@mcoda/mswarm 0.1.90 → 0.1.94
This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
- package/dist/codali-executor.d.ts +107 -6
- package/dist/codali-executor.d.ts.map +1 -1
- package/dist/codali-executor.js +500 -26
- package/dist/codali-executor.js.map +1 -1
- package/dist/runtime.d.ts +4 -2
- package/dist/runtime.d.ts.map +1 -1
- package/dist/runtime.js +18 -5
- package/dist/runtime.js.map +1 -1
- package/dist/vendor/codali/cli/DatasetCommand.d.ts +38 -0
- package/dist/vendor/codali/cli/DatasetCommand.d.ts.map +1 -0
- package/dist/vendor/codali/cli/DatasetCommand.js +959 -0
- package/dist/vendor/codali/cli/EvalCommand.d.ts +3 -0
- package/dist/vendor/codali/cli/EvalCommand.d.ts.map +1 -1
- package/dist/vendor/codali/cli/EvalCommand.js +55 -1
- package/dist/vendor/codali/cli/ImprovementCommand.d.ts +97 -0
- package/dist/vendor/codali/cli/ImprovementCommand.d.ts.map +1 -0
- package/dist/vendor/codali/cli/ImprovementCommand.js +1879 -0
- package/dist/vendor/codali/cli.d.ts.map +1 -1
- package/dist/vendor/codali/cli.js +15 -0
- package/dist/vendor/codali/dataset-cli.d.ts +3 -0
- package/dist/vendor/codali/dataset-cli.d.ts.map +1 -0
- package/dist/vendor/codali/dataset-cli.js +30 -0
- package/dist/vendor/codali/eval/CodaliGatewayLiveHarness.d.ts +90 -1
- package/dist/vendor/codali/eval/CodaliGatewayLiveHarness.d.ts.map +1 -1
- package/dist/vendor/codali/eval/CodaliGatewayLiveHarness.js +443 -9
- package/dist/vendor/codali/eval/GatewayDatasetEval.d.ts +70 -0
- package/dist/vendor/codali/eval/GatewayDatasetEval.d.ts.map +1 -0
- package/dist/vendor/codali/eval/GatewayDatasetEval.js +477 -0
- package/dist/vendor/codali/eval/GatewayEvalSuite.d.ts +52 -0
- package/dist/vendor/codali/eval/GatewayEvalSuite.d.ts.map +1 -1
- package/dist/vendor/codali/eval/GatewayEvalSuite.js +135 -1
- package/dist/vendor/codali/gateway/CodaliGateway.d.ts +4 -0
- package/dist/vendor/codali/gateway/CodaliGateway.d.ts.map +1 -1
- package/dist/vendor/codali/gateway/CodaliGateway.js +37 -1
- package/dist/vendor/codali/gateway/CodaliGatewaySchemas.d.ts.map +1 -1
- package/dist/vendor/codali/gateway/CodaliGatewaySchemas.js +0 -4
- package/dist/vendor/codali/gateway/CodaliGatewayStore.d.ts.map +1 -1
- package/dist/vendor/codali/gateway/CodaliGatewayStore.js +1 -0
- package/dist/vendor/codali/gateway/GatewayPlanner.d.ts.map +1 -1
- package/dist/vendor/codali/gateway/GatewayPlanner.js +355 -41
- package/dist/vendor/codali/gateway/GatewayStateMachine.d.ts +1 -0
- package/dist/vendor/codali/gateway/GatewayStateMachine.d.ts.map +1 -1
- package/dist/vendor/codali/gateway/GatewayStateMachine.js +221 -5
- package/dist/vendor/codali/improvement/CandidateReleaseBuilder.d.ts +201 -0
- package/dist/vendor/codali/improvement/CandidateReleaseBuilder.d.ts.map +1 -0
- package/dist/vendor/codali/improvement/CandidateReleaseBuilder.js +1175 -0
- package/dist/vendor/codali/improvement/DatasetEligibilityGate.d.ts +75 -0
- package/dist/vendor/codali/improvement/DatasetEligibilityGate.d.ts.map +1 -0
- package/dist/vendor/codali/improvement/DatasetEligibilityGate.js +497 -0
- package/dist/vendor/codali/improvement/DatasetExportManifestReader.d.ts +88 -0
- package/dist/vendor/codali/improvement/DatasetExportManifestReader.d.ts.map +1 -0
- package/dist/vendor/codali/improvement/DatasetExportManifestReader.js +333 -0
- package/dist/vendor/codali/improvement/DocdexRetrievalCandidateBuilder.d.ts +192 -0
- package/dist/vendor/codali/improvement/DocdexRetrievalCandidateBuilder.d.ts.map +1 -0
- package/dist/vendor/codali/improvement/DocdexRetrievalCandidateBuilder.js +625 -0
- package/dist/vendor/codali/improvement/EvalReplayCandidateBuilder.d.ts +173 -0
- package/dist/vendor/codali/improvement/EvalReplayCandidateBuilder.d.ts.map +1 -0
- package/dist/vendor/codali/improvement/EvalReplayCandidateBuilder.js +363 -0
- package/dist/vendor/codali/improvement/FineTuneJobPlanner.d.ts +281 -0
- package/dist/vendor/codali/improvement/FineTuneJobPlanner.d.ts.map +1 -0
