Authored strict paired-IRQ ARM guests — original qualification
Corrected final-head guest gate
The generated ignored-test inventory required a correction after the first guest
pass. Run38158310794,
attempt1, then passed all ten cases at final head
b4c986f361079f9a6ffb5eb9446f67ee97252733. All recorded ELF identities and
case results match the original pass below; source/policy/bounds are unchanged.
This was automatic CI after a required gate fix, not a diagnostic replay for
reading evidence. The original failed default-member index gate remains in
run38153786840.
The sole artifact11709799464
is authored-paired-irq-guest-results, 16,642 bytes, ZIP SHA256
4381f1d0ac7805eb99ed7add47016fac4cbabac16bf5ae55bce1c5b80cd687ff.
The unchanged final-head result is 4,291 bytes, SHA256
9a319216f6546080802e3281a23c19105e7661e80186769ce2a73014b28d4117.
The same separate no-producer-import packet/source audit passed on its official
identity, six source files, snapshots and both execution logs. Official Git
metadata confirms qualification merge 0621e95366b2f55a1369f373f96c3ec639eaa956,
tree 539beada886c0fd51425842ac03061a65296e638, parents
78ec4af563b70466d8e33c79193c0e270d513230 and final head b4c986f....
All 20 enabled final-head checks passed, including both aggregate gates, with
only the four declared optional jobs skipped. PR198
merged as 46aed9ba8232b3cceaec8a33e15230f9d973d486; the predicted, reviewed,
actual-guest and landed trees all equal 539beada886c0fd51425842ac03061a65296e638.
Landed parents are the same base and final head listed above. This remains a
bounded authored-control result. Both original packets/results are preserved.
Earlier unchanged original pass
Run38153786907,
attempt1, passed at head e7589656998d3a08cd59670ff09a4f3214310d2e.
Qualification merge f074b54d48cf4381b6832aab7ce5abc4aaea45d3 has tree
3c2c12a69ed4abdab301b5528eeb5cd8d462e6a6 and parents
78ec4af563b70466d8e33c79193c0e270d513230 and that tested head.
The earlier pass alone did not satisfy the broader merge gate; that gate was
subsequently completed at the corrected final head described above.
The guest consumed CODAL source 3c66f0d27c865192a2b22217e71c1f88f2b3fe91,
after all eight enabled source checks passed. CODAL PR10
merged as c2274c7ed397787820c92a91438c4c0a902612b8, with reviewed/landed tree
8ed57db57606e4ea644609e33c6eaf385beb623b equal. That merge did not select
these opt-in components in the default runtime CMake build.
Original artifact and separate audit
The sole official artifact11707374717
is authored-paired-irq-guest-results, a 16,643-byte ZIP, SHA256
374e6bd22a8c7eec1de729a9e14e63f837e79bb5d4309f723b360aac2570a83f.
The unchanged original result is 4,291 bytes, SHA256
ff7a67aabfcf7410da46020fc2b067f914d8808a6d05e9a915b663b62b444d4e.
The original contains both logs, source snapshots, maps and symbol reports;
ELFs are deliberately not bundled. This durable result is not a new execution.
A separate standard-library audit without importing the producer checked official run/head/attempt/success and artifact identity/size, bounded unique ZIP members, qualification merge/tree/parents, equal before/after snapshots, six actual CODAL source hashes, both log hashes, all ten result rows and identical recorded ELF identities across profiles. It passed. Official Git commit metadata also confirmed the qualification merge's tree and parents. This is packet/source audit, not replay, independent-agent testimony, full compiler-component closure or independent ELF rehashing from absent binaries.
Ten observed cases
Both default and event-scheduler profiles produced the same following results:
| Authored case | Status | Fault | Guest steps, each profile | Terminal refusal instructions |
|---|---|---|---|---|
| Positive nested/restoration/IRQ | 0x600d |
0 | 3791 | 0 |
| Underflow | 0x201 |
3 | 369 | 64 |
| Changed diagnostic identity | 0x202 |
5 | 624 | 64 |
| Depth limit | 0x203 |
4 | 16644 | 64 |
| Lost masking | 0x204 |
6 | 638 | 64 |
The tested harness checks vectors, startup BSS/data initialization, mask restoration, one positive stage2/mask0/exception16 IRQ and thread return. Refusals require unchanged exact diagnostics, masked thread mode and zero IRQ witnesses. Only the actual compiled mask-restore helper is allowed during transition, for at most 32 instructions; 64 consecutive terminal observations reset across it, with a 192-observation transition bound. Caller escape, unrelated code, stuck helper and stopped PC refuse. Four pure progress controls ran before the actual guest cases.
This is bounded liveness, not indefinite behavior. The diagnostic identity is not real fiber switching. Additional guest-level policy mutants and full recording-entry-point adversaries remain open. Real Timer/Fiber/Heap and startup/panic/scheduler integration, compatible core/board inputs, a named no-ASF build, required USB/WebUSB, component notices, peripheral IRQ/DMA, native Arcade boot, debugger/package adoption and active RTx remain unqualified. Continue through the interface contract and replacement lanes.