betterwithage Claude Opus 4.7 commited on
Commit
542b994
·
verified ·
1 Parent(s): c500355

deploy(hf): sync szl-holdings/a11oy@main derived COPY set

Browse files

Reusable Dockerfile-COPY-derived deploy from szl-holdings/a11oy main.
Files: 1131 Pruned: 0
Derived from Dockerfile COPY sources (NO hand-maintained allowlist).

Signed-off-by: SZL Holdings <noreply@szlholdings.ai>
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>

model_release/receipt-agent/MODEL_CARD_DRAFT.md CHANGED
@@ -80,12 +80,12 @@ The existing adapter showed a directional loss improvement on a 16-row observati
80
  |---|---:|---:|---|
81
  | Existing SFT dataset | 167 rows | 0 approved for reuse | Rights/privacy/contamination review incomplete |
82
  | A11oy Brain | 9,464 raw nodes | 0 | All raw nodes training-quarantined |
83
- | Formula crosswalk | 148 rows | 0 | Frozen holdout; namespace/status evaluation only |
84
  | ORPO candidate | 12 qualification checks | 0 | 0/12 passed; quarantined |
85
 
86
  The Brain and formula system can make the released model more useful **without being copied into the weights**. They remain external retrieval, holdout, contradiction, freshness, and proof-status infrastructure. A future training curriculum must be newly assembled from owned or explicitly licensed rows with item-level provenance, deduplication, contamination analysis, and immutable split receipts.
87
 
88
- No card may claim training on "all 200 formulas," the Brain, Lean, mathlib, GitHub, publications, or Hugging Face until row-level admission evidence proves the exact statement. The current repository evidence supports 148 formula crosswalk rows, not a verified 200-formula training set.
89
 
90
  ## Two-stage output contract
91
 
 
80
  |---|---:|---:|---|
81
  | Existing SFT dataset | 167 rows | 0 approved for reuse | Rights/privacy/contamination review incomplete |
82
  | A11oy Brain | 9,464 raw nodes | 0 | All raw nodes training-quarantined |
83
+ | Formula admission | 146 formula crosswalk records + 2 SZL-Lake evidence records = 148 holdout rows | 0 | Frozen holdout; namespace/status evaluation only |
84
  | ORPO candidate | 12 qualification checks | 0 | 0/12 passed; quarantined |
85
 
86
  The Brain and formula system can make the released model more useful **without being copied into the weights**. They remain external retrieval, holdout, contradiction, freshness, and proof-status infrastructure. A future training curriculum must be newly assembled from owned or explicitly licensed rows with item-level provenance, deduplication, contamination analysis, and immutable split receipts.
87
 
88
+ No card may claim training on "all 200 formulas," the Brain, Lean, mathlib, GitHub, publications, or Hugging Face until row-level admission evidence proves the exact statement. The current repository evidence supports 146 formula crosswalk records plus 2 SZL-Lake evidence records in a 148-row holdout tranche, not a verified 200-formula training set.
89
 
90
  ## Two-stage output contract
91
 
model_release/receipt-agent/README.md CHANGED
@@ -27,7 +27,7 @@ turn retrieved text into a mathematical proof.
27
  | Existing evaluation | Incomplete | A 16-row loss direction exists, but general quality is not established. |
28
  | ORPO candidate | 0/12 qualification checks | Quarantined; cannot influence release. |
29
  | Brain | 9,464 raw nodes | All 9,464 are training-quarantined; zero raw-node text rows are admitted to train. |
30
- | Formula crosswalk | 148 rows | All 148 are holdout; zero are admitted to train. Formula IDs are namespace-scoped. |
31
  | Proposed model | No ReceiptAgent weight artifact | `NOT_PROMOTED`, `NOT_ESTABLISHED`, and unavailable for inference. |
32
 
33
  The current Brain and formula corpus are valuable as a governed retrieval and
@@ -63,7 +63,7 @@ and permits zero catastrophic errors. Current blockers are:
63
 
64
  - incomplete row-level governance for the existing 167-row dataset;
65
  - 9,464/9,464 raw Brain nodes quarantined from training;
66
- - 148/148 formula rows reserved for holdout, with F1-F23 namespace collisions;
67
  - the ORPO candidate failing all 12 qualification checks;
68
  - no ReceiptAgent-specific training receipt or weights;
69
  - no frozen three-way evaluation;
 