- package/dist/vendor/codali/improvement/FineTuneJobPlanner.js +823 -0
- package/dist/vendor/codali/improvement/ImprovementEvalRunner.d.ts +50 -0
- package/dist/vendor/codali/improvement/ImprovementEvalRunner.d.ts.map +1 -0
- package/dist/vendor/codali/improvement/ImprovementEvalRunner.js +648 -0
- package/dist/vendor/codali/improvement/ImprovementPolicy.d.ts +272 -0
- package/dist/vendor/codali/improvement/ImprovementPolicy.d.ts.map +1 -0
- package/dist/vendor/codali/improvement/ImprovementPolicy.js +1450 -0
- package/dist/vendor/codali/improvement/ModelRouterCandidateBuilder.d.ts +186 -0
- package/dist/vendor/codali/improvement/ModelRouterCandidateBuilder.d.ts.map +1 -0
- package/dist/vendor/codali/improvement/ModelRouterCandidateBuilder.js +880 -0
- package/dist/vendor/codali/improvement/OperatorInspector.d.ts +182 -0
- package/dist/vendor/codali/improvement/OperatorInspector.d.ts.map +1 -0
- package/dist/vendor/codali/improvement/OperatorInspector.js +604 -0
- package/dist/vendor/codali/improvement/ProductionGovernance.d.ts +91 -0
- package/dist/vendor/codali/improvement/ProductionGovernance.d.ts.map +1 -0
- package/dist/vendor/codali/improvement/ProductionGovernance.js +329 -0
- package/dist/vendor/codali/improvement/PromptSchemaToolMetadataCandidateBuilder.d.ts +152 -0
- package/dist/vendor/codali/improvement/PromptSchemaToolMetadataCandidateBuilder.d.ts.map +1 -0
- package/dist/vendor/codali/improvement/PromptSchemaToolMetadataCandidateBuilder.js +480 -0
- package/dist/vendor/codali/improvement/PublishOrchestrator.d.ts +113 -0
- package/dist/vendor/codali/improvement/PublishOrchestrator.d.ts.map +1 -0
- package/dist/vendor/codali/improvement/PublishOrchestrator.js +590 -0
- package/dist/vendor/codali/improvement/ReleaseOutcomeReporter.d.ts +133 -0
- package/dist/vendor/codali/improvement/ReleaseOutcomeReporter.d.ts.map +1 -0
- package/dist/vendor/codali/improvement/ReleaseOutcomeReporter.js +501 -0
- package/dist/vendor/codali/improvement/StorageServiceImprovementClient.d.ts +71 -0
- package/dist/vendor/codali/improvement/StorageServiceImprovementClient.d.ts.map +1 -0
- package/dist/vendor/codali/improvement/StorageServiceImprovementClient.js +245 -0
- package/dist/vendor/codali/index.d.ts +43 -0
- package/dist/vendor/codali/index.d.ts.map +1 -1
- package/dist/vendor/codali/index.js +21 -0
- package/dist/vendor/codali/runtime/CodaliRuntime.d.ts +0 -2
- package/dist/vendor/codali/runtime/CodaliRuntime.d.ts.map +1 -1
- package/dist/vendor/codali/runtime/CodaliRuntime.js +0 -5
- package/dist/vendor/codali/storage/CodaliDatasetPrivacyEngine.d.ts +138 -0
- package/dist/vendor/codali/storage/CodaliDatasetPrivacyEngine.d.ts.map +1 -0
- package/dist/vendor/codali/storage/CodaliDatasetPrivacyEngine.js +417 -0
- package/dist/vendor/codali/storage/CodaliFeedbackReviewIngestion.d.ts +50 -0
- package/dist/vendor/codali/storage/CodaliFeedbackReviewIngestion.d.ts.map +1 -0
- package/dist/vendor/codali/storage/CodaliFeedbackReviewIngestion.js +141 -0
- package/dist/vendor/codali/storage/CodaliStorageContracts.d.ts +337 -0
- package/dist/vendor/codali/storage/CodaliStorageContracts.d.ts.map +1 -0
- package/dist/vendor/codali/storage/CodaliStorageContracts.js +1555 -0
- package/dist/vendor/codali/storage/DatasetExportJob.d.ts +51 -0
- package/dist/vendor/codali/storage/DatasetExportJob.d.ts.map +1 -0
- package/dist/vendor/codali/storage/DatasetExportJob.js +358 -0
- package/dist/vendor/codali/storage/DatasetReviewQueue.d.ts +98 -0
- package/dist/vendor/codali/storage/DatasetReviewQueue.d.ts.map +1 -0
- package/dist/vendor/codali/storage/DatasetReviewQueue.js +426 -0
- package/dist/vendor/codali/storage/GatewayDatasetStore.d.ts +233 -0
- package/dist/vendor/codali/storage/GatewayDatasetStore.d.ts.map +1 -0
- package/dist/vendor/codali/storage/GatewayDatasetStore.js +2070 -0
- package/package.json +3 -3
|
@@ -0,0 +1,173 @@
|
|
|
1
|
+
