Signature
No abort, no panic, no lockup marker. SCHED_STRAND_ORACLE prints normally through the run
(stranded=0 every sample, worst_nonprogress_ms=0), and BLOCK_EINTR_ORACLE:PASS prints, but the
serial goes silent after [heartbeat] tid=1203 uptime_ms=19959 — roughly 21s of wall-clock heartbeat
progress logged, then nothing — and the boot is truncated at the gate's per-boot timeout before
[STRAND_INJECT_ORACLE: (or, later, [CENSUS_WIDEN_ORACLE:) is ever printed. The service-sequence
gate's classifier already has an explicit clause for exactly this (run-aarch64-service-sequence-gate.sh,
"scheduler strand oracle marker absent" / "census widening mutation oracle marker absent"), so the
boot correctly reds as UNATTRIBUTED rather than being silently counted GREEN — it was never absorbed,
but it also was never filed with its own number until now.
No P5b/init-group markers (INIT_GROUP_WALK, INIT_GROUP_REFUSAL) appear either, so whatever the
boot is doing in its last ~21s of wall clock, it never reaches the init-group quiesce phase that a
healthy boot on this profile reaches by then.
Rate
1 occurrence in this branch's own 100-boot/profile starved gate (fix/609-early-kthread-dispatch,
cortex-a72 profile, boot 7). Not otherwise characterized; no powered control.
Preserved serial
scratchpad/t3g/preserved-serials/starved100-cortex-a72-UNATTR-strand-serial-7.txt (T3-G session,
2026-08-21), 796 lines total; last non-heartbeat progress at uptime_ms=19959.
Context
Filed as part of ruling R34 (in-repo record + gate correction for the #609 falsification round). Not
established as related to #609, #612, or the two new abort signatures filed alongside it — recorded
separately because a starved 10-hog nice -n 19 profile is a distinct stress shape from #609's clean
gate and from #612/the new DATA_ABORT/INSTRUCTION_ABORT shapes' fault mechanism.
Signature
No abort, no panic, no lockup marker.
SCHED_STRAND_ORACLEprints normally through the run(
stranded=0every sample,worst_nonprogress_ms=0), andBLOCK_EINTR_ORACLE:PASSprints, but theserial goes silent after
[heartbeat] tid=1203 uptime_ms=19959— roughly 21s of wall-clock heartbeatprogress logged, then nothing — and the boot is truncated at the gate's per-boot timeout before
[STRAND_INJECT_ORACLE:(or, later,[CENSUS_WIDEN_ORACLE:) is ever printed. The service-sequencegate's classifier already has an explicit clause for exactly this (
run-aarch64-service-sequence-gate.sh,"scheduler strand oracle marker absent" / "census widening mutation oracle marker absent"), so the
boot correctly reds as UNATTRIBUTED rather than being silently counted GREEN — it was never absorbed,
but it also was never filed with its own number until now.
No P5b/init-group markers (
INIT_GROUP_WALK,INIT_GROUP_REFUSAL) appear either, so whatever theboot is doing in its last ~21s of wall clock, it never reaches the init-group quiesce phase that a
healthy boot on this profile reaches by then.
Rate
1 occurrence in this branch's own 100-boot/profile starved gate (
fix/609-early-kthread-dispatch,cortex-a72profile, boot 7). Not otherwise characterized; no powered control.Preserved serial
scratchpad/t3g/preserved-serials/starved100-cortex-a72-UNATTR-strand-serial-7.txt(T3-G session,2026-08-21), 796 lines total; last non-heartbeat progress at
uptime_ms=19959.Context
Filed as part of ruling R34 (in-repo record + gate correction for the #609 falsification round). Not
established as related to #609, #612, or the two new abort signatures filed alongside it — recorded
separately because a starved 10-hog
nice -n 19profile is a distinct stress shape from #609's cleangate and from #612/the new DATA_ABORT/INSTRUCTION_ABORT shapes' fault mechanism.