27
  | Existing evaluation | Incomplete | A 16-row loss direction exists, but general quality is not established. |
28
  | ORPO candidate | 0/12 qualification checks | Quarantined; cannot influence release. |
29
  | Brain | 9,464 raw nodes | All 9,464 are training-quarantined; zero raw-node text rows are admitted to train. |
30
+ | Formula admission | 146 formula crosswalk records + 2 SZL-Lake evidence records = 148 holdout tranche rows | All 148 are holdout; zero are admitted to train. Formula IDs are namespace-scoped. |
31
  | Proposed model | No ReceiptAgent weight artifact | `NOT_PROMOTED`, `NOT_ESTABLISHED`, and unavailable for inference. |
32
 
33
  The current Brain and formula corpus are valuable as a governed retrieval and
 
63
 
64
  - incomplete row-level governance for the existing 167-row dataset;
65
  - 9,464/9,464 raw Brain nodes quarantined from training;
66
+ - 146 formula records plus 2 SZL-Lake evidence records reserved as 148/148 holdout rows, with F1-F23 namespace collisions;
67
  - the ORPO candidate failing all 12 qualification checks;
68
  - no ReceiptAgent-specific training receipt or weights;
69
  - no frozen three-way evaluation;
model_release/receipt-agent/admission-manifest.json CHANGED
@@ -59,7 +59,7 @@
59
  "state": "PASS",
60
  "required_for_promotion": true,
61
  "exact_requirement": "Formula IDs must be namespace-scoped, proof transfer must require an explicit semantic binding, and no holdout formula may enter training.",
62
- "observed": "The current crosswalk has 148 holdout rows and zero train rows; F1-F23 namespace collisions block implicit proof transfer.",
63
  "evidence": [
64
  "research/formula-training-admission/admission-manifest.json",
65
  "research/formula-training-admission/formula-id-crosswalk.json"
 
59
  "state": "PASS",
60
  "required_for_promotion": true,
61
  "exact_requirement": "Formula IDs must be namespace-scoped, proof transfer must require an explicit semantic binding, and no holdout formula may enter training.",
62
+ "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.",
63
  "evidence": [
64
  "research/formula-training-admission/admission-manifest.json",
65
  "research/formula-training-admission/formula-id-crosswalk.json"
model_release/receipt-agent/release-manifest.json CHANGED
@@ -17,7 +17,7 @@
17
  "brain_raw_nodes": 9464,
18
  "brain_raw_nodes_training_quarantined": 9464,
19
  "brain_raw_rows_admitted_to_train": 0,
20
- "formula_crosswalk_rows": 148,
21
  "formula_holdout_rows": 148,
22
  "formula_train_rows": 0,
23
  "orpo_checks_passed": 0,
 
17
  "brain_raw_nodes": 9464,
18
  "brain_raw_nodes_training_quarantined": 9464,
19
  "brain_raw_rows_admitted_to_train": 0,
20
+ "formula_crosswalk_rows": 146,
21
  "formula_holdout_rows": 148,
22
  "formula_train_rows": 0,
23
  "orpo_checks_passed": 0,
model_release/receipt-agent/release-manifest.schema.json CHANGED
@@ -95,7 +95,7 @@
95
  "const": 0
96
  },
97
  "formula_crosswalk_rows": {
98
- "const": 148
99
  },
100
  "formula_holdout_rows": {
101
  "const": 148
 
95
  "const": 0
96
  },
97
  "formula_crosswalk_rows": {
98
+ "const": 146
99
  },
100
  "formula_holdout_rows": {
101
  "const": 148
model_release/szl-forge-family.json CHANGED
@@ -114,12 +114,18 @@
114
  "raw_nodes_admitted_to_gradients": 0,
115
  "current_state": "ROW_LEVEL_ADMISSION_ENGINE_IMPLEMENTED_CURRENT_RAW_ROWS_QUARANTINED",
116
  "admission_engine": "szl_brain_training_admission.py",
117
- "admission_contract": "szl.brain-training-admission-report.v1",
118
- "admission_requirement": "Each gradient row requires Ed25519-signed allowlisted evidence for rights, provenance, immutable revision, freshness, deduplication, contamination analysis, and a frozen cross-run split assignment.",
119
  "required_signed_inputs": [
120
  "protected_eval_content_sha256_list",
121
- "signed_evidence_trust_store",
122
- "signed_prior_split_ledger_descriptor"
 