import { type CodaliStorageDatasetKind, type CodaliStorageExportKind, type CodaliStorageObjectPrivacyFlags } from "../storage/CodaliStorageContracts.js";
|
|
2
|
+
import { CODALI_DATASET_REPLAY_FIXTURE_SCHEMA_VERSION } from "../storage/DatasetExportJob.js";
|
|
3
|
+
import type { DatasetEligibilityGateAcceptedExample, DatasetEligibilityGateRejectedExample } from "./DatasetEligibilityGate.js";
|
|
4
|
+
import type { DatasetExportManifestReaderResult } from "./DatasetExportManifestReader.js";
|
|
5
|
+
export declare const CODALI_EVAL_REPLAY_CANDIDATE_SCHEMA_VERSION: "codali.improvement.eval_replay_candidate.v1";
|
|
6
|
+
export type CodaliEvalReplayProposalArtifact = "eval";
|
|
7
|
+
export interface CodaliEvalReplayObjectRefSummary {
|
|
8
|
+
refId: string;
|
|
9
|
+
kind: string;
|
|
10
|
+
contentHash: string;
|
|
11
|
+
byteSize: number;
|
|
12
|
+
mimeType: string;
|
|
13
|
+
deletionGroupId: string;
|
|
14
|
+
privacyFlags: CodaliStorageObjectPrivacyFlags;
|
|
15
|
+
uri?: string;
|
|
16
|
+
mediaType?: string;
|
|
17
|
+
metadataKeys: string[];
|
|
18
|
+
}
|
|
19
|
+
export interface CodaliEvalReplayRecordShape {
|
|
20
|
+
requiredFields: string[];
|
|
21
|
+
optionalRefFields: string[];
|
|
22
|
+
presentFields: string[];
|
|
23
|
+
missingRequiredFields: string[];
|
|
24
|
+
bodyPolicy: "object_refs_only";
|
|
25
|
+
}
|
|
26
|
+
export interface CodaliEvalReplayRecordRefSummary {
|
|
27
|
+
recordId: string;
|
|
28
|
+
datasetKind?: CodaliStorageDatasetKind;
|
|
29
|
+
sourceGatewayRecordId?: string;
|
|
30
|
+
inputRef?: CodaliEvalReplayObjectRefSummary;
|
|
31
|
+
outputRef?: CodaliEvalReplayObjectRefSummary;
|
|
32
|
+
evidenceRefs: CodaliEvalReplayObjectRefSummary[];
|
|
33
|
+
quality?: {
|
|
34
|
+
score?: number;
|
|
35
|
+
reviewed?: boolean;
|
|
36
|
+
labels: string[];
|
|
37
|
+
};
|
|
38
|
+
metadataShape: {
|
|
39
|
+
keys: string[];
|
|
40
|
+
evalStage?: string;
|
|
41
|
+
exampleType?: string;
|
|
42
|
+
role?: string;
|
|
43
|
+
tool?: string;
|
|
44
|
+
sourceType?: string;
|
|
45
|
+
};
|
|
46
|
+
expectedShape: CodaliEvalReplayRecordShape;
|
|
47
|
+
}
|
|
48
|
+
export interface CodaliEvalReplayAcceptedEvidence {
|
|
49
|
+
recordId: string;
|
|
50
|
+
sourceGatewayRecordId?: string;
|
|
51
|
+
lineageKey: DatasetEligibilityGateAcceptedExample["lineageKey"];
|
|
52
|
+
deletionGroupIds: string[];
|
|
53
|
+
runIds: string[];
|
|
54
|
+
artifactTypes: string[];
|
|
55
|
+
preferenceSignals: string[];
|
|
56
|
+
priorityScore: number;
|
|
57
|
+
recordShape: CodaliEvalReplayRecordShape;
|
|
58
|
+
objectRefs: {
|
|
59
|
+
inputRef?: CodaliEvalReplayObjectRefSummary;
|
|
60
|
+
outputRef?: CodaliEvalReplayObjectRefSummary;
|
|
61
|
+
evidenceRefs: CodaliEvalReplayObjectRefSummary[];
|
|
62
|
+
};
|
|
63
|
+
}
|
|
64
|
+
export interface CodaliEvalReplayRejectedEvidence {
|
|
65
|
+
targetType: DatasetEligibilityGateRejectedExample["targetType"];
|
|
66
|
+
recordId?: string;
|
|
67
|
+
dedupeKey?: string;
|
|
68
|
+
failureLabels: string[];
|
|
69
|
+
reasons: Array<{
|
|
70
|
+
code: string;
|
|
71
|
+
message: string;
|
|
72
|
+
path?: string;
|
|
73
|
+
details?: Record<string, unknown>;
|
|
74
|
+
}>;
|
|
75
|
+
}
|
|
76
|
+
export interface CodaliEvalReplayExpectedShape {
|
|
77
|
+
schemaVersion: typeof CODALI_EVAL_REPLAY_CANDIDATE_SCHEMA_VERSION;
|
|
78
|
+
artifact: CodaliEvalReplayProposalArtifact;
|
|
79
|
+
evalFixture: {
|
|
80
|
+
requiredFields: string[];
|
|
81
|
+
caseRequiredFields: string[];
|
|
82
|
+
evidenceFields: string[];
|
|
83
|
+
bodyPolicy: "object_refs_only";
|
|
84
|
+
};
|
|
85
|
+
replayFixture: {
|
|
86
|
+
schemaVersion: typeof CODALI_DATASET_REPLAY_FIXTURE_SCHEMA_VERSION;
|
|
87
|
+
requiredFields: string[];
|
|
88
|
+
recordRequiredFields: string[];
|
|
89
|
+
bodyPolicy: "object_refs_only";
|
|
90
|
+
};
|
|
91
|
+
}
|
|
92
|
+
export interface CodaliEvalFixtureCandidate {
|
|
93
|
+
fixtureId: string;
|
|
94
|
+
sourceExportId: string;
