Observed failure
The pinned Bun 1.4.2 macOS ARM64 runtime/sensor regression run on October 22, 2026 passed 45 cases, skipped the Linux-only FIFO case, and failed the unchanged runtime test invalid CLI parameters cannot silently execute a default bench run. The parent test reached its default five-second deadline at 5,020.21 ms, Bun reported one killed dangling process, and the subsequently observed child exit was 143 rather than the expected validation exit1. The command was bun bin/drone.ts runtime:bench --ticks nan.
The failure is retained in runtime-cold-build-checks-409/runtime-sensors.log: 7,218 bytes, SHA-256 c5d60f0b344a44456d10cb920b155b26c51546bf4e6dcec0c7dd76738d1df1b0. Another sensor CLI case in that same run took 28.56 seconds and passed under its existing budget. The log does not establish why startup was slow, which import or operation consumed the time, or whether an output-pipe dependency contributed. No passing retry replaces the result.
This is distinct from issue #65's demonstrated cold CMake/setup-hook mismatch: the native build and controlled 21-second configure qualification passed, and the runtime CLI assertion itself was unchanged.
Acceptance
- Reproduce the actual native standalone CLI validation boundary and retain its parent/child runtime, lifecycle, bounded output and operation timings.
- Diagnose why validation did not settle within the existing test budget. Distinguish imports, startup, parsing/validation, output and cleanup instead of attributing a cause from elapsed time alone.
- Implement a causal fix while preserving rejection of invalid arguments and proving that an invalid value cannot start a default native bench run.
- Verify cleanup of actual owned processes and run the relevant full regression suite. A passing warm retry or arbitrary deadline increase is insufficient.
No hardware or physical runtime timing is qualified by this test result.