 
 
 
 
 
123
  ]
124
  },
125
  "formula_policy": {
 
114
  "raw_nodes_admitted_to_gradients": 0,
115
  "current_state": "ROW_LEVEL_ADMISSION_ENGINE_IMPLEMENTED_CURRENT_RAW_ROWS_QUARANTINED",
116
  "admission_engine": "szl_brain_training_admission.py",
117
+ "admission_contract": "szl.brain-training-admission-report.v2",
118
+ "admission_requirement": "Each gradient row requires Ed25519-signed, purpose-scoped evidence for source identity, authorship/rightsholder permission, privacy/PII clearance, immutable revision, freshness, deduplication, held-out contamination analysis, reviewer-owned approval, and a frozen cross-run split assignment rooted in a signed policy bundle; train admission is disabled by default and terminal artifacts require a distinct signing key.",
119
  "required_signed_inputs": [
120
  "protected_eval_content_sha256_list",
121
+ "purpose_scoped_evidence_trust_store",
122
+ "policy_root_signer",
123
+ "root_signed_policy_bundle",
124
+ "signed_prior_split_ledger_descriptor",
125
+ "exact_split_ledger_head_sha256",
126
+ "reviewer_allowlist",
127
+ "artifact_signing_key",
128
+ "explicit_train_admission_switch"
129
  ]
130
  },
131
  "formula_policy": {
model_release/szl-khipu-second-brain.json CHANGED
@@ -50,21 +50,32 @@
50
  "current_state": "ROW_LEVEL_ADMISSION_ENGINE_IMPLEMENTED_CURRENT_RAW_ROWS_QUARANTINED",
51
  "admission_engine": "szl_brain_training_admission.py",
52
  "admission_test": "tests/test_brain_training_admission.py",
53
- "admission_contract": "szl.brain-training-admission-report.v1",
54
  "admission_evidence_security": {
55
  "signature": "ED25519",
56
- "trust_model": "ALLOWLISTED_ISSUER_TOOL_KEY_TRIPLE",
57
- "content_binding": "EXACT_CONTENT_SOURCE_REVISION_RIGHTS_AND_CONTAMINATION",
58
- "cross_run_isolation": "SIGNED_DETERMINISTIC_SPLIT_LEDGER",
 
 
59
  "self_attestation_allowed": false
60
  },
61
  "required_admission_inputs": [
62
  "protected_eval_content_sha256_list",
63
- "signed_evidence_trust_store",
64
- "signed_prior_split_ledger_descriptor"
 
 
 
 
 
 
65
  ],
66
  "current_inventory_obligations": {
67
  "license_or_rights_unknown": 9464,
 
 
 
68
  "immutable_source_revision_unknown": 9464,
69
  "freshness_unknown": 9430,
70
  "duplicate_candidates": 3747,
@@ -85,24 +96,37 @@
85
  "file_sha256": "6f5e0d9ecb60d8aa52fc229a42ad37d62fc7fa7406993a0b8352b2b574fb40a4"
86
  }
87
  },
88
- "candidate_schema": "szl.brain-training-candidate.v1",
89
- "decision_schema": "szl.brain-training-admission-decision.v1",
90
- "report_schema": "szl.brain-training-admission-report.v1",
91
  "candidate_required_inputs": [
92
  "node_id",
93
  "content",
94
  "content_sha256",
95
  "source.uri",
 
96
  "source.revision",
97
  "source.timestamp_utc",
98
  "source.evidence",
 
 
99
  "rights.basis",
100
  "rights.license",
 
101
  "rights.evidence",
 
 
 
 
102
  "contamination.result",
103
  "contamination.method",
104
  "contamination.checked_against",
105
  "contamination.evidence",
 
 
 
 
 