|
|
95
|
+
caseCount: number;
|
|
96
|
+
cases: Array<{
|
|
97
|
+
caseId: string;
|
|
98
|
+
sourceRecordId: string;
|
|
99
|
+
datasetKind?: CodaliStorageDatasetKind;
|
|
100
|
+
expectedShape: CodaliEvalReplayRecordShape;
|
|
101
|
+
acceptedEvidenceRecordId: string;
|
|
102
|
+
rejectedEvidenceRecordIds: string[];
|
|
103
|
+
failureLabels: string[];
|
|
104
|
+
objectRefs: {
|
|
105
|
+
inputRef?: CodaliEvalReplayObjectRefSummary;
|
|
106
|
+
outputRef?: CodaliEvalReplayObjectRefSummary;
|
|
107
|
+
evidenceRefs: CodaliEvalReplayObjectRefSummary[];
|
|
108
|
+
};
|
|
109
|
+
}>;
|
|
110
|
+
}
|
|
111
|
+
export interface CodaliReplayFixtureCandidate {
|
|
112
|
+
fixtureId: string;
|
|
113
|
+
schemaVersion: typeof CODALI_DATASET_REPLAY_FIXTURE_SCHEMA_VERSION;
|
|
114
|
+
sourceExportId: string;
|
|
115
|
+
exportKind: CodaliStorageExportKind;
|
|
116
|
+
generatedAt: string;
|
|
117
|
+
bodyStorage: "object_ref" | "record_refs_only";
|
|
118
|
+
bodyRef?: CodaliEvalReplayObjectRefSummary;
|
|
119
|
+
recordCount: number;
|
|
120
|
+
records: CodaliEvalReplayRecordRefSummary[];
|
|
121
|
+
}
|
|
122
|
+
export interface CodaliEvalReplayImprovementCandidateSummary {
|
|
123
|
+
candidateId: string;
|
|
124
|
+
candidateKind: "eval_replay";
|
|
125
|
+
status: "proposed" | "blocked";
|
|
126
|
+
fixtureIds: {
|
|
127
|
+
evalFixtureId: string;
|
|
128
|
+
replayFixtureId: string;
|
|
129
|
+
};
|
|
130
|
+
sourceExportIds: string[];
|
|
131
|
+
sourceRecordIds: string[];
|
|
132
|
+
artifactIds: string[];
|
|
133
|
+
exampleCount: number;
|
|
134
|
+
objectBytes: number;
|
|
135
|
+
failureLabels: string[];
|
|
136
|
+
}
|
|
137
|
+
export interface CodaliEvalReplayCandidateBundle {
|
|
138
|
+
schemaVersion: typeof CODALI_EVAL_REPLAY_CANDIDATE_SCHEMA_VERSION;
|
|
139
|
+
artifact: CodaliEvalReplayProposalArtifact;
|
|
140
|
+
source: {
|
|
141
|
+
exportId: string;
|
|
142
|
+
manifestId: string;
|
|
143
|
+
manifestPath: string;
|
|
144
|
+
exportKind: CodaliStorageExportKind;
|
|
145
|
+
checksum: string;
|
|
146
|
+
recordCount: number;
|
|
147
|
+
primaryArtifactRef: CodaliEvalReplayObjectRefSummary;
|
|
148
|
+
replayFixtureRef?: CodaliEvalReplayObjectRefSummary;
|
|
149
|
+
};
|
|
150
|
+
generationPolicy: {
|
|
151
|
+
deterministic: true;
|
|
152
|
+
modifiesRuntimePrompts: false;
|
|
153
|
+
modifiesRuntimeCode: false;
|
|
154
|
+
bodyPolicy: "object_refs_only";
|
|
155
|
+
};
|
|
156
|
+
expectedShape: CodaliEvalReplayExpectedShape;
|
|
157
|
+
fixtureIds: {
|
|
158
|
+
evalFixtureId: string;
|
|
159
|
+
replayFixtureId: string;
|
|
160
|
+
};
|
|
161
|
+
acceptedEvidence: CodaliEvalReplayAcceptedEvidence[];
|
|
162
|
+
rejectedEvidence: CodaliEvalReplayRejectedEvidence[];
|
|
163
|
+
failureLabels: string[];
|
|
164
|
+
evalFixture: CodaliEvalFixtureCandidate;
|
|
165
|
+
replayFixture: CodaliReplayFixtureCandidate;
|
|
166
|
+
candidates: CodaliEvalReplayImprovementCandidateSummary[];
|
|
167
|
+
}
|
|
168
|
+
export interface BuildCodaliEvalReplayCandidateBundleInput {
|
|
169
|
+
inspection: DatasetExportManifestReaderResult;
|
|
170
|
+
artifact?: CodaliEvalReplayProposalArtifact;
|
|
171
|
+
}
|
|
172
|
+
export declare const buildCodaliEvalReplayCandidateBundle: (input: BuildCodaliEvalReplayCandidateBundleInput) => CodaliEvalReplayCandidateBundle;
|
|
173
|
+
//# sourceMappingURL=EvalReplayCandidateBuilder.d.ts.map
|
|
@@ -0,0 +1 @@
|
|
|
1
|
+
