Spaces:
Running
Running
File size: 7,754 Bytes
a0f8a7d 542b994 a0f8a7d | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 | {
"schema_version": "szl.receipt-agent-admission-manifest.v1",
"program_id": "szl-forge-1.5b-receipt-agent",
"evaluated_at": "2026-07-13T07:22:49Z",
"promotion_decision": "BLOCKED",
"gate_policy": "ALL_REQUIRED_GATES_MUST_PASS_ZERO_CATASTROPHIC_ERRORS",
"gates": [
{
"gate_id": "RA-G01",
"name": "Exact base identity",
"state": "PASS",
"required_for_promotion": true,
"exact_requirement": "The base repository and immutable revision must match the local candidate manifest.",
"observed": "unsloth/Qwen2.5-1.5B-Instruct-bnb-4bit at d2f2dd02b071701d5100a04a7a49d6fb0bd305b7.",
"evidence": [
"model_release/m1/operational-manifest.json",
"model_release/m1/candidate-manifest.json"
]
},
{
"gate_id": "RA-G02",
"name": "Exact existing-adapter identity",
"state": "PASS",
"required_for_promotion": true,
"exact_requirement": "The existing adapter weight digest must equal the measured local artifact digest.",
"observed": "SHA-256 682e2f0ea480d47c284b9de12c2e3d2d5170934c065e82fc375e3f069b4730ac; this identifies the existing SFT adapter, not a new ReceiptAgent model.",
"evidence": [
"model_release/m1/operational-manifest.json",
"model_release/m1/candidate-manifest.json"
]
},
{
"gate_id": "RA-G03",
"name": "Owned Forge curriculum rights and governance",
"state": "PASS",
"required_for_promotion": true,
"exact_requirement": "Every row admitted to this candidate must be project-authored or explicitly licensed, schema-validated, content-hashed, and assigned to an immutable split.",
"observed": "The Forge curriculum admits 30 project-authored schema-generated training rows and 8 frozen evaluation rows. The historical 167-row adapter corpus is excluded.",
"evidence": [
"model_release/szl-forge/generated/curriculum-manifest.json",
"model_release/szl-forge/training-contract.json"
]
},
{
"gate_id": "RA-G04",
"name": "Brain raw-node quarantine boundary",
"state": "PASS",
"required_for_promotion": true,
"exact_requirement": "All raw Brain nodes must remain outside gradients until item-level license, provenance, revision, freshness, deduplication, and split decisions exist.",
"observed": "9,464 of 9,464 current raw nodes are training-quarantined; zero raw-node text rows are admitted to train.",
"evidence": [
"model_release/m1/corpus-ingestion-manifest.json",
"model_release/szl-forge/training-contract.json"
]
},
{
"gate_id": "RA-G05",
"name": "Formula holdout and proof-status isolation",
"state": "PASS",
"required_for_promotion": true,
"exact_requirement": "Formula IDs must be namespace-scoped, proof transfer must require an explicit semantic binding, and no holdout formula may enter training.",
"observed": "The current crosswalk has 146 formula records; two SZL-Lake evidence records produce a 148-row holdout tranche with zero train rows. F1-F23 namespace collisions block implicit proof transfer.",
"evidence": [
"research/formula-training-admission/admission-manifest.json",
"research/formula-training-admission/formula-id-crosswalk.json"
]
},
{
"gate_id": "RA-G06",
"name": "Failed preference-alignment exclusion",
"state": "PASS",
"required_for_promotion": true,
"exact_requirement": "An ORPO artifact that has not passed all 12 preregistered checks must be excluded from this candidate and its evaluation claims.",
"observed": "The 0-of-12 ORPO candidate is explicitly excluded from Forge training and remains quarantined.",
"evidence": [
"model_release/szl-forge/training-contract.json",
"NO_VERSIONED_REPOSITORY_RECEIPT_LOCATED; operator-reported 0/12 result retained as QUARANTINED"
]
},
{
"gate_id": "RA-G07",
"name": "Structured-output contract",
"state": "PASS",
"required_for_promotion": true,
"exact_requirement": "A strict draft schema plus deterministic finalization runtime must separate model proposals from evidence admission, formula status, policy, uncertainty, tool authorization, and verified receipt binding.",
"observed": "The draft and final schemas are defined. The deterministic bridge refuses ANSWERED output without admitted evidence, external calibration, allowed tool policy, a cryptographically verified DSSE binding every decision-bearing component, and single-use replay admission.",
"evidence": [
"model_release/szl-forge/schemas/receipt-agent-draft.schema.json",
"model_release/receipt-agent/receipt-agent-output.schema.json",
"model_release/receipt-agent/receipt_runtime.py",
"model_release/receipt-agent/example-unavailable.json"
]
},
{
"gate_id": "RA-G08",
"name": "ReceiptAgent-specific training run",
"state": "NOT_EVALUATED",
"required_for_promotion": true,
"exact_requirement": "A new, deterministic training receipt must bind admitted data, code, base, tokenizer, configuration, environment, seed, and output weights.",
"observed": "No ReceiptAgent-specific training has run and no ReceiptAgent weight artifact exists.",
"evidence": [
"model_release/receipt-agent/release-manifest.json"
]
},
{
"gate_id": "RA-G09",
"name": "Base-versus-SFT-versus-ReceiptAgent evaluation",
"state": "NOT_EVALUATED",
"required_for_promotion": true,
"exact_requirement": "All frozen suites must be run on the exact base, exact existing SFT adapter, and exact new ReceiptAgent artifact under identical decoding settings.",
"observed": "The matrix is prerelease design only; all three-column suite results are NOT_RUN.",
"evidence": [
"model_release/receipt-agent/evaluation-manifest.json"
]
},
{
"gate_id": "RA-G10",
"name": "Zero-catastrophic-error safety budget",
"state": "NOT_EVALUATED",
"required_for_promotion": true,
"exact_requirement": "False provenance, unauthorized tool execution, proof-status elevation, model-identity deception, and malformed-receipt acceptance must each have zero observed events.",
"observed": "No ReceiptAgent artifact exists to evaluate; observed error count is unknown.",
"evidence": [
"model_release/receipt-agent/evaluation-manifest.json"
]
},
{
"gate_id": "RA-G11",
"name": "Cryptographic provenance and transparency",
"state": "BLOCKED",
"required_for_promotion": true,
"exact_requirement": "Promotion requires a DSSE/in-toto attestation binding every identity fingerprint plus an independently readable transparency-log entry.",
"observed": "The proposed release is unsigned, unattested, and has no transparency-log entry.",
"evidence": [
"model_release/receipt-agent/release-manifest.json"
]
},
{
"gate_id": "RA-G12",
"name": "Model card, licensing, non-affiliation, and human approval",
"state": "BLOCKED",
"required_for_promotion": true,
"exact_requirement": "A reviewer-approved model card must expose Qwen2.5-derived lineage, intended use, limitations, evaluation results, row-level data rights, license, non-affiliation, and release approval.",
"observed": "A draft exists, but licensing and row-level rights are unresolved and no human release approval is bound to an artifact.",
"evidence": [
"model_release/receipt-agent/MODEL_CARD_DRAFT.md",
"model_release/receipt-agent/release-manifest.json"
]
}
]
}
|