106
  "split"
107
  ],
108
  "derived_decision_fields": [
@@ -112,7 +136,9 @@
112
  "freshness.state",
113
  "source.status",
114
  "rights.status",
 
115
  "contamination.observed_result",
 
116
  "training_eligible",
117
  "evaluation_eligible",
118
  "reason_codes",
@@ -123,6 +149,9 @@
123
  "COPY_EVALUATION_QRELS_INTO_TRAINING",
124
  "TREAT_RETRIEVAL_AS_TRAINING",
125
  "LABEL_INDEX_NODES_AS_PARAMETERS",
 
 
 
126
  "WEAKEN_GATES_TO_REACH_A_TARGET_ROW_COUNT"
127
  ]
128
  },
 
50
  "current_state": "ROW_LEVEL_ADMISSION_ENGINE_IMPLEMENTED_CURRENT_RAW_ROWS_QUARANTINED",
51
  "admission_engine": "szl_brain_training_admission.py",
52
  "admission_test": "tests/test_brain_training_admission.py",
53
+ "admission_contract": "szl.brain-training-admission-report.v2",
54
  "admission_evidence_security": {
55
  "signature": "ED25519",
56
+ "trust_model": "ROOT_SIGNED_PURPOSE_SCOPED_ISSUER_TOOL_KEY",
57
+ "content_binding": "EXACT_CONTENT_SOURCE_IDENTITY_REVISION_AUTHOR_RIGHTSHOLDER_PERMISSION_PRIVACY_REVIEW_AND_CONTAMINATION",
58
+ "cross_run_isolation": "POLICY_PINNED_SIGNED_SPLIT_LEDGER_HEAD",
59
+ "review_signer_binding": "REVIEWER_OWNED_KEY",
60
+ "terminal_artifact": "ED25519_SIGNED_MANIFEST",
61
  "self_attestation_allowed": false
62
  },
63
  "required_admission_inputs": [
64
  "protected_eval_content_sha256_list",
65
+ "purpose_scoped_evidence_trust_store",
66
+ "policy_root_signer",
67
+ "root_signed_policy_bundle",
68
+ "signed_prior_split_ledger_descriptor",
69
+ "exact_split_ledger_head_sha256",
70
+ "reviewer_allowlist",
71
+ "artifact_signing_key",
72
+ "explicit_train_admission_switch"
73
  ],
74
  "current_inventory_obligations": {
75
  "license_or_rights_unknown": 9464,
76
+ "author_or_rightsholder_permission_unknown": 9464,
77
+ "privacy_or_pii_clearance_unknown": 9464,
78
+ "allowlisted_review_unknown": 9464,
79
  "immutable_source_revision_unknown": 9464,
80
  "freshness_unknown": 9430,
81
  "duplicate_candidates": 3747,
 
96
  "file_sha256": "6f5e0d9ecb60d8aa52fc229a42ad37d62fc7fa7406993a0b8352b2b574fb40a4"
97
  }
98
  },
99
+ "candidate_schema": "szl.brain-training-candidate.v2",
100
+ "decision_schema": "szl.brain-training-admission-decision.v2",
101
+ "report_schema": "szl.brain-training-admission-report.v2",
102
  "candidate_required_inputs": [
103
  "node_id",
104
  "content",
105
  "content_sha256",
106
  "source.uri",
107
+ "source.identity",
108
  "source.revision",
109
  "source.timestamp_utc",
110
  "source.evidence",
111
+ "rights.author",
112
+ "rights.rightsholder",
113
  "rights.basis",
114
  "rights.license",
115
+ "rights.permission_scope",
116
  "rights.evidence",
117
+ "privacy.classification",
118
+ "privacy.pii_result",
119
+ "privacy.method",
120
+ "privacy.evidence",
121
  "contamination.result",
122
  "contamination.method",
123
  "contamination.checked_against",
124
  "contamination.evidence",
125
+ "review.state",
126
+ "review.reviewer",
127
+ "review.reviewed_at_utc",
128
+ "review.reasons",
129
+ "review.evidence",
130
  "split"
131
  ],
132
  "derived_decision_fields": [
 