{"version":3,"file":"EvalReplayCandidateBuilder.d.ts","sourceRoot":"","sources":["../../src/improvement/EvalReplayCandidateBuilder.ts"],"names":[],"mappings":"AACA,OAAO,EACL,KAAK,wBAAwB,EAC7B,KAAK,uBAAuB,EAC5B,KAAK,+BAA+B,EAErC,MAAM,sCAAsC,CAAC;AAC9C,OAAO,EAAE,4CAA4C,EAAE,MAAM,gCAAgC,CAAC;AAC9F,OAAO,KAAK,EACV,qCAAqC,EACrC,qCAAqC,EAEtC,MAAM,6BAA6B,CAAC;AACrC,OAAO,KAAK,EAAE,iCAAiC,EAAE,MAAM,kCAAkC,CAAC;AAE1F,eAAO,MAAM,2CAA2C,EACtD,6CAAsD,CAAC;AAEzD,MAAM,MAAM,gCAAgC,GAAG,MAAM,CAAC;AAEtD,MAAM,WAAW,gCAAgC;IAC/C,KAAK,EAAE,MAAM,CAAC;IACd,IAAI,EAAE,MAAM,CAAC;IACb,WAAW,EAAE,MAAM,CAAC;IACpB,QAAQ,EAAE,MAAM,CAAC;IACjB,QAAQ,EAAE,MAAM,CAAC;IACjB,eAAe,EAAE,MAAM,CAAC;IACxB,YAAY,EAAE,+BAA+B,CAAC;IAC9C,GAAG,CAAC,EAAE,MAAM,CAAC;IACb,SAAS,CAAC,EAAE,MAAM,CAAC;IACnB,YAAY,EAAE,MAAM,EAAE,CAAC;CACxB;AAED,MAAM,WAAW,2BAA2B;IAC1C,cAAc,EAAE,MAAM,EAAE,CAAC;IACzB,iBAAiB,EAAE,MAAM,EAAE,CAAC;IAC5B,aAAa,EAAE,MAAM,EAAE,CAAC;IACxB,qBAAqB,EAAE,MAAM,EAAE,CAAC;IAChC,UAAU,EAAE,kBAAkB,CAAC;CAChC;AAED,MAAM,WAAW,gCAAgC;IAC/C,QAAQ,EAAE,MAAM,CAAC;IACjB,WAAW,CAAC,EAAE,wBAAwB,CAAC;IACvC,qBAAqB,CAAC,EAAE,MAAM,CAAC;IAC/B,QAAQ,CAAC,EAAE,gCAAgC,CAAC;IAC5C,SAAS,CAAC,EAAE,gCAAgC,CAAC;IAC7C,YAAY,EAAE,gCAAgC,EAAE,CAAC;IACjD,OAAO,CAAC,EAAE;QACR,KAAK,CAAC,EAAE,MAAM,CAAC;QACf,QAAQ,CAAC,EAAE,OAAO,CAAC;QACnB,MAAM,EAAE,MAAM,EAAE,CAAC;KAClB,CAAC;IACF,aAAa,EAAE;QACb,IAAI,EAAE,MAAM,EAAE,CAAC;QACf,SAAS,CAAC,EAAE,MAAM,CAAC;QACnB,WAAW,CAAC,EAAE,MAAM,CAAC;QACrB,IAAI,CAAC,EAAE,MAAM,CAAC;QACd,IAAI,CAAC,EAAE,MAAM,CAAC;QACd,UAAU,CAAC,EAAE,MAAM,CAAC;KACrB,CAAC;IACF,aAAa,EAAE,2BAA2B,CAAC;CAC5C;AAED,MAAM,WAAW,gCAAgC;IAC/C,QAAQ,EAAE,MAAM,CAAC;IACjB,qBAAqB,CAAC,EAAE,MAAM,CAAC;IAC/B,UAAU,EAAE,qCAAqC,CAAC,YAAY,CAAC,CAAC;IAChE,gBAAgB,EAAE,MAAM,EAAE,CAAC;IAC3B,MAAM,EAAE,MAAM,EAAE,CAAC;IACjB,aAAa,EAAE,MAAM,EAAE,CAAC;IACxB,iBAAiB,EAAE,MAAM,EAAE,CAAC;IAC5B,aAAa,EAAE,MAAM,CAAC;IACtB,WAAW,EAAE,2BAA2B,CAAC;IACzC,UAAU,EAAE;QACV,QAAQ,CAAC,EAAE,gCAAgC,CAAC;QAC5C,SAAS,CAAC,EAAE,gCAAgC,CAAC;QAC7C,YAAY,EAAE,gCAAgC,EAAE,CAAC;KAClD,CAAC;CACH;AAED,MAAM,WAAW,gCAAgC;IAC/C,UAAU,EAAE,qCAAqC,CAAC,YAAY,CAAC,CAAC;IAChE,QAAQ,CAAC,EAAE,MAAM,CAAC;IAClB,SAAS,CAAC,EAAE,MAAM,CAAC;IACnB,aAAa,EAAE,MAAM,EAAE,CAAC;IACxB,OAAO,EAAE,KAAK,CAAC;QACb,IAAI,EAAE,MAAM,CAAC;QACb,OAAO,EAAE,MAAM,CAAC;QAChB,IAAI,CAAC,EAAE,MAAM,CAAC;QACd,OAAO,CAAC,EAAE,MAAM,CAAC,MAAM,EAAE,OAAO,CAAC,CAAC;KACnC,CAAC,CAAC;CACJ;AAED,MAAM,WAAW,6BAA6B;IAC5C,aAAa,EAAE,OAAO,2CAA2C,CAAC;IAClE,QAAQ,EAAE,gCAAgC,CAAC;IAC3C,WAAW,EAAE;QACX,cAAc,EAAE,MAAM,EAAE,CAAC;QACzB,kBAAkB,EAAE,MAAM,EAAE,CAAC;QAC7B,cAAc,EAAE,MAAM,EAAE,CAAC;QACzB,UAAU,EAAE,kBAAkB,CAAC;KAChC,CAAC;IACF,aAAa,EAAE;QACb,aAAa,EAAE,OAAO,4CAA4C,CAAC;QACnE,cAAc,EAAE,MAAM,EAAE,CAAC;QACzB,oBAAoB,EAAE,MAAM,EAAE,CAAC;QAC/B,UAAU,EAAE,kBAAkB,CAAC;KAChC,CAAC;CACH;AAED,MAAM,WAAW,0BAA0B;IACzC,SAAS,EAAE,MAAM,CAAC;IAClB,cAAc,EAAE,MAAM,CAAC;IACvB,SAAS,EAAE,MAAM,CAAC;IAClB,KAAK,EAAE,KAAK,CAAC;QACX,MAAM,EAAE,MAAM,CAAC;QACf,cAAc,EAAE,MAAM,CAAC;QACvB,WAAW,CAAC,EAAE,wBAAwB,CAAC;QACvC,aAAa,EAAE,2BAA2B,CAAC;QAC3C,wBAAwB,EAAE,MAAM,CAAC;QACjC,yBAAyB,EAAE,MAAM,EAAE,CAAC;QACpC,aAAa,EAAE,MAAM,EAAE,CAAC;QACxB,UAAU,EAAE;YACV,QAAQ,CAAC,EAAE,gCAAgC,CAAC;YAC5C,SAAS,CAAC,EAAE,gCAAgC,CAAC;YAC7C,YAAY,EAAE,gCAAgC,EAAE,CAAC;SAClD,CAAC;KACH,CAAC,CAAC;CACJ;AAED,MAAM,WAAW,4BAA4B;IAC3C,SAAS,EAAE,MAAM,CAAC;IAClB,aAAa,EAAE,OAAO,4CAA4C,CAAC;IACnE,cAAc,EAAE,MAAM,CAAC;IACvB,UAAU,EAAE,uBAAuB,CAAC;IACpC,WAAW,EAAE,MAAM,CAAC;IACpB,WAAW,EAAE,YAAY,GAAG,kBAAkB,CAAC;IAC/C,OAAO,CAAC,EAAE,gCAAgC,CAAC;IAC3C,WAAW,EAAE,MAAM,CAAC;IACpB,OAAO,EAAE,gCAAgC,EAAE,CAAC;CAC7C;AAED,MAAM,WAAW,2CAA2C;IAC1D,WAAW,EAAE,MAAM,CAAC;IACpB,aAAa,EAAE,aAAa,CAAC;IAC7B,MAAM,EAAE,UAAU,GAAG,SAAS,CAAC;IAC/B,UAAU,EAAE;QACV,aAAa,EAAE,MAAM,CAAC;QACtB,eAAe,EAAE,MAAM,CAAC;KACzB,CAAC;IACF,eAAe,EAAE,MAAM,EAAE,CAAC;IAC1B,eAAe,EAAE,MAAM,EAAE,CAAC;IAC1B,WAAW,EAAE,MAAM,EAAE,CAAC;IACtB,YAAY,EAAE,MAAM,CAAC;IACrB,WAAW,EAAE,MAAM,CAAC;IACpB,aAAa,EAAE,MAAM,EAAE,CAAC;CACzB;AAED,MAAM,WAAW,+BAA+B;IAC9C,aAAa,EAAE,OAAO,2CAA2C,CAAC;IAClE,QAAQ,EAAE,gCAAgC,CAAC;IAC3C,MAAM,EAAE;QACN,QAAQ,EAAE,MAAM,CAAC;QACjB,UAAU,EAAE,MAAM,CAAC;QACnB,YAAY,EAAE,MAAM,CAAC;QACrB,UAAU,EAAE,uBAAuB,CAAC;QACpC,QAAQ,EAAE,MAAM,CAAC;QACjB,WAAW,EAAE,MAAM,CAAC;QACpB,kBAAkB,EAAE,gCAAgC,CAAC;QACrD,gBAAgB,CAAC,EAAE,gCAAgC,CAAC;KACrD,CAAC;IACF,gBAAgB,EAAE;QAChB,aAAa,EAAE,IAAI,CAAC;QACpB,sBAAsB,EAAE,KAAK,CAAC;QAC9B,mBAAmB,EAAE,KAAK,CAAC;QAC3B,UAAU,EAAE,kBAAkB,CAAC;KAChC,CAAC;IACF,aAAa,EAAE,6BAA6B,CAAC;IAC7C,UAAU,EAAE;QACV,aAAa,EAAE,MAAM,CAAC;QACtB,eAAe,EAAE,MAAM,CAAC;KACzB,CAAC;IACF,gBAAgB,EAAE,gCAAgC,EAAE,CAAC;IACrD,gBAAgB,EAAE,gCAAgC,EAAE,CAAC;IACrD,aAAa,EAAE,MAAM,EAAE,CAAC;IACxB,WAAW,EAAE,0BAA0B,CAAC;IACxC,aAAa,EAAE,4BAA4B,CAAC;IAC5C,UAAU,EAAE,2CAA2C,EAAE,CAAC;CAC3D;AAED,MAAM,WAAW,yCAAyC;IACxD,UAAU,EAAE,iCAAiC,CAAC;IAC9C,QAAQ,CAAC,EAAE,gCAAgC,CAAC;CAC7C;AAuTD,eAAO,MAAM,oCAAoC,GAC/C,OAAO,yCAAyC,KAC/C,+BAyHF,CAAC"}
