15 lines
1.8 KiB
JSON
15 lines
1.8 KiB
JSON
{
|
|
"actor": "research-abi-independent",
|
|
"basis": "aaddeb76f74bbd1a43671f882770ec84a7d690b4060d060ef1a68f918e8edbb0",
|
|
"contract": "research-completion-abi",
|
|
"decision": "d-b51f3f76803e852ed250846a",
|
|
"evidence_digest": "e5073be0f1187ee65c203eb9ad20adc64d18ff2d3314d554f58bd8487dafa931",
|
|
"explanation": "Scoped integrated PASS over the current two-record integrated evidence array. Fresh handoff reproduction executed 22 nonempty static windows plus strict archived save parse/checksum: 24 positive subprocesses, empty stderr, no skips; manifest matches the integrated artifact except session and state is byte-identical. Seven narrow/wide boundaries reproduce complete ret 4/ret 8 bytes. Held-out direct-PE ablation confirmed each exact return and rejected all minus-one/plus-one shifted starts; static inspection exposed distinct 0x2c/0x74 strides, deep-copy calls, three guarded long-string deletes, description inequality and three unordered-or-unequal float branches. Actual state inspection found EvNxID 4, turn-3 count 2, event ID 3 fields/defaults, exact 609080-byte rebuild, and a 2503-byte RNG leaf with digest 0978fdf34ff7962f76c2de810dc93e0a. Integrated source/input/binary/outcome hashes remain bound and campaign tests/validation pass. Scope is independent static reproduction plus one archived state only: no live allocator safety, original differential, replacement or replay acceptance; runtime branch residuals remain. Historical failed provenance/description predictions remain failures resolved by Astra.",
|
|
"model": "openai/gpt-5.6-sol",
|
|
"role": "verifier",
|
|
"schema": "sots-verdict/1",
|
|
"session": "run-d94d4516b9d898793318805e",
|
|
"source_bindings_digest": "bea3fe491275d544850c432ddcae2473a8fb226b5c97ec8151d3b75f7add7f73",
|
|
"timestamp": "2026-09-10T01:39:48.486681+00:00",
|
|
"verdict": "invalidated"
|
|
}
|