136
  "freshness.state",
137
  "source.status",
138
  "rights.status",
139
+ "privacy.status",
140
  "contamination.observed_result",
141
+ "review.status",
142
  "training_eligible",
143
  "evaluation_eligible",
144
  "reason_codes",
 
149
  "COPY_EVALUATION_QRELS_INTO_TRAINING",
150
  "TREAT_RETRIEVAL_AS_TRAINING",
151
  "LABEL_INDEX_NODES_AS_PARAMETERS",
152
+ "AUTO_APPROVE_UNKNOWN_PRIVACY_OR_RIGHTS",
153
+ "SELF_REVIEW_WITHOUT_ALLOWLISTED_SIGNED_EVIDENCE",
154
+ "ENABLE_TRAIN_ADMISSION_BY_DEFAULT",
155
  "WEAKEN_GATES_TO_REACH_A_TARGET_ROW_COUNT"
156
  ]
157
  },
schemas/quant-claims/benchmark-run.v1.schema.json ADDED
@@ -0,0 +1,217 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "$schema": "https://json-schema.org/draft/2020-12/schema",
3
+ "$id": "https://a-11-oy.com/schemas/quant-claims/benchmark-run.v1.schema.json",
4
+ "title": "SZL Quant Benchmark Run v1",
5
+ "type": "object",
6
+ "additionalProperties": false,
7
+ "required": [
8
+ "schema_version", "run_id", "claim_id", "protocol", "subject", "baseline",
9
+ "dataset", "hardware", "software", "execution", "raw_trials", "measurement",
10
+ "correctness", "energy", "freshness", "review", "artifacts"
11
+ ],
12
+ "properties": {
13
+ "schema_version": {"const": "szl.quant.benchmark-run/v1"},
14
+ "run_id": {"type": "string", "pattern": "^[A-Za-z0-9][A-Za-z0-9._:-]{2,127}$"},
15
+ "claim_id": {"type": "string", "pattern": "^[a-z0-9][a-z0-9-]{2,95}$"},
16
+ "protocol": {
17
+ "type": "object", "additionalProperties": false,
18
+ "required": ["id", "version", "sha256", "preregistered_at", "manifest"],
19
+ "properties": {
20
+ "id": {"type": "string"}, "version": {"type": "string"},
21
+ "sha256": {"type": "string", "pattern": "^sha256:[0-9a-f]{64}$"},
22
+ "preregistered_at": {"type": "string", "format": "date-time"},
23
+ "manifest": {
24
+ "type": "object", "additionalProperties": false,
25
+ "required": [
26
+ "id", "version", "metric", "unit", "preregistered_at", "minimum_trials",
27
+ "hardware_class", "requires_energy", "dataset_manifest_sha256"
28
+ ],
29
+ "properties": {
30
+ "id": {"type": "string"}, "version": {"type": "string"},
31
+ "metric": {"type": "string"}, "unit": {"type": "string"},
32
+ "preregistered_at": {"type": "string", "format": "date-time"},
33
+ "minimum_trials": {"type": "integer", "minimum": 3},
34
+ "hardware_class": {"type": "string"},
35
+ "requires_energy": {"type": "boolean"},
36
+ "dataset_manifest_sha256": {"type": "string", "pattern": "^sha256:[0-9a-f]{64}$"}
37
+ }
38
+ }
39
+ }
40
+ },
41
+ "subject": {"$ref": "#/$defs/artifactSubject"},
42
+ "baseline": {"$ref": "#/$defs/artifactSubject"},
43
+ "dataset": {
44
+ "type": "object", "additionalProperties": false,
45
+ "required": ["manifest_sha256", "split", "seeds", "rights_admitted"],
46
+ "properties": {
47
+ "manifest_sha256": {"type": "string", "pattern": "^sha256:[0-9a-f]{64}$"},
48
+ "split": {"type": "string"},
49
+ "seeds": {"type": "array", "minItems": 1, "items": {"type": "integer"}},
50
+ "rights_admitted": {"const": true}
51
+ }
52
+ },
53
+ "hardware": {
54
+ "type": "object", "additionalProperties": false,
55
+ "required": ["class", "devices", "driver", "cuda"],
56
+ "properties": {
57
+ "class": {"type": "string"},
58
+ "devices": {"type": "array", "minItems": 1, "items": {"$ref": "#/$defs/device"}},
59
+ "driver": {"type": "string"}, "cuda": {"type": "string"}
60
+ }
61
+ },
62
+ "software": {
63
+ "type": "object", "additionalProperties": false,
64
+ "required": ["git_commit", "container_digest", "lock_sha256"],
65
+ "properties": {
66
+ "git_commit": {"type": "string", "pattern": "^[0-9a-f]{40}$"},
67
+ "container_digest": {"type": "string", "pattern": "^sha256:[0-9a-f]{64}$"},
68
+ "lock_sha256": {"type": "string", "pattern": "^sha256:[0-9a-f]{64}$"}
69
+ }
70
+ },
71
+ "execution": {
72
+ "type": "object", "additionalProperties": false,
73
+ "required": ["started_at", "completed_at", "warmup_trials", "measured_trials", "config"],
74
+ "properties": {
75
+ "started_at": {"type": "string", "format": "date-time"},
76
+ "completed_at": {"type": "string", "format": "date-time"},
77
+ "warmup_trials": {"type": "integer", "minimum": 0},
78
+ "measured_trials": {"type": "integer", "minimum": 3},
79
+ "config": {"type": "object"}
80
+ }
81
+ },
82
+ "raw_trials": {
83
+ "type": "array", "minItems": 3,
84
+ "items": {"$ref": "#/$defs/rawTrial"}
85
+ },
86
+ "measurement": {"$ref": "#/$defs/measurement"},
87
+ "correctness": {"$ref": "#/$defs/correctness"},
88
+ "energy": {"$ref": "#/$defs/energy"},
89
+ "freshness": {
90
+ "type": "object", "additionalProperties": false,
91
+ "required": ["fresh_until"],
92
+ "properties": {"fresh_until": {"type": "string", "format": "date-time"}}
93
+ },
94
+ "review": {
95
+ "type": "object", "additionalProperties": false,
96
+ "required": ["status", "reviewer", "reviewed_at"],
97
+ "properties": {
98
+ "status": {"const": "APPROVED"}, "reviewer": {"type": "string", "minLength": 1},
99
+ "reviewed_at": {"type": "string", "format": "date-time"}
100
+ }
101
+ },
102
+ "artifacts": {
103
+ "type": "object", "additionalProperties": false,
104
+ "required": ["result_sha256", "result_bundle"],
105
+ "properties": {
106
+ "result_sha256": {"type": "string", "pattern": "^sha256:[0-9a-f]{64}$"},
107
+ "result_bundle": {
108
+ "type": "object", "additionalProperties": false,
109
+ "required": ["raw_trials", "measurement", "correctness", "energy"],
110
+ "properties": {
111
+ "raw_trials": {"type": "array", "minItems": 3, "items": {"$ref": "#/$defs/rawTrial"}},
112
+ "measurement": {"$ref": "#/$defs/measurement"},
113
+ "correctness": {"$ref": "#/$defs/correctness"},
114
+ "energy": {"$ref": "#/$defs/energy"}
115
+ }
116
+ }
117
+ }
118
+ }
119
+ },
120
+ "$defs": {
121
+ "artifactSubject": {
122
+ "type": "object", "additionalProperties": false,
123
+ "required": ["artifact_id", "revision", "sha256", "license"],
124
+ "properties": {
125
+ "artifact_id": {"type": "string"}, "revision": {"type": "string"},
126
+ "sha256": {"type": "string", "pattern": "^sha256:[0-9a-f]{64}$"},
127
+ "license": {"type": "string", "minLength": 1}
128
+ }
129
+ },
130
+ "device": {
131
+ "type": "object", "additionalProperties": false,
132
+ "required": ["kind", "model", "count"],
133
+ "properties": {
134
+ "kind": {"enum": ["CPU", "GPU", "REMOTE_ACCELERATOR"]},
135
+ "model": {"type": "string", "minLength": 1},
136
+ "count": {"type": "integer", "minimum": 1}
137
+ }
138
+ },
139
+ "rawTrial": {
140
+ "type": "object", "additionalProperties": false,
141
+ "required": ["trial", "value", "unit"],
142
+ "properties": {
143
+ "trial": {"type": "integer", "minimum": 1},
144
+ "value": {"type": "number"},
145
+ "unit": {"type": "string", "minLength": 1}
146
+ }
147
+ },
148
+ "measurement": {
149
+ "type": "object", "additionalProperties": false,
150
+ "required": ["metric", "value", "unit", "estimator", "confidence_interval", "summary_tolerance"],
151
+ "properties": {
152
+ "metric": {"type": "string", "minLength": 1},
153
+ "value": {"type": "number"},
154
+ "unit": {"type": "string", "minLength": 1},
155
+ "estimator": {"enum": ["mean", "median"]},
156
+ "summary_tolerance": {"type": "number", "minimum": 0},
157
+ "confidence_interval": {
158
+ "oneOf": [
159
+ {"type": "null"},
160
+ {"type": "array", "prefixItems": [{"type": "number"}, {"type": "number"}], "items": false, "minItems": 2, "maxItems": 2}
161
+ ]
162
+ }
163
+ }
164
+ },
165
+ "correctness": {
166
+ "type": "object", "additionalProperties": false,
167
+ "required": ["passed", "checks"],
168
+ "properties": {
169
+ "passed": {"const": true},
170
+ "checks": {
171
+ "type": "array", "minItems": 1,
172
+ "items": {
173
+ "type": "object", "additionalProperties": false,
174
+ "required": ["name", "passed", "tolerance"],
175
+ "properties": {
176
+ "name": {"type": "string", "minLength": 1},
177
+ "passed": {"const": true},
178
+ "tolerance": {"type": "string", "minLength": 1}
179
+ }
180
+ }
181
+ }
182
+ }
183
+ },
184
+ "energy": {
185
+ "oneOf": [
186
+ {
187
+ "type": "object", "additionalProperties": false,
188
+ "required": ["status", "blocker"],
189
+ "properties": {
190
+ "status": {"const": "NOT_OBSERVABLE"},
191
+ "blocker": {"type": "string", "minLength": 1}
192
+ }
193
+ },
194
+ {
195
+ "type": "object", "additionalProperties": false,
196
+ "required": [
197
+ "status", "exporter_id", "sampled_at_start", "sampled_at_end",
198
+ "joules_before", "joules_after", "work_units_before", "work_units_after",
199
+ "work_unit", "fresh"
200
+ ],
201
+ "properties": {
202
+ "status": {"const": "MEASURED_DELTA"},
203
+ "exporter_id": {"type": "string", "minLength": 1},
204
+ "sampled_at_start": {"type": "string", "format": "date-time"},
205
+ "sampled_at_end": {"type": "string", "format": "date-time"},
206
+ "joules_before": {"type": "number", "minimum": 0},
207
+ "joules_after": {"type": "number", "minimum": 0},
208
+ "work_units_before": {"type": "number", "minimum": 0},
209
+ "work_units_after": {"type": "number", "minimum": 0},
210
+ "work_unit": {"type": "string", "minLength": 1},
211
+ "fresh": {"const": true}
212
+ }
213
+ }
214
+ ]
215
+ }
216
+ }
217
+ }
schemas/quant-claims/claim.v1.schema.json ADDED
@@ -0,0 +1,118 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "$schema": "https://json-schema.org/draft/2020-12/schema",
3
+ "$id": "https://a-11-oy.com/schemas/quant-claims/claim.v1.schema.json",
4
+ "title": "SZL Quant External Claim Resolution v1",
5
+ "type": "object",
6
+ "additionalProperties": false,
7
+ "required": [
8
+ "schema_version",
9
+ "claim_id",
10
+ "claim",
11
+ "external_report",
12
+ "protocol",
13
+ "required_hardware_class",
14
+ "measurement_receipt_id",
15
+ "blocked_by",
16
+ "szl_label",
17
+ "szl_measured",
18
+ "measurement_gate"
19
+ ],
20
+ "properties": {
21
+ "schema_version": {"const": "szl.quant.claim/v1"},
22
+ "claim_id": {"type": "string", "pattern": "^[a-z0-9][a-z0-9-]{2,95}$"},
23
+ "claim": {"type": "string", "minLength": 1, "maxLength": 240},
24
+ "external_report": {
25
+ "type": "object",
26
+ "additionalProperties": false,
27
+ "required": ["value", "units", "scope", "source"],
28
+ "properties": {
29
+ "value": {"type": "string", "minLength": 1, "maxLength": 120},
30
+ "units": {"type": "string", "minLength": 1, "maxLength": 80},
31
+ "scope": {"type": "string", "minLength": 1, "maxLength": 500},
32
+ "source": {
33
+ "type": "object",
34
+ "additionalProperties": false,
35
+ "required": ["organization", "title", "url", "kind"],
36
+ "properties": {
37
+ "organization": {"type": "string", "minLength": 1, "maxLength": 120},
38
+ "title": {"type": "string", "minLength": 1, "maxLength": 240},
39
+ "url": {"type": "string", "pattern": "^https://"},
40
+ "kind": {
41
+ "enum": ["OFFICIAL_VENDOR_REPORT", "OFFICIAL_RESEARCH", "OFFICIAL_DOCS", "OFFICIAL_MODEL_CARD", "ACADEMIC_PAPER"]
42
+ }
43
+ }
44
+ }
45
+ }
46
+ },
47
+ "protocol": {
48
+ "type": "object",
49
+ "additionalProperties": false,
50
+ "required": ["id", "metric", "unit", "description", "requires_energy"],
51
+ "properties": {
52
+ "id": {"type": "string", "pattern": "^[a-z0-9][a-z0-9-]{2,95}$"},
53
+ "metric": {"type": "string", "pattern": "^[a-z][a-z0-9_]{2,95}$"},
54
+ "unit": {"type": "string", "minLength": 1, "maxLength": 80},
55
+ "description": {"type": "string", "minLength": 1, "maxLength": 500},
56
+ "requires_energy": {"type": "boolean"}
57
+ }
58
+ },
59
+ "required_hardware_class": {"type": "string", "pattern": "^[A-Z][A-Z0-9_]{2,95}$"},
60
+ "measurement_receipt_id": {
61
+ "oneOf": [
62
+ {"type": "null"},
63
+ {"type": "string", "pattern": "^sha256:[0-9a-f]{64}$"}
64
+ ]
65
+ },
66
+ "blocked_by": {
67
+ "type": "array",
68
+ "minItems": 1,
69
+ "items": {"type": "string", "minLength": 1, "maxLength": 240},
70
+ "uniqueItems": true
71
+ },
72
+ "szl_label": {"enum": ["ROADMAP", "MEASURED", "UNVERIFIED", "CONFLICT"]},
73
+ "szl_measured": {
74
+ "oneOf": [
75
+ {"type": "null"},
76
+ {
77
+ "type": "object",
78
+ "additionalProperties": false,
79
+ "required": ["metric", "value", "unit", "estimator", "confidence_interval", "summary_tolerance"],
80
+ "properties": {
81
+ "metric": {"type": "string"},
82
+ "value": {"type": "number"},
83
+ "unit": {"type": "string"},
84
+ "estimator": {"enum": ["mean", "median"]},
85
+ "summary_tolerance": {"type": "number", "minimum": 0},
86
+ "confidence_interval": {
87
+ "oneOf": [
88
+ {"type": "null"},
89
+ {
90
+ "type": "array",
91
+ "prefixItems": [{"type": "number"}, {"type": "number"}],
92
+ "items": false,
93
+ "minItems": 2,
94
+ "maxItems": 2
95
+ }
96
+ ]
97
+ }
98
+ }
99
+ }
100
+ ]
101
+ },
102
+ "measurement_gate": {
103
+ "type": "object",
104
+ "additionalProperties": false,
105
+ "required": ["promoted", "reason", "receipt_id"],
106
+ "properties": {
107
+ "promoted": {"type": "boolean"},
108
+ "reason": {"type": "string", "minLength": 1, "maxLength": 240},
109
+ "receipt_id": {
110
+ "oneOf": [
111
+ {"type": "null"},
112
+ {"type": "string", "pattern": "^sha256:[0-9a-f]{64}$"}
113
+ ]
114
+ }
115
+ }
116
+ }
117
+ }
118
+ }