|
|
@@ -0,0 +1,363 @@
|
|
|
1
|
+
import { createHash } from "node:crypto";
|
|
2
|
+
import { CODALI_DATASET_REPLAY_FIXTURE_SCHEMA_VERSION } from "../storage/DatasetExportJob.js";
|
|
3
|
+
export const CODALI_EVAL_REPLAY_CANDIDATE_SCHEMA_VERSION = "codali.improvement.eval_replay_candidate.v1";
|
|
4
|
+
const isRecord = (value) => Boolean(value && typeof value === "object" && !Array.isArray(value));
|
|
5
|
+
const stableJson = (value) => {
|
|
6
|
+
if (Array.isArray(value))
|
|
7
|
+
return `[${value.map(stableJson).join(",")}]`;
|
|
8
|
+
if (!isRecord(value))
|
|
9
|
+
return JSON.stringify(value);
|
|
10
|
+
return `{${Object.keys(value).sort().map((key) => `${JSON.stringify(key)}:${stableJson(value[key])}`).join(",")}}`;
|
|
11
|
+
};
|
|
12
|
+
const sha256Hex = (value) => createHash("sha256").update(stableJson(value)).digest("hex");
|
|
13
|
+
const stableId = (prefix, value) => `${prefix}-${sha256Hex(value).slice(0, 16)}`;
|
|
14
|
+
const uniqueSorted = (values) => Array.from(new Set(values
|
|
15
|
+
.map((value) => value?.trim())
|
|
16
|
+
.filter((value) => Boolean(value))))
|
|
17
|
+
.sort();
|
|
18
|
+
const readString = (record, keys) => {
|
|
19
|
+
for (const key of keys) {
|
|
20
|
+
const value = record?.[key];
|
|
21
|
+
if (typeof value === "string" && value.trim())
|
|
22
|
+
return value.trim();
|
|
23
|
+
}
|
|
24
|
+
return undefined;
|
|
25
|
+
};
|
|
26
|
+
const readNumber = (record, key) => {
|
|
27
|
+
const value = record?.[key];
|
|
28
|
+
return typeof value === "number" && Number.isFinite(value) ? value : undefined;
|
|
29
|
+
};
|
|
30
|
+
const readBoolean = (record, key) => {
|
|
31
|
+
const value = record?.[key];
|
|
32
|
+
return typeof value === "boolean" ? value : undefined;
|
|
33
|
+
};
|
|
34
|
+
const stringList = (value) => Array.isArray(value)
|
|
35
|
+
? value.filter((entry) => typeof entry === "string")
|
|
36
|
+
: [];
|
|
37
|
+
const objectRefFromValue = (value) => isRecord(value) &&
|
|
38
|
+
typeof value.refId === "string" &&
|
|
39
|
+
typeof value.contentHash === "string" &&
|
|
40
|
+
typeof value.byteSize === "number" &&
|
|
41
|
+
typeof value.mimeType === "string" &&
|
|
42
|
+
typeof value.deletionGroupId === "string" &&
|
|
43
|
+
isRecord(value.privacyFlags)
|
|
44
|
+
? value
|
|
45
|
+
: undefined;
|
|
46
|
+
const summarizeObjectRef = (ref) => ({
|
|
47
|
+
refId: ref.refId,
|
|
48
|
+
kind: ref.kind,
|
|
49
|
+
contentHash: ref.contentHash,
|
|
50
|
+
byteSize: ref.byteSize,
|
|
51
|
+
mimeType: ref.mimeType,
|
|
52
|
+
deletionGroupId: ref.deletionGroupId,
|
|
53
|
+
privacyFlags: ref.privacyFlags,
|
|
54
|
+
...(ref.uri ? { uri: ref.uri } : {}),
|
|
55
|
+
...(ref.mediaType ? { mediaType: ref.mediaType } : {}),
|
|
56
|
+
metadataKeys: isRecord(ref.metadata) ? Object.keys(ref.metadata).sort() : [],
|
|
57
|
+
});
|
|
58
|
+
const refsForRow = (row) => {
|
|
59
|
+
const record = isRecord(row) ? row : {};
|
|
60
|
+
const evidenceRefs = Array.isArray(record.evidenceRefs)
|
|
61
|
+
? record.evidenceRefs.map(objectRefFromValue)
|
|
62
|
+
.filter((ref) => Boolean(ref))
|
|
63
|
+
: [];
|
|
64
|
+
return {
|
|
65
|
+
inputRef: objectRefFromValue(record.inputRef),
|
|
66
|
+
outputRef: objectRefFromValue(record.outputRef),
|
|
67
|
+
evidenceRefs,
|
|
68
|
+
};
|
|
69
|
+
};
|
|
70
|
+
const recordIdForRow = (row) => isRecord(row) ? readString(row, ["recordId", "record_id"]) : undefined;
|
|
71
|
+
const rowsByRecordId = (rows) => {
|
|
72
|
+
const output = new Map();
|
|
73
|
+
for (const row of rows) {
|
|
74
|
+
if (!isRecord(row))
|
|
75
|
+
continue;
|
|
76
|
+
const recordId = recordIdForRow(row);
|
|
77
|
+
if (recordId)
|
|
78
|
+
output.set(recordId, row);
|
|
79
|
+
}
|
|
80
|
+
return output;
|
|
81
|
+
};
|
|
82
|
+
const expectedRecordShape = (row) => {
|
|
83
|
+
const presentFields = isRecord(row) ? Object.keys(row).sort() : [];
|
|
84
|
+
const requiredFields = ["datasetKind", "inputRef", "privacy", "recordId"];
|
|
85
|
+
return {
|
|
86
|
+
requiredFields,
|
|
87
|
+
optionalRefFields: ["outputRef", "evidenceRefs"],
|
|
88
|
+
presentFields,
|
|
89
|
+
missingRequiredFields: requiredFields
|
|
90
|
+
.filter((field) => !presentFields.includes(field)),
|
|
91
|
+
bodyPolicy: "object_refs_only",
|
|
92
|
+
};
|
|
93
|
+
};
|
|
94
|
+
const metadataShapeForRow = (row) => {
|
|
95
|
+
const metadata = isRecord(row.metadata) ? row.metadata : {};
|
|
96
|
+
return {
|
|
97
|
+
keys: Object.keys(metadata).sort(),
|
|
98
|
+
...(readString(metadata, ["evalStage", "eval_stage", "gatewayEvalStage", "gateway_eval_stage"])
|
|
99
|
+
? { evalStage: readString(metadata, ["evalStage", "eval_stage", "gatewayEvalStage", "gateway_eval_stage"]) }
|
|
100
|
+
: {}),
|
|
101
|
+
...(readString(metadata, ["exampleType", "example_type"])
|
|
102
|
+
? { exampleType: readString(metadata, ["exampleType", "example_type"]) }
|
|
103
|
+
: {}),
|
|
104
|
+
...(readString(metadata, ["role"]) ? { role: readString(metadata, ["role"]) } : {}),
|
|
105
|
+
...(readString(metadata, ["tool"]) ? { tool: readString(metadata, ["tool"]) } : {}),
|
|
106
|
+
...(readString(metadata, ["sourceType", "source_type"])
|
|
107
|
+
? { sourceType: readString(metadata, ["sourceType", "source_type"]) }
|
|
108
|
+
: {}),
|
|
109
|
+
};
|
|
110
|
+
};
|
|
111
|
+
const summarizeRecordRow = (row) => {
|
|
112
|
+
const refs = refsForRow(row);
|
|
113
|
+
const quality = isRecord(row.quality) ? row.quality : undefined;
|
|
114
|
+
return {
|
|
115
|
+
recordId: readString(row, ["recordId", "record_id"]) ?? "unknown-record",
|
|
116
|
+
...(readString(row, ["datasetKind", "dataset_kind"])
|
|
117
|
+
? { datasetKind: readString(row, ["datasetKind", "dataset_kind"]) }
|
|
118
|
+
: {}),
|
|
119
|
+
...(readString(row, ["sourceGatewayRecordId", "source_gateway_record_id"])
|
|
120
|
+
? { sourceGatewayRecordId: readString(row, ["sourceGatewayRecordId", "source_gateway_record_id"]) }
|
|
121
|
+
: {}),
|
|
122
|
+
...(refs.inputRef ? { inputRef: summarizeObjectRef(refs.inputRef) } : {}),
|
|
123
|
+
...(refs.outputRef ? { outputRef: summarizeObjectRef(refs.outputRef) } : {}),
|
|
124
|
+
evidenceRefs: refs.evidenceRefs.map(summarizeObjectRef),
|
|
125
|
+
...(quality ? {
|
|
126
|
+
quality: {
|
|
127
|
+
score: readNumber(quality, "score"),
|
|
128
|
+
reviewed: readBoolean(quality, "reviewed"),
|
|
129
|
+
labels: stringList(quality.labels).sort(),
|
|
130
|
+
},
|
|
131
|
+
} : {}),
|
|
132
|
+
metadataShape: metadataShapeForRow(row),
|
|
133
|
+
expectedShape: expectedRecordShape(row),
|
|
134
|
+
};
|
|
135
|
+
};
|
|
136
|
+
const failureLabelsForReasons = (reasons) => uniqueSorted(reasons.map((reason) => reason.code));
|
|
137
|
+
const rejectedEvidence = (rejected) => rejected.map((item) => ({
|
|
138
|
+
targetType: item.targetType,
|
|
139
|
+
...(item.recordId ? { recordId: item.recordId } : {}),
|
|
140
|
+
...(item.dedupeKey ? { dedupeKey: item.dedupeKey } : {}),
|
|
141
|
+
failureLabels: failureLabelsForReasons(item.reasons),
|
|
142
|
+
reasons: item.reasons.map((reason) => ({
|
|
143
|
+
code: reason.code,
|
|
144
|
+
message: reason.message,
|
|
145
|
+
...(reason.path ? { path: reason.path } : {}),
|
|
146
|
+
...(reason.details ? { details: reason.details } : {}),
|
|
147
|
+
})),
|
|
148
|
+
}));
|
|
149
|
+
const acceptedEvidence = (accepted, rowMap) => accepted.map((item) => {
|
|
150
|
+
const row = rowMap.get(item.recordId);
|
|
151
|
+
const refs = refsForRow(row);
|
|
152
|
+
return {
|
|
153
|
+
recordId: item.recordId,
|
|
154
|
+
...(item.sourceGatewayRecordId ? { sourceGatewayRecordId: item.sourceGatewayRecordId } : {}),
|
|
155
|
+
lineageKey: item.lineageKey,
|
|
156
|
+
deletionGroupIds: item.deletionGroupIds,
|
|
157
|
+
runIds: item.runIds,
|
|
158
|
+
artifactTypes: item.artifactTypes,
|
|
159
|
+
preferenceSignals: item.preferenceSignals,
|
|
160
|
+
priorityScore: item.priorityScore,
|
|
161
|
+
recordShape: expectedRecordShape(row),
|
|
162
|
+
objectRefs: {
|
|
163
|
+
...(refs.inputRef ? { inputRef: summarizeObjectRef(refs.inputRef) } : {}),
|
|
164
|
+
...(refs.outputRef ? { outputRef: summarizeObjectRef(refs.outputRef) } : {}),
|
|
165
|
+
evidenceRefs: refs.evidenceRefs.map(summarizeObjectRef),
|
|
166
|
+
},
|
|
167
|
+
};
|
|
168
|
+
});
|
|
169
|
+
const allFailureLabels = (inspection) => {
|
|
170
|
+
const curationLabels = inspection.curationReport.rejected
|
|
171
|
+
.flatMap((item) => item.reasons.map((reason) => reason.code));
|
|
172
|
+
const warningLabels = inspection.warnings.map((warning) => warning.code);
|
|
173
|
+
return uniqueSorted([
|
|
174
|
+
...curationLabels,
|
|
175
|
+
...warningLabels,
|
|
176
|
+
inspection.curationReport.acceptedCount === 0 ? "no_accepted_examples" : undefined,
|
|
177
|
+
inspection.manifest.exportKind !== "eval-replay" ? "export_kind_not_eval_replay" : undefined,
|
|
178
|
+
inspection.curationReport.artifactReadAllowed ? undefined : "artifact_payload_read_blocked",
|
|
179
|
+
inspection.curationReport.lineageValid ? undefined : "lineage_invalid",
|
|
180
|
+
]);
|
|
181
|
+
};
|
|
182
|
+
const replayFixtureRefForInspection = (inspection) => inspection.manifest.artifactRefs.find((ref) => {
|
|
183
|
+
const metadata = isRecord(ref.metadata) ? ref.metadata : {};
|
|
184
|
+
return readString(metadata, ["artifactType", "artifact_type"]) === "replay_fixture";
|
|
185
|
+
});
|
|
186
|
+
const primaryArtifactRefForInspection = (inspection) => {
|
|
187
|
+
const ref = inspection.primaryArtifact?.ref ?? inspection.manifest.artifactRefs[0];
|
|
188
|
+
if (!ref) {
|
|
189
|
+
throw new Error("CODALI_EVAL_REPLAY_PRIMARY_ARTIFACT_MISSING");
|
|
190
|
+
}
|
|
191
|
+
return ref;
|
|
192
|
+
};
|
|
193
|
+
const expectedShape = (artifact) => ({
|
|
194
|
+
schemaVersion: CODALI_EVAL_REPLAY_CANDIDATE_SCHEMA_VERSION,
|
|
195
|
+
artifact,
|
|
196
|
+
evalFixture: {
|
|
197
|
+
requiredFields: [
|
|
198
|
+
"fixtureId",
|
|
199
|
+
"cases",
|
|
200
|
+
"acceptedEvidence",
|
|
201
|
+
"rejectedEvidence",
|
|
202
|
+
"failureLabels",
|
|
203
|
+
],
|
|
204
|
+
caseRequiredFields: [
|
|
205
|
+
"caseId",
|
|
206
|
+
"sourceRecordId",
|
|
207
|
+
"expectedShape",
|
|
208
|
+
"objectRefs",
|
|
209
|
+
],
|
|
210
|
+
evidenceFields: [
|
|
211
|
+
"recordId",
|
|
212
|
+
"lineageKey",
|
|
213
|
+
"artifactTypes",
|
|
214
|
+
"preferenceSignals",
|
|
215
|
+
],
|
|
216
|
+
bodyPolicy: "object_refs_only",
|
|
217
|
+
},
|
|
218
|
+
replayFixture: {
|
|
219
|
+
schemaVersion: CODALI_DATASET_REPLAY_FIXTURE_SCHEMA_VERSION,
|
|
220
|
+
requiredFields: ["fixtureId", "schemaVersion", "records"],
|
|
221
|
+
recordRequiredFields: ["recordId", "datasetKind", "inputRef"],
|
|
222
|
+
bodyPolicy: "object_refs_only",
|
|
223
|
+
},
|
|
224
|
+
});
|
|
225
|
+
const objectBytesForRecords = (rows, replayFixtureRef) => {
|
|
226
|
+
const refs = new Map();
|
|
227
|
+
for (const row of rows) {
|
|
228
|
+
const rowRefs = refsForRow(row);
|
|
229
|
+
for (const ref of [
|
|
230
|
+
rowRefs.inputRef,
|
|
231
|
+
rowRefs.outputRef,
|
|
232
|
+
...rowRefs.evidenceRefs,
|
|
233
|
+
]) {
|
|
234
|
+
if (ref)
|
|
235
|
+
refs.set(`${ref.refId}:${ref.contentHash}`, ref);
|
|
236
|
+
}
|
|
237
|
+
}
|
|
238
|
+
if (replayFixtureRef) {
|
|
239
|
+
refs.set(`${replayFixtureRef.refId}:${replayFixtureRef.contentHash}`, replayFixtureRef);
|
|
240
|
+
}
|
|
241
|
+
return Array.from(refs.values())
|
|
242
|
+
.reduce((total, ref) => total + ref.byteSize, 0);
|
|
243
|
+
};
|
|
244
|
+
export const buildCodaliEvalReplayCandidateBