Runtime bench: qualify cold build setup timing and interrupted compiler cleanup

Problem and original failure

The original Linux FIFO command in CI 39819353675 failed an unrelated 20-second setup hook at 20,000.86 ms, before the test body or full suite qualified. A dangling process was killed and later cleanup reported compiler descendants. Each real configure, compile and CTest command already permits 120 seconds. The reason the original compilation was slow remains unknown.

The terminal logs remain retained in evidence 401. The original completed check-job log is 34,258 bytes, SHA-256 99f3015b0d36a485cbb6a11b0ac70c90bdc6f7208813474ee5a3921ac08e03cf.

Implemented in PR #69, published with PR #70

Commit 6bdf28b544fb821f9cab54088c7178196b3674c1 awaits runtime/sensor build initialization before test timers start and names failed build stages. Every production command and runtime/test-body deadline remains unchanged. A fixture-only aggregate supervisor derives its budget from the three separate commands, settlement, the selected test's existing deadline and startup allowance. Concurrent output remains bounded to 2 MB per stream. Actual parent and detached tool groups are observed and cleaned.

Retained causal campaigns:

• 406/407: the unchanged original hook fails a real 21-second configure at 20,001.98 ms with an observed CMake orphan. The initialization correction passes the same controlled boundary in 37.68 seconds, followed by the unchanged native boot case. An initial controller text-classifier error was corrected from the same retained bytes, without rerunning that subject.
• 414/415/419: an initial aggregate attempt lost its receipt after a cleanup probe error; it remains retained. Reporting-only correction then establishes the old outer 120-second cutoff after actual configure and compile exited0 in 46.185 and 45.134 seconds. The derived aggregate supervisor completes all three real stages in 139.18 seconds, each individually below 120 seconds, with no owned groups before controller cleanup.
• 416/417/418: the first inherited-pipe regression failed. Nested actual errors identified a second group kill after the first succeeded. Idempotent termination passes all five native process cases.
• 420/421: making the early receipt path a directory and interrupting the actual test parent left an actual detached CMake group before the reporting correction. The corrected qualifier retains EISDIR and settles the group before external cleanup. Its explicit CI fault mode requires that refusal and an actual live tool group, rejects unexpected cleanup errors, and cannot count as ordinary build success.

Final local validation 422: Pickier, application/runtime types, fresh cold build and reporting-fault mode pass. Relevant tests: 51 passed, one Linux-only skipped, zero failed, 347 assertions. Independent Spec and Standards reviews are clean.

Actual Linux evidence

Current-head CI 39838472116, Bun 1.4.3 on Linux, has passed both cold qualification modes and the separate original kernel-confirmed FIFO command.

• Cold-build artifact: actual configure 24.258 seconds, compile 11.307 seconds and CTest 2.720 seconds, all exit 0; selected FIFO test exits0. No parent/tool groups require cleanup or remain afterward. Downloaded once; digest 64bf5fae8cd44c2f5b08d19be760b0813336fd0ccc6d3f71f8208a86dc3aebcb matches.
• Reporting-refusal artifact: actual subject exit 137, EISDIR retained, one live detached tool group before qualifier cleanup, none afterward, no cleanup errors. Downloaded once; digest 16d1b119796af636f3d21b74035a18e922cbf2820ed691f3db13fe33034cc923 matches.

The workflow is now terminal. Its entire check job passed, including the full Buddy suite: 1,515 passed, 41 skipped, zero failed, 1,556 tests across 196 files in 3,528.73 seconds. All subsequent standalone CLI gates passed. The overall workflow failed only in the two separate browser jobs, retained under #44 and #64. Full terminal log was collected once: 3,234,312 bytes, SHA-256 afb3c4617d50d464b09aab76f84bc3e6796acfe40ef08f2b4dd62530cf1fbdbf. The original workflow was not restarted. The combined source in PR #70 has since passed its complete exact-head workflow, as recorded below. This later Linux pass does not explain the retained original local #67/#68 failures.

Separate failed gates

Original local full suite 410: exit 1, 1,431 passed, 27 skipped, 82 failed, 474,778 assertions, 194 files, 2,770.67 seconds. Source/Bun pins stayed unchanged. Complete log: 428,106 bytes, SHA-256 daf568d9b4373f1b4e0d589c968081df09e9c1257b051b704aa6401257c88995. Issue #68 retains every terminal entry and missing inline diagnostics. Observed disk exhaustion is not an established cause. The earlier invalid-CLI timeout remains separately unresolved in #67; later focused passes do not explain it.

Acceptance

• Retain original compiler/process evidence and reproduce real hook/aggregate preemption before changing orchestration.
• Qualify the corrected cold-build contract and original Linux FIFO case, retaining all original failures.
• Prove actual Linux parent/tool settlement in normal and interrupted reporting-failure modes, with owned source/evidence retained.
• Verify a complete current-head passing CI workflow and full existing suite before main publication.

No board, sensor, actuator or physical timing acceptance is claimed. The original Linux compilation slowness and distinct failures remain unresolved.

Qualified combined source published to main

PR #70 adds bounded browser-launch evidence to this runtime correction. Its exact head 1ef2f6a86ed77ae9e7924d4fd3f76ebd19a8648e passed CI 39843840771: all five jobs succeeded, including cold build, reporting-fault cleanup, original FIFO, both actual rendered browser campaigns, firmware and namespace. Full Buddy suite: 1,519 passed, 41 skipped, zero failed, 489,307 assertions, 1,560 tests across 198 files in 3,210.49 seconds. All subsequent standalone CLI gates passed.

That qualified combined commit was published to origin/main by normal fast-forward on October 22, 2026 as Chris Breuer, with no coauthors. PRs #69 and #70 are merged. The complete terminal log was collected once: 3,081,463 bytes, SHA-256 cc486891ca288870412844dd3b48a89cd655db33d1b4984a96546be78872a88b. The separate main push qualification has now also passed, as recorded below.

The retained original compilation slowness, macOS browser exits and earlier local #67/#68 failures remain unexplained. This passing qualification does not establish their causes or replace any failed evidence.

Main publication qualification completed

Main push CI 39852572090 qualified the published exact source 1ef2f6a86ed77ae9e7924d4fd3f76ebd19a8648e: all five jobs succeeded. This includes the cold-build and interrupted-reporting cases, original Linux FIFO boundary, both real browser campaigns, firmware and namespace qualification. The full Buddy suite recorded 1,519 passed, 41 skipped, zero failed, 489,307 assertions and 1,560 tests across 198 files in 3,556.52 seconds; all subsequent copied standalone CLI gates passed.

The full terminal raw log was collected once: 3,057,579 bytes, SHA-256 65d7e8175efdcf55bb09f200c99006c0d58f7acb7b1a83928841fbd4dcc50f12. All four acceptance items above are fulfilled for this orchestration correction, so this qualification ticket can close. The original slow compilation is not explained by this passing run, and distinct #44/#60/#63/#64/#67/#68 failures retain their own evidence and open investigation. The later TCP branch browser-startup failure is separately tracked in #72; it does not replace this exact-main receipt.