699307. Production source and compiled modules remain byte-identical. VSIX1079814B/SHA256416ea51fa31471545b022ca37f7ab30901a2e36b871ba230b54bbeadb97943ec; checksum81B/SHA256969312065029ae5763370a0687a50c2e385af0afea0379bf3a3aeffc5efbf7ab; provenance267B/SHA2565755d71840acb366010171792889fce565bf1f1a800ee89d74832b7863390e4c. At initial publication no rebuild, repackage or new origin was created; H4 is the later icon-only packaging derivation documented above.

The34 passing receipts prove actual merge65d98 checkout, H3/modules, FRESH5 / CONFIGURED4 / PREPARED8 shared prefixes plus only their announced tails, completed assertions and scoped cleanup. The45 fresh bootstrap cases remain qualified. Core resolution is phase-specific: fresh General1.1.1 is registered/tree749 in both Controller and Pluto; configured/prepared use candidate1.1.1/d079efd5c24c3dcac5b7550781822038e0ff780f/tree749f8eac351ed2a1f0bc110456f2eb37a3bc08ee, not a uniform registry installation. Authentication, GPU rendering, unavailable TCP cases and arbitrary callbacks retain their recorded limits. Positive availability3600s/shared330min plus30min reserve/job360min do not change deliberate timeout/budget/cancel or product/RPC limits.

The Copilot-based product passed one local TypeScript compilation and 373 Node tests: 345 passed, 28 skipped, zero failed/cancelled. The new harness diff passed three syntax checks and one complete controller-preflight suite: 47 passed, zero failed/skipped/cancelled. Node 24.14.0 used CPU19, inherited single-thread pools, MemoryMax512 MiB / High448 MiB, swap0 and OOMGroup1. Those Node checks did not execute Julia, Electron or browser workloads. A subsequent targeted local Julia 1.13.1 diagnostic used four sequential Malt workers, CPU19, MemoryMax2 GiB / High1792 MiB, swap0 and OOMGroup1. Fresh and reuse wall profiles passed the real immutable wait_task fixture, oracle42, exact source roots and nonempty target filtering (sums of reported stack-row sample counts81 and207 across3 rows respectively). Both CPU profile cases returned Malt.RemoteException; their cause is unknown because only class/length/hash were exported. The initial local montage failed before workers because PerfChecker.Malt was not bound; the corrected montage used parentmodule(PerfChecker.Worker). These local outcomes establish neither Windows1.13.2 causality nor performance qualification.

Previous U962 native run38799879202 remains globally failed: Windows stable Workbench116660198609/receipt12013486616 failed only native-complete-suite-measurements because baseline wait_task_wall_profile had no samples rooted in target source (5.563s); Testing and the Workbench tail passed and do not compensate for the failed lane. Its17 physically inspected native passes and two successful general CIs remain historical evidence on U, not qualification of this new source. The precise Windows cause is unknown. No rerun, cancellation, manual dispatch, new archive origin or product change was made for this fixture adjustment.

The two versioned workflows were temporarily disabled through the API for the normal merge/tag/release, then restored active. No versioned workflow or product was edited for publication. The historical evidence below is retained verbatim and is not current qualification.

T15a failures retained

T15a PR CI38779088038, Windows job116574300310, failed test/codex-connector.test.mjs:295 at its FIRST inspectCodex --version preflight:10.052s, stdout/stderr0, code null/SIGKILL, JS-entry marker not observed. This does not prove an MCP RPC failure. CLI connector/owner/PowerShell source is identical to v1.0.2/fa5634e7290514e0f43363f2188711308a029129; cause before the missing marker remains unknown and no new1.0.3 product defect is established. Independent automatic T15a push CI38779079286 passed4/4, including two Windows367-test passes (338PASS/29SKIP/0FAIL), browser2/2 and four real CLI preflights. This does not erase the PR failure. No targeted Windows PR POST was sent.

T15a native38779087808 failed Linux1.96 Studio job116587059873 before CONFIGURED/PREPARED: fresh5 passed; cache preparation retained an inspect ENOENT for unqualified PID4890 despite its existing same-survey current=null/proved-absent observation. Receipt12002295317/SHA256876bb3ad1d4e6d09cd4038d25218d2ef61dc0c3bf1dd875facaeaebe10e3d098 preserves that original refusal and cleanupQualified=false; no Studio-tail, leak or product failure is inferred. The reviewed harness correction addresses that precise sticky refusal. T15a remains globally failed; its other passes and outstanding lanes cannot qualify this new SHA. Old jobs were not cancelled.

Subsequent historical outcomes retained

The old source131 attempt1 ended with 31 physically inspected functional passes and four failures, all retained: Linux1.96 Budgets official Julia extension installation HTTP503 after CONFIGURED and before PREPARED; Windows stable Narrative's FRESH Studio idle-baseline assertion (subsequent Pluto blocking was secondary); Intel Studio's PREPARED idle-baseline equality; and Intel MCP/Pluto's FRESH Feature suite click timing out before installation. Other passing prefixes or tails do not compensate for a failed lane. Copilot's external retry of old131 is historical and does not qualify this source or H3. The three automatic EE-only runs ended action_required with zero jobs and supplied no validation.

All earlier failures, diagnostics, limitations and snapshots remain below verbatim, including the original 10-second CLI failure, preceding native failures, diagnostic-only outcomes, transport failures, cancellation histories and the stale firstImportBoundSeconds annotation. None is converted into a PASS by the new code or bootstrap evidence.

Previous PR description and historical evidence, preserved verbatim

Fresh controller setup and Pluto installation/upgrades previously selected Core 1.1.0, whose CpuId dependency prevented ARM loading. Both installers now resolve Core 1.1.1 or newer below 2.0 with their requested packages and verify the result. Existing-environment compatibility checks and explicit installation consent remain unchanged. Core 1.1.1 is [registered in General](https://github.com/JuliaRegistries/General/pull/171712) at tree `749f8eac351ed2a1f0bc110456f2eb37a3bc08ee`.

Studio and Testing discovery now share equivalent in-flight requests, freeze invocation settings and reject stale results. Trust, folder membership and current settings are rechecked before spawning/publication. The packaged README consistently recommends Core 1.1.1 for ARM and read-only worker fixes while preserving historical compatibility evidence.

Qualification also strengthens two test fixtures without changing product code or the intentional timeout/budget/cancellation cases: profile/wall-profile operations accumulate 64 checked reductions inside the target module; narrative reload waits for the exact saved measurement's timestamp and deeply identical rendered runs, with idle UI and no stale explanation, before Explain. These fix demonstrated fixture deficiencies; neither establishes the deeper cause of a historical failure.

The native harness now gives positive availability/readiness observations 3,600,000 ms (one hour): Suite designer cards and Investigation navigation, Testing appearance and RunTest evidence, plan/checktypes, direct native command results, preparations and Playwright auto-waits. This is the requested qualification policy, not a product fix or a calibration claim. Product/RPC durations and intentional timeout, budget, Cancel, ownership, inspection, drain and teardown assertions retain their distinct bounds. Positive analyzer fixtures requiring completion allow one hour; deliberately expiring fixtures remain separate.

Each native job has one 330-minute shared activity budget, anchored before runner setup and including preparations plus fresh → configured → prepared, followed by a 30-minute technical reserve within the 360-minute GitHub job cap. A late availability wait is clipped by remaining activity, so a full hour is not guaranteed near exhaustion. Infrastructure expiry is explicit. Observation expiry does not prove an operation stopped: unqualified preparation trees or pending download/extraction cause session retention. No phase renews the budget. Historical diagnostic flags are refused on this source; their original branches remain immutable.

The seven functional groups remain unchanged, with all eight prepared prefixes before each declared tail:

Function	Tail covered
Studio & suites	Studio controls
Julia workbench	Workbench controls
Investigations & reports	Investigation controls
Budgets & deadlines	Investigation limits
Narrative & HTTP lifecycle	Narrative protocol
Providers & model setup	Advisor controls
MCP + Pluto + discovery	MCP → Pluto → discovery/sync in one physical session

They run on Linux x64 and Windows x64 with VS Code 1.96/stable, plus macOS Intel stable: 35 native jobs, seven per configuration. MCP → Pluto → discovery/sync stays in one session. Qualification source changes; the canonical VSIX is reused unchanged.

Exact source and archive

• PR head: 131430d; tree c0db27cfcc6c30645fc4a2e26654d495467da9e0; parent AA aa36296. Only 21 qualification files changed; no product/resource/package/lock/README/changelog delta since Q.

• Merge-test: 7979385; exact parents main fa5634e and head 131430d. Merge tree equals head.

• Immutable VSIX H2: 1,079,048 bytes, SHA256 bc61f9fa172069c61722e0691900d3ddf8a39ce143fd9d553b252304dbc55850, packaged once on Linux from Q 73b99deb87efaf45bd5e96036f755e42c748bdab / tree 3722727a870402497078e63da78be90d606f5ed5 in origin R2 38653410110.

• Original artifact 11935253207, name native-candidate-73b99deb87efaf45bd5e96036f755e42c748bdab; ZIP 1,079,814 bytes, SHA256 6b457845b680b2b54e84677439071a2ae2ebeef8d9849989d6654b6c8b23a85b. VSIX/checksum/provenance and packaged README were inspected.

• Current merge-test reuse artifact 11968813750: ZIP 1,079,814 bytes, SHA256 3e7d18aa5363b4cfac94b2f853665d41af0c81a61e3a14cd727afbd25fa03296. Its three assets were physically inspected: H2 1,079,048 bytes/bc61f9fa…, exact 81-byte checksum/41452082… and 267-byte provenance/75fc9267…, all matching original Q/tree372/R2. This is a reuse upload, not a new package origin or native-lane PASS.

Current validation — publication blocked

Both current automatic general CIs are terminal SUCCESS, 4/4 jobs each: push 38729843666 and PR 38729851787. The native campaign 38729851490 remains attempt 1 on head 131430d / merge-test 7979385, with identical tree c0db27cfcc6c30645fc4a2e26654d495467da9e0. At 12:35:08.728 UTC on 2026-10-15, all 35 native lanes were accounted for: 2 physically verified PASS, 1 infrastructure FAIL, 20 active and 12 queued. This is not 35/35 qualification; publication remains blocked.

Current lane	Job / receipt	Physical ZIP and result
Windows x64 / VS Code 1.96.0 / Julia workbench	116404755074 / 11972230935	1,503,025 bytes; SHA256 215245f292b1cde98fa4c8ffe60f10a909ed22ef77f5911b40c08088967087c4. Fresh 5/5, configured 4/4, all eight prepared prefixes and the declared workbench tail PASS.
Windows x64 / VS Code 1.96.0 / MCP + Pluto + discovery	116404755122 / 11973240780	2,171,407 bytes; SHA256 b8feca8dd2a7386efe911439b0d92b877c32a04bf888baa9f7fa94d7af2df2f1. Fresh 5/5, configured 4/4, all eight prepared prefixes and MCP → Pluto → discovery/sync PASS in one session.
Linux x64 / VS Code 1.96.0 / Budgets & deadlines	116404755007 / 11970331572	1,347,888 bytes; SHA256 227e6a02065b3b4a52aa6be1064700aa9f84a0b87732c46782eb95f2f6dabf1d. Fresh 5/5 and configured 4/4 PASS; prepared never started, so its eight prefixes and Budgets tail were not executed.

All three receipts contain the exact physical H2 VSIX (1,079,048 bytes / SHA256 bc61f9fa172069c61722e0691900d3ddf8a39ce143fd9d553b252304dbc55850), Q/tree372/R2 provenance and Core candidate d079efd5c24c3dcac5b7550781822038e0ff780f / tree749f8eac351ed2a1f0bc110456f2eb37a3bc08ee. These are separate lane receipts, not reuse-package success borrowed as native evidence. The original VSIX, 81-byte checksum and 267-byte provenance remain intact.

The Linux Budgets infrastructure failure is HTTP 503 while installing a VS Code extension, at 12:04:16 UTC, followed by code exit 1. The log names no extension or upstream endpoint; the candidate/full source sequence identifies the pending installation as the official julialang.language-julia extension, not installation of the Julia runtime. This is not a demonstrated product defect or 3,600-second timeout. Preparation owner PID 6875 physically closed, but its process tree was unqualified and the private session was retained; no global child-absence claim follows. The original failure and cleanup limitation remain recorded.

Exactly one targeted rerun POST was attempted for job 116404755007 at 12:28:50 UTC and was refused with HTTP 403: the workflow run was already running. POST count 1, accepted count 0; no other job was rerun or cancelled. A future targeted attempt must wait for authoritative whole-workflow completion and preservation of the other 34 physical lane receipts. Any future PASS requires a new upload ID and digest plus new physical H2/Q/R2/Core and fresh/configured/prepared-prefix/Budgets evidence; the previous receipt cannot qualify a rerun. Artifact attempt-scoping remains unknown; no overwrite or deletion is authorized.

One proof annotation is stale: official-julia-repl reports firstImportBoundSeconds=360, while the actual source wait in native-workbench-controls.cjs:128 uses NATIVE_READINESS_MS=3600000. The observed import took 192,024 ms. No assertion or result was changed, and this annotation does not create a failure or a new global claim. Authenticated Codex and GPU rendering remain outside the acquired scope. MCP/Pluto cancellation, foreign-process preservation and owned-worker cleanup are qualified only for their observed cases; arbitrary user cleanup callbacks are not guaranteed, and no global process-absence claim is made.

Local Node 24.14 validation: 19 initial JavaScript syntax checks plus corrected Results-adapter syntax passed; the complete pure-Node suite passed 343 tests, 0 failures, 28 skips, in 66.968309766 seconds. Skips are Julia/browser/real-Codex opt-ins and platform-specific Windows cases. Tests exercise real readiness helpers and actual dispatch paths: exact document identity and one click after 30 simulated minutes; shared-budget consumption/clip/expiry; pending preparation retention; original errors; protected Cancel bounds and unchanged polling. The preceding local run's seven Results-adapter ReferenceErrors were corrected by binding the real constant; those were local validation failures, not CI or product verdicts. No compiler, Julia, browser, benchmark or packaging ran locally for this change.

Publication requires both current general CIs and every current native lane with physical H2/Q/R2/Core provenance, all three phases, eight shared prefixes and its declared tail. Their union must cover all nine historical tails on every configuration. The acquired General bootstrap evidence covers 45 fresh x64/ARM cases under its separate contract; nativeUpgrade and full ARM remain unqualified. No merge, tag, GitHub release or Marketplace publication of 1.0.3 has occurred.

Initial 131 validation snapshot (historical, retained verbatim)

The source has one automatic push CI 38729843666, PR CI 38729851787 and 35-group native PR campaign 38729851490, all attempt 1. At the initial 11:49:53 UTC snapshot, API listed 37 native-workflow jobs: package reuse SUCCESS, connector ownership SKIPPED and 35 uniquely named native lanes active/queued, seven per configuration. General CI jobs were active/queued. Current qualification remains pending. Package success and earlier-source passes cannot qualify these lanes. No manual full dispatch, global rerun, rebuild or new package origin was performed.

Subsequent historical AA and diagnostic outcomes

• AA Studio job-only rerun, attempt 2 job 116338254119: FAILURE at 10:23:12 UTC, distinct from attempt 1's official Julia-extension installation read ETIMEDOUT and secondary upload ENOTFOUND. Fresh/configured passed. The fourth prepared prefix native-suite-run-button began 09:34:17.541, clicked Feature suite 09:34:18.778, then failed locating real #cards at 09:36:19.854 under the original 120-second oracle. Checktypes readiness was not reached. The capture showed visible Studio/planning status without a designer frame; observer inspection errors leave actual plan PID/close/completion and internal cause unknown. Seven other prefixes and the entire Studio tail passed; these do not compensate. Testing and RunTest correctness passed. Receipt 11964393089: 3,133,800 bytes, ZIP SHA256 3f1a581aa2ba5ace7a072d06deef857b9d58357e2d4d0068233e1b07ce084d19; physical H2/Q/R2/Core exact.
• Nonqualifying Investigation UI diagnostic a86, run 38717906470: FAILURE at 11:16:52 UTC. Unique failure: the prepared native-suite-run-button / real #cards acquisition under the original 120-second bound (10:35:38.611 → failure capture 10:37:39.688), without a deeper cause established. Fresh 5/configured 4 and seven other prefixes passed. Prototype Testing-positive oracle was 180 seconds, RunTest evidence 180 and Investigation navigation 120. Testing tree-ready was observed 10:44:46.300 after the 10:43:35.533 oracle journal; RunTest clicked 10:44:46.706; complete/correctness-passed evidence recorded 10:45:59.772. Investigation controls themselves passed 10:53:00.376 → 11:16:27.590, including nine complete analyzers. No Investigation acquisition failure was reproduced. The observer used zero failure snapshots, reached its 256-event cap and retired drained/listenersRemoved=true; unassociated requests and the cap limit exhaustiveness. This does not qualify AA, the new source or H2 globally. Receipt 11968067549: 2,130,611 bytes, ZIP SHA256 eb772227f912db025fa7bf625d8e6ced7947f7d5496fd94527db1f2c797b9dd8; physical H2/Q/R2/Core exact.

Historical AA snapshot retained verbatim

The following earlier snapshot and all preceding-source failures are preserved. Its pending Studio-attempt-2/diagnostic statements are historical, superseded by the terminal outcomes above. No historical FAIL is requalified by a later diagnostic or the new waiting policy.

The exact aa36296 35-job full native campaign 38690836321 finished its first attempt with 32 physically verified SUCCESS jobs and 3 FAILURE jobs. The last Intel MCP/Pluto/discovery job finished at 08:39:10 UTC on 15 October. Both push CI 38690829709 and PR CI 38690836372 finished SUCCESS, four jobs each. These passes do not compensate for failed lanes or qualify the complete release.

Configuration	Physically passing functions	Remaining failures from attempt 1
Linux x64 / VS Code 1.96	All seven	None
Windows x64 / VS Code 1.96	All seven	None
Linux x64 / VS Code stable	All seven	None
Windows x64 / VS Code stable	All seven	None
macOS Intel / VS Code stable	Workbench, Narrative, Providers, MCP/Pluto/discovery	Studio transport; Budgets Testing prefix; Investigations fresh initialization and prepared webview

Every passing receipt contains physical H2/Q/tree372/R2 and Core candidate d079/tree749, all three phases, all eight prepared prefixes and its declared tail. The Intel MCP→Pluto→discovery/sync chain passed in one session; receipt 11958901154 is 2,669,480 bytes, ZIP SHA256 84c23709676723bd05a15021c6b66a84d5d104cb65920caf3d8f722417f1af1c. Cancellation/restart/stop/close/home assertions verify owned-worker and owned-file cleanup while preserving foreign processes. Its external-browser export was skipped on Intel, plotting prerequisites were unavailable, and arbitrary user cleanup callbacks are not guaranteed after forced termination.

Only the Intel Studio transport job was explicitly rerun after the campaign became terminal: one job-only POST returned HTTP 201 at 09:01:12 UTC. Attempt 2 job 116338254119 started at 09:01:19 and was active at the 10:02:13 UTC snapshot, with the same SHA/REF/archive/Core. No other job rerun was requested. GitHub's attempt-2 catalog also copies prior results under new IDs with their original timestamps; those aliases are not new executions or new physical qualifications.

Current failures are retained:

• Intel Studio, job 116276184252: fresh and the configured missing-target negative passed. Before prepared started, installation of the official Julia VS Code extension failed with read ETIMEDOUT at 06:40:59.673 UTC, followed by CLI exit 1. No prepared/Studio oracle ran. The later artifact upload independently failed with ENOTFOUND, leaving no receipt for this job. The initial explicitly authorized job-only rerun request received HTTP 403 because the workflow was still running and created no attempt. After the workflow finished, the separately authorized single POST described above created attempt 2; its result remains pending. This transport failure is distinct from Testing discovery.

• Intel Budgets, job 116276184375: fresh/configured passed; the seventh prepared prefix, native-testitem-measurement, failed with an actual locator.waitFor timeout of 120000 ms (case started 06:51:13.003 UTC; failure recorded 06:53:16.917). The six earlier prefixes passed, and complete-suite subsequently passed with 30 passing runs and four expected unavailable runs. The Budgets tail itself passed, including its intentional configured timeout at 07:06:23.502 and elapsed-budget refusal at 07:08:14.297, then case PASS at 07:08:16.333. No 130-minute global limit was reached. Receipt 11954715907, 1,902,059 bytes, ZIP SHA256 310344666d35d445b943d31e60861def66286bff54082b3efe6d53210067a58d, physically matches H2/Q/R2/Core. A qualified expected Julia mapping for PID 5758 was observed before the UI failure. Final retained data contain a valid listing for one UUID (listing SHA256 aeb6af849110001c429f29abd6c0ff3623269b140f17799d864c55bcddce21a6, UUID-directory SHA256 70465e3274ce8106fbef3e1bfe62c21d27a020fcf777adddc9f8297e93ff436c) and its matching Evidence line. This establishes child close and JSON read/parse for that UUID; it does not establish code 0, PID-to-UUID binding, exact completion after 120 seconds, refresh guards or UI publication. Copied ZIP/file timestamps are not CLI completion timestamps. The internal cause remains unknown. Scoped preflight cleanup passed; this is not a claim that an unobserved product tree was globally empty.

• Intel Investigations, job 116276184456: fresh initialization reached its explicit 600-second plan deadline (06:55:44.443→07:05:44.443), with the plan-ready refusal recorded at 07:05:44.543; the following fresh Pluto case was blocked. Configured passed. Prepared passed all eight prefixes, including Testing readiness at 07:34:50.035 and RunTest correctness, then four investigation subgroups passed. Eight subsequent subgroups failed locating the real Investigation webview, from 07:51:33.809 through 08:05:34.422. The aggregate reports nine failures, with the additional child error not independently serialized; the source finally can add teardown failure. These are not nine analyzer failures. A retained real benchmark for the adopted scenario has availability complete, correctness passed and two observations per metric. No prepared failure image exists; empty visibleDesigner and the retained renderer logs do not distinguish closed/inactive/blank/modal/stale-frame causes. Receipt 11956897911, 2,030,440 bytes, ZIP SHA256 938d062d81092cd74edc14eeeab6ec38d79bc817bee23365a73c5b41828990b5, physically verifies H2/Q/R2/Core. No global 130-minute limit or deeper cause is established.

Publication requires every current functional job, its physical source/archive provenance, all eight shared prefixes and its declared tail to pass. Their union must cover all nine historical tails on each configuration. Previous successes or a nonqualifying diagnostic cannot replace these gates. No 1.0.3 merge, tag, release or Marketplace publication has occurred.

• Local validation of the test-only diagnostic: five syntax checks and 51/51 Node tests passed, zero skips.
• Local validation of the functional split: two syntax checks and 54/54 Node tests passed, zero skips. Tests execute the actual package matrix-output script and host dispatch paths with controlled callbacks; native effects still require remote qualification.
• Isolated Intel discovery diagnostic 38689728030, source 789/H2, finished FAILURE with the original 120-second Testing UI timeout after fresh/configured and six prepared prefixes passed. Exactly one listing Julia was launched. Its initial process snapshot qualified; observations at 60/85 seconds became unknown at image canonicalisation with ENOENT, before CPU/stack sampling. No sample ran. Observers physically closed/drained; product-close was not observed, which is not evidence of a leak. Receipt 11953645404 physically verifies H2/Q/R2/Core. The full prelude and 120-second Testing/180-second worker oracles were unchanged; complete-suite/Studio/Workbench were intentionally excluded after the target. This diagnostic is perturbing and always nonqualifying.
• A separate test-only diagnostic correction, commit 61318f209818a0d7f621cc68e68146b8dcead43a on diagnostic/testitems-discovery-20261015, selects the exact expected mapped Julia file instead of requiring every unrelated mapped library to remain resolvable. Exact PID, canonical file identity before/after and between observations, process incarnation, deadlines, abort and cleanup refusals remain strict. Two independent reviews approved diff SHA256 f4bb9a507b92a5257781d2cc66e9a313d0b3a93961fbd02fbdc0f3f9a0dd7f9b; two syntax checks and 50/50 Node tests passed, zero skips. This commit is not part of PR head aa36296 and does not change the product or fix/prove the internal UI-timeout cause.
• Corrected isolated diagnostic 38698147856, source 61318f2, attempt 1, finished SUCCESS but remains nonqualifying. One Julia PID 3013 physically closed with code 0 after 104.603 seconds; its exact UUID listing was valid after close, and the Testing row was visible 104.265 seconds after the explicit UI-oracle start. RunTest correctness passed. The qualified process had accumulated 55.05 seconds CPU at the 60-second snapshot. The 85-second sample attempt was inconclusive (SIGTERM after 4.829 seconds), with no valid stack projection. Observers physically closed and drained. Receipt 11957367126 is 1,625,411 bytes, ZIP SHA256 6b3d91d665b12b823b18b26cef76ca9cec22c630c8ddf6d69772c51d2178b738, and physically retains H2/Q/R2/Core. This perturbing observation does not calibrate the UI oracle or establish the historical timeout cause.
• Separate diagnostic-only commit f46cc2037bf3ae8908367d47dfdefe73f788c973 (tree 07cd2f5f30e2294531428682441c991e4d2a5baa, parent 613) adds passive completion observation. Four syntax checks and 63/63 Node tests passed, zero skips, after two reviews of diff SHA256 8f5d05315728ce97b6168b76281920c65cd0b9b3b4805946e30ceddda7c13a3e. It follows the original ChildProcess and exact output UUID without extra Julia, ps, lsof, CPU or sample/manual inspection during the measured discovery case. The original UI120 failure remains a failure; observation may continue only until spawn+180 seconds, capped by the unchanged phase bound. There is no late RunTest, retry, warmup or product timeout change. A single late visibility query cannot establish TestController publication or which refresh published. Earlier full-prelude actions and ordinary teardown remain intact; suppressed process observations make this diagnostic different from prior probes.
• Passive isolated diagnostic 38710233332, exact source f46cc20, attempt 1, finished FAILURE at 09:40:51 UTC on 15 October and remains always nonqualifying, outside PR head aa36296. Fresh/configured and six prepared prefixes passed; the original positive Testing appearance oracle failed after 120008.645 ms at 09:39:51.179. One captured ChildProcess (Julia PID 76290, parent 64030) started at 09:37:50.514 and physically closed with code 0 at 09:40:30.173, 159658.945 ms after spawn. Its exact UUID directory was bound at 09:37:50.529, before close. Reading the original result after close returned schema perfchecker-testitems/1, one exact expected target, matching root and source SHA, 456 bytes, listing SHA256 883ce415f8d0c5e75107abd80348a714fb406fd33823cf67d4854d38f7a7f6cb. This proves correct CLI completion after the UI bound for this invocation, not an internal JIT cause or a calibrated canonical UI deadline. A single isVisible query at 09:40:30.230–.239 returned false with unchanged context; visibility at that instant is absent, while TestController publication and which refresh published remain unknown. No late RunTest ran, and the original failure is preserved. Observation drained with zero external observer processes and no sample. Receipt 11962840416 is 1,820,129 bytes, ZIP SHA256 cb5298af927d596bd675e83018a73c7b2e7dc2c2b6742553ee479e29b27af073, physically retaining H2/Q/tree372/R2/Core. Reuse artifact 11959219964 (1,079,814 bytes, ZIP SHA256 84e2089fb525df4b8d4c9e6d9de5aa3224eae47c767b992a7c6e6dfeaffad713) is a source-named reupload, not a new package origin.
• Isolated Investigation UI diagnostic 38717906470, source a86ab45 (tree 4927e07c88017407c6703fda317b795de0059832, parent aa36296), attempt 1, has exactly one Intel stable Investigation job 116365145532, active at the initial 10:07:11 UTC snapshot. This separate diagnostic branch is not part of PR head aa36296 and is always perturbing/nonqualifying. It keeps the full fresh/configured preparation, eight prepared prefixes and complete Investigation tail. Only its positive Testing appearance wait is prototyped at 180 seconds; negative Testing remains 120 seconds, Investigation navigation remains 120 seconds, RunTest evidence remains 180 seconds, and phase/job bounds remain 130/180 minutes. Lightweight callsite markers distinguish reportAfter acquisition from the later Findings acquisition; the first acquisition failure can trigger one bounded private-data-free DOM observation, within a maximum of four. Frame/document generations, exact script attribution, bounded latest-eight request candidates, idempotent listener retirement and original-error preservation were reviewed twice. Six syntax checks and 38/38 portable Node tests passed with zero skips; this does not qualify native UI behavior or justify changing canonical AA's deadline. No retry, extra Julia, OS sampler, product edit or package rebuild is included. H2/Q/R2 and exact Core candidate d079/tree749 are reused.

Acquired scoped evidence:

• Previous 5e8 full campaign: five physical passing receipts were acquired (Intel investigate 11948087934, Linux 1.96 11948054757, Intel agentic 11950235692, Windows 1.96 11949911186, Linux stable 11950639073), all with exact H2/Q/R2 and Core candidate d079/tree749. Each passed its three phases, eight prepared prefixes and declared tail. Intel measure failed its Testing prefix; Windows stable subsequently ended CANCELLED at GitHub's three-hour job limit; its retained prepared phase had already exceeded 130 minutes inside Pluto, with global cleanup unqualified (details below). Five passes do not compensate for the failure or qualify the new source. Linux stable's 155,533,028-byte ZIP SHA256 is e73baf86a072e447e664cfa3461c2d448f88f92e4f21e81a998cdf4d082ab23b; all nine tail groups passed on that host, including strict Narrative reload/cancel/deadline and MCP/Pluto/sync. Windows-specific ownership fields are not inferred from Linux.
• H2 bootstrap: 25/25 fresh x64 cases across five hosts and 20/20 fresh ARM cases across Linux/Apple Silicon × Julia 1.10/stable, using registered General Core 1.1.1. All nine physical receipts were inspected. These cases are unchanged by the later fixture edits. Native Pluto upgrade and full ARM workflows remain unqualified.
• Focused Windows profile fixture 38667136233, source fd2: actual H2/Core candidate d079/tree749, 32 runs passed and two known unavailable. Both dev/baseline profile and wall-profile passed their strict numeric oracle with 25/12 and 34/10 retained grouped stack rows. Rows are not raw sample counts; the unchanged Core target-root/nonempty-group assertions establish rooted nonempty results. Receipt 11942522942 was physically inspected; an explicitly exceptional second read inspected previously omitted success reports under suite/final.
• Focused Intel narrative 38673134401, source 5e8: two exact measured-report DOM acknowledgements passed for the same four runs. A real held second HTTP request was cancelled, its socket closed before fixture teardown and controller cleanup preserved configuration/source/measurements. A third request reached its unchanged 45-second deadline, reporting timeout after 45.668 seconds with its socket closed; fixture close and settings restoration completed. Receipt 11945912255 (1,149,586 bytes, ZIP SHA256 4f38550f041d9176b9d91991dbdcef142c075d57058fdff9000ef4f733a823fc) retained physical H2/Q/R2, exact Core candidate and internal run/narrative reports. This focused environment does not qualify a full campaign or the current 35-group coverage or prove the historical 130-minute cause.
• TypeScript compilation of the unchanged product passed locally; initial discovery/cancellation checks were 3 passed/0 failed/1 explicit Julia skip. The narrative ACK change passed both syntax checks and 3 targeted Node tests with no failures/skips. No local Julia, native browser or benchmark qualification is claimed.
• R2's package test ran Node 22.23.3/npm 10.9.9: 339 tests, 311 passed, 0 failed, 28 explicit platform/runtime/authentication/live-browser skips. Its separate focused Linux producer case passed with physical evidence.
• Windows diagnostics 38660395125 (instrumented temporary copy) and 38664980295 (external observer of unchanged H2) passed targeted tests, with original files and diagnostic cleanup verified. Both explicitly set qualifiesCanonicalProduct=false; observation can perturb timing and does not explain an earlier 10-second failure.

Failures retained on preceding sources

Source/campaign	Real failure and limits
Previous 5e8 full 38675933380, Intel measure 116229487703	Fresh/configured passed. Prepared native-testitem-measurement failed Testing visibility under 120 seconds (04:13:31.819–04:15:35.928 UTC): empty tree and no observed listing result. One qualified controller Julia PID11509, created 04:13:33, remained present with the same parent/incarnation/image before cleanup; workspace trust, empty tag/exclusion filters, source and controller files were stable. The earlier preflight listing took 149.971 seconds in a different preparation state, which does not establish the UI cause. Complete suite, Studio and Workbench subsequently passed; no global limit or provider error occurred. Receipt 11949586841 was physically verified; an explicitly exceptional second read inspected previously omitted owned-process/configuration snapshots. Initial cause, CPU state and eventual close/exit remain unknown.
Previous 5e8 full 38675933380, Windows stable 116229487754	GitHub verdict CANCELLED, not PASS. Fresh/configured and all eight prepared prefixes passed, followed by seven tail groups through MCP. Prepared reached its unchanged 130-minute bound at 03:41:58.504–05:51:58.640 UTC, with lastObservedCase=native-pluto-controls, hostStopRequested=true and productCleanupQualified=false. Pluto has no final verdict; discovery/sync did not execute. The last observed Pluto milestone at 05:24:25.281 had verified active-allocation cancellation, worker termination, owned-directory removal and no new memory report; this scoped proof does not qualify later completion or global cleanup. The job then reached GitHub's annotated maximum execution time of 3h0m0s (job 03:22:53–06:23:05 UTC; Windows step stopped at 06:22:57). The cause of the intervening wait remains unknown. Receipt 11951834469, 2,462,772 bytes, ZIP SHA256 92e79951f74c6b88c2e635a68be6a9a93b8d7719481b326c951c51b3cdfddc87, was inspected once and retained exact physical H2/Q/tree372/R2 and Core candidate d079/tree749.
F2 push CI 38654332092	Windows authenticated Codex fixture's first --version exceeded the unchanged 10-second bound without observed JavaScript entry. Cause undetermined. Separate PR CI 38654338128 passed 4/4 and does not compensate.
F2 full 38654338136, Windows stable 116160495667	Fresh/configured passed; prepared suite exited 1. Its real compatibility report passed, then sum_squares_wall_profile dev@0.1.0 errored: “CPU profiler found no samples rooted in target source” at Core profile168. Baseline counterpart passed; 31 other runs passed/two unavailable. Prepared Testing and all nine later groups passed but do not compensate. Generic UI cleanup/inventory wording is not evidence of Core cleanup failure. Raw samples/path/compilation cause remains unproved. Receipt 11940211346 had an explicitly exceptional second read of previously omitted first-failure reports.
F2 full, Intel measure 116160495734	Five defects: fresh initialization exceeded 600s and dependent Pluto was blocked; prepared designer cards absent, Testing visibility exceeded 120s and Workbench lacked Julia REPL evidence. Configured, complete suite and final Studio passed; no global phase bound reached. Receipt 11940117876; deeper causes unknown.
F2 full, Intel agentic 116160495704	Fresh passed; configured Testing visibility exceeded 120s. Prepared passed all eight prefixes, then reached its real 130-minute bound (00:02:10.525–02:12:10.563 UTC) inside narrative, with hostStopRequested=true/productCleanupQualified=false. The first response/export completed at 00:40:01.316; later cancel/deadline/finally were not distinguished. No agentic tail has a final verdict. Receipt 11943770104 had an explicitly exceptional second read of omitted global-limit/last-check metadata. This is not evidence of a server deadlock or the cause of the earlier Testing delay.
fd2 full 38669330756, Intel measure 116208559616	Fresh/smoke passed. Before configured/prepared, cache preflight failed its strict identity/cleanup qualification: owned PID32329 inspection via lsof ended SIGTERM/killed, then unanchored-private-group; cleanupQualified=false despite survivors=[]. Initial inspection versus bounded recheck was not identified, so no cause or deadline defect is claimed. Receipt 11944321687 physically verified H2/Core candidate. Configured/prepared did not execute.
fd2 full, Windows stable 116208559706	Fresh/configured and eight prepared prefixes passed; only native-mcp-controls failed its 180-second completion oracle. The generated BenchmarkTools suite passed internally (04:11:24–04:11:44 UTC), but actual Chat/MCP projection did not complete. At 04:15:23.107 the controlled provider recorded ERR_ASSERTION: No saved measurement reaches the server without explicit Attach; the final failure was recorded at 04:15:27.423. The fixture clears its mutable Attach oracle in finally before waiting for idle, so a pending request can trigger this secondary assertion; the initial 180-second cause is undetermined. Eight other tail groups passed and no global limit was reached. Receipt 11948941000 had an explicitly exceptional second read of the omitted provider error. No MCP cleanup success is inferred from later Pluto cleanup.

F2 full is terminal with four GitHub successes and three failures. Three passing receipts were physically inspected; Linux stable is API-only because its 150,764,181-byte receipt was not downloaded. The fd2 full campaign is entirely terminal: five GitHub successes (four physically verified, Linux stable artifact 11946796825 metadata only/nonphysical) and two failures retained above. Both fd2 CIs passed all four jobs, but their terminal Windows job logs were unavailable to the bounded read, so no uninspected TAP counts or first---version timings are claimed. None of these older passes qualifies 5e8.

Earlier H/068a qualification history, failures and scoped limits

Historical H evidence, superseded by the packaged README correction:

The superseded H archive was packaged once on Linux from product commit 7937002c262b8d9b12014e1e13a9e30af295ed0b / tree 40853c0cce3c6416b63bd68dfbf64878037a73ee in origin run 38643783540, artifact 11930759442, named native-candidate-7937002c262b8d9b12014e1e13a9e30af295ed0b. The actual VSIX is 1,079,004 bytes, SHA256 b4cc9352bb408441a6550cbc35ef79139f8f70eff3f727ca6cd30b1f6f4ba74e. Its complete ZIP (1,079,770 bytes, SHA256 4a5bc8f7df807928127b35e5f3e3a05e9f2d327fb99cd4c0fee83318161f2115), checksum, provenance and packaged discovery implementation were inspected. Workflow commit a6e9a40ea506248f92a2e4c8c9431dc223550e5a changes only the frozen-origin anchor after the product commit. The former 068a archive and published 1.0.2 archive remain immutable historical origins.

• H's push CI 38645066703 and PR CI 38645073740 passed all four jobs. Its producer focused Linux test passed and its physical receipt was inspected. Its x64 bootstrap passed 25/25 cases across five hosts and ARM bootstrap passed 20/20 cases across four lanes; all nine physical receipts were inspected. These cover fresh installation, not a native Pluto upgrade.
• H's seven-lane full campaign 38645073798 is terminal with four GitHub successes and three Intel failures. Measure retains its configured Testing visibility and prepared designer-card failures; investigate retains a Testing visibility failure and a JET analyzer timeout. Agentic retains its failed prepared Testing visibility prefix even though fresh/configured and its five later groups passed. No passing tail compensates for a failed phase or partition. The large Linux stable passing receipt was deliberately not downloaded, so its result remains API-only. H is not the final publication archive, and none of its passes qualifies H2.

Historical evidence for the previous 068a origin follows; it does not qualify the new archive:

• TypeScript compilation and the initial targeted product checks passed. Earlier qualification-only changes passed 29/29 targeted Node checks, including the independent historical case inventory, the actual host/suite dispatch prefix, workflow matrix routing, discovery receipts and strict ownership regressions. Both native drivers passed syntax checks. These do not replace native execution.
• Fresh General bootstrap: five x64 hosts, 25/25 native cases passed. Native ARM bootstrap: four lanes, 20/20 passed. All nine receipts and archive digests were inspected, including real controller/Pluto Core versions, source trees and stopped listeners. These cover fresh installations; upgrade unit fixtures prove orchestration, not a native upgrade.
• Both push CI and PR CI passed all four jobs on preceding head 1059d1753fd2fe7e61a422374f59df3f8097aeb3.
• Historical failures remain failures. Run 38519958480 retains Windows stable job 115708720295 failing its inventory assertion and recording cleanupQualified=false. Run 38526097661 passed four full hosts, whose receipts, ZIP digests and actual retained VSIX bytes were inspected. Intel job 115726777065 failed the configured missing-target and prepared test-item visibility checks, then hit the real 130-minute prepared bound during progressing MCP work. Its productCleanupQualified=false is preserved. Neither failure is reclassified as a pass.
• PR CI 38548507330 on e782a33 retains its Windows test failure, job 115800429188: the expected authentication rejection returned, but its duration was 12.710 seconds against the fixture’s unchanged 10-second bound. The isolated fixture passed its three targeted Git tests locally; this does not establish the cause of that runner delay. The native campaigns on e782a33 remain historical and are not reclassified.
• Push CI 38550223095 on adebd5e retains Windows job 115806081289 failing the owned-Job cleanup retry test during start(), with an MCP request timeout before the injected stop failure. The log does not identify which startup RPC expired. The other campaigns on that source remain historical and are not cancelled or reclassified.
• The Windows fixture change passed syntax validation and three existing neutral transport/refusal tests locally: 3 passed, none failed or skipped. The Windows test itself was not executed on Linux, so these checks do not establish Windows success.
• On 09439a0, both push CI and PR CI passed all four jobs. The modified Windows MCP cleanup test passed four real executions across their test/package steps; exact logs show living owner/server before original RPCs, unchanged 5-second RPC timers and intact retry/foreign-process/cleanup assertions.
• Native run 38552737649 on 09439a0 has a failed required Intel investigate partition, job 115824654355. Its receipt 11902626538 retains all seven defects: fresh initialization exceeding its 600-second oracle and consequently blocked Pluto; configured Pluto dashboard absent and Testing item absent after 120 seconds; prepared designer frame absent, Testing item absent after 120 seconds and diagnosis failing its strict analyzer assertion. The real diagnosis report records JET and AllocCheck timeouts at 180.063 and 180.046 seconds; seven other analyzers completed. No global 130-minute bound was reached. Cache/import cleanup and the separate deadline/cancel checks passed, which does not qualify global cleanup of the failed scenario or prove the cause of its delays.
• The same historical 09439a0 native run also retains Windows stable job 115824654409 failing its global prepared-phase bound, receipt 11903743819. Fresh and configured passed. Prepared started at 12:38:57.665 UTC and its 130-minute limit expired at 14:48:57.613 UTC, while native-pluto-controls was still progressing. All eight shared-prefix cases and the first seven control groups through MCP had passed without an earlier assertion failure. The last observed Pluto Close check at 14:48:34.698 records its worker stopped and owned files removed, but Pluto has no final verdict and discovery/sync was not executed. The receipt explicitly records hostStopRequested=true and productCleanupQualified=false; individual cleanup proofs do not qualify the interrupted campaign. The complete ZIP digest and retained physical VSIX/provenance were verified. This failure is not reclassified as a pass, and the report does not establish a deadlock or other cause.
• The same native run has inspected passing receipts for Linux 1.96 (11896813368), Linux stable (11899976588), Windows 1.96 (11899347741) and Intel measure (11900548819). Complete ZIP digests, retained physical VSIX bytes and original provenance were verified in memory. Intel measure covers its eight shared-prefix cases and two measure groups; it does not cover the failed investigate partition or the agentic partition. Intel agentic finished with the distinct failed prefix described below; Windows stable remains the separate historical failure described above. All prerequisite/unverified controls and callback/ownership limits remain explicit; no historical failure is reclassified as a pass.
• The new observation delta passed syntax checks and 29/29 targeted Node tests, including exact result/error delegation, ownership/privacy refusal, coalescence, observer abort/drain, late observation rejection and executable revalidation after state collection. These unit checks do not qualify new native execution.
• Historical 09439a0 Intel agentic job 115824654647 also failed, receipt 11906066661 (3,187,148 bytes, ZIP SHA256 77eaa472c2971ba8a73e3b12955c49562d1e479a58d4ca7cdbbb39624cd98d22). Fresh and configured passed. Prepared retained two failed prefix cases: Studio #cards was absent at 13:44:15.111 UTC after the case began at 13:42:13.859, with the plan CLI still visible; the Testing item remained absent after its 120-second oracle at 13:50:47.843. All five agentic control groups subsequently passed, including Pluto and discovery/sync at 15:06:45.133, but that does not qualify the failed phase or the partition union. No global phase bound was reached. The original VSIX/provenance and cache/import cleanup were verified; individual Pluto cleanup proofs do not qualify every process. The cause of the two prefix delays remains unknown.
• On b96d58a, push CI 38583585491 failed Windows job 115919549229: the authenticated Codex fixture test timed out during its first inspectCodex --version. The actual PowerShell owner was killed after 10,015 ms with zero stdout/stderr bytes, before any HTTP RPC. This does not identify whether script entry, launcher preparation or another step was delayed. PR CI 38583597681 separately passed all four jobs; its success does not reclassify the push failure.
• On the same b96d58a source, native run 38583597627 Linux 1.96 job 115920828600 failed before prepared host launch: the VS Code CLI rejected the downloaded julialang.language-julia extension with “End of central directory record signature not found. Either not a zip file, or file is truncated.” Fresh and configured had passed; prepared has no receipt and was not executed. Receipt 11904589217 is 1,347,521 bytes with ZIP SHA256 d6cf7f97fba43fd5ae5de09929444037c54c69eae8feb167337f997174fd80e2; the original physical PerfChecker VSIX/provenance and controller cache/import cleanup were verified. This dependency archive failure remains a failure; it is not evidence of a changed PerfChecker package or a passing prepared campaign. Other native lanes retain their separately recorded conclusions.
• The final test-only diagnostic now records private fixture JavaScript entry per sacrificial CLI invocation, alongside the real owner PID, safe spawn-error codes, timestamps and byte counts. Entry files are exclusive, bounded and tied to their nonce and file identity; stale, foreign, linked or arbitrary records are refused. The marker proves historical script entry only and is not used for ownership, signals or readiness. Diagnostic errors do not replace the original operation result. There is no warmup, retry, new installation precondition or change to the product’s 10-second preflight deadline. The new direct-Node diagnostic test waits for physical close before cleanup and explicitly rejects unobserved close within a separate bounded drain. Syntax and five targeted Node tests passed locally, with no failures or skips; these include three diagnostic tests and the existing discovery/authentication contracts, not native Windows execution. This change collects missing evidence and does not establish or fix the cause of the earlier delay.
• Historical native run 38583597627 on b96 is terminal with three passing and four failing lanes. Besides the dependency-archive failure above, Intel investigate receipt 11912919212 retains four defects: first Run returned no suite evidence after initialization had completed in 518.114 seconds, its dependent fresh Pluto case was blocked, and configured/prepared Testing visibility each exceeded 120 seconds. Intel measure receipt 11913747367 retains six defects: first Run/reload evidence failed after initialization completed in 599.163 seconds, dependent fresh Pluto was blocked, configured Testing exceeded 120 seconds, and prepared designer cards, Testing visibility and the Julia REPL evidence failed. Those initialization durations were below their 600-second oracle; they are not themselves initialization timeouts. Intel agentic receipt 11918571319 retains configured/prepared Testing visibility failures and a separate MCP Suite projection/request completion failure (diagnostic 17:36:45 UTC); four other agentic groups passed. No global phase bound was reached in those Intel jobs. Controller cache/import cleanup was qualified, which does not qualify every process of a failed campaign. Linux stable, Windows 1.96 and Windows stable passed with their own inspected receipts and explicit callback/ownership limits.
• Historical e485 push CI 38589434780 and PR CI 38589444994 passed all four jobs each. Its native run 38589445608 is terminal with three passing and four failing lanes: Linux 1.96, Linux stable and Windows 1.96 passed; all three Intel partitions and Windows stable failed the Testing visibility prefix. Intel investigate failed configured and prepared visibility; Intel measure failed prepared visibility; Intel agentic failed configured and prepared visibility, although all five agentic groups subsequently passed. Windows stable failed prepared visibility, although its other seven prefixes and all nine groups passed. Each failed visibility check retained its 120-second oracle and empty Testing tree; no global phase bound was reached. Physical 068a bytes/provenance and receipt ZIP digests were verified. Preflight cleanup and individual Pluto cleanup remain scoped proofs, with callback and second-cancellation limits preserved. None reclassifies a failed phase or partition union.
• The separate e485 Intel focused diagnostic 38639042070 also failed Testing visibility after 120 seconds, without Aqua/SnoopCompile or the fresh/configured/full phases. Its actual VS Code was 1.142.0. Case start was 21:55:25.746 UTC, failure diagnostic 21:57:29.608; preflight had discovered the correct item in 99.148 seconds without executing or preparing the target. Receipt 11931378917 (1,228,816 bytes, ZIP SHA256 405e909193fa13493dc20cf0eec943499b475c90e9034568b7f2fe4724fe6632) retained the exact physical 068a origin. Two direct listing children remained present at the deadline and no listing result was observed in the bounded pre-close snapshots. This reproduces the symptom in an isolated environment; it does not identify loading, JIT, parsing or shutdown as the cause, and does not qualify the new archive.