3.5 KiB
Row 299 gate run 1
Grant sage-3268. Candidate ffa93b054f, base 80793d824, manifest 277ace0b4274970d5bd4c05cf8fe99aa1e7ebc469a719e4cefcb334a75d3865d. Run 2026-10-23T07:26:52Z to 2026-10-23T07:50:16Z, 1404 seconds by UTC timestamps. No carried result or retry.
The preflight passed. All three case1 files and all 59 CLI files ran whole. Unmutated total: 1076 pass, 0 fail, 0 skip, 0 cancelled, 0 todo. Mutants: 6/6 killed by named assertions, no external timeout. All six restores match their original SHA256 and have clean git status. Sweep: 0 gate residuals; four MCP harness processes predate the gate, with start times in sweep.json.
| Stage | Pass | Fail | Skip | Wall seconds | Start load |
|---|---|---|---|---|---|
| case1-whole-1 | 53 | 0 | 0 | 239.134 | 1.05 |
| case1-whole-2 | 53 | 0 | 0 | 240.044 | 3.41 |
| case1-whole-3 | 53 | 0 | 0 | 235.011 | 2.08 |
| cli-whole | 917 | 0 | 0 | 631.848 | 1.91 |
CLI sampled 1-minute load range: 1.54 to 7.17. Ambient >=4 was a target, not a wait condition. No stress process. Each of the three case1 start loads and the CLI start/load samples are recorded. The direct readings below come from the test diagnostics, with fixed inner [1.25,2.75) and outer [3.25,4.75) bands.
| Stage | Inner exec | Inner hop | Outer exec | Outer hop |
|---|---|---|---|---|
| case1-whole-1 | 2.25 | 2.27 | 4 | 4 |
| case1-whole-2 | 2.29 | 2.28 | 4 | 4.01 |
| case1-whole-3 | 2.25 | 2.26 | 4 | 4 |
| cli-whole | 2.27 | 2.27 | 4.01 | 4.01 |
| Mutant | Named killer | Assertion | Direct interval | Wall seconds |
|---|---|---|---|---|
| M1 | row253 C4 P10 inner reason holds under both sudo models | case1-accept.test.mjs:450 lower band | 1.23 | 4.774 |
| M2 | row253 C4 P10 inner reason holds under both sudo models | case1-accept.test.mjs:450 upper band | 3.24 | 6.768 |
| M3 | row253 C4 P10 outer reason holds under both sudo models | case1-accept.test.mjs:452 outer lower band | 3.01 | 6.618 |
| M4 | row253 C4 P10 outer reason holds under both sudo models | case1-accept.test.mjs:445 exact probe-outer-timeout reason | not reached | 9.023 |
| M5 | row253 C4 P10 inner reason holds under both sudo models | case1-accept.test.mjs:448 required complete pair (doesNotThrow) | not reached | 5.816 |
| M6 | row299 P10 observer records are complete and ordered | case1-accept.test.mjs:418 equal readings must refuse (throws) | not reached | 0.164 |
M1 measured 1.23 s, below 1.25 by only 0.02 s. This is the anticipated startup-overhead risk. It died at the intended lower-band assertion; no tuning or retry was used.
M4 classification follows sage-3271, sage-3273 and mos-1426. The unchanged 7000 ms fixture guard returned; the named outer test failed with ERR_ASSERTION / match at line 445 because P10 reported reason=probe-inner-command instead of exact reason=probe-outer-timeout. It returned rc 1 in 9.023 s including the baseline. No external command timeout or cleanup-only failure was counted. Raw evidence: M4.log, M4.command.json, results.json guard_return_receipt and clean restore.
sage-3274 reported a read-only canonical queue polling overlap for receipts window 07:20–07:26Z. It wrote nothing and touched no worktree or test path. This is accepted ambient load, not a rerun condition; ambient-overlap.json keeps the report.
All raw logs, command headers with timeout -k 10, TMPDIRs, per-stage load samples, controller, results, and restores are in this directory. Every run used a fresh short TMPDIR directly under /mnt/storage/scratch/tmp. No source or assertion changed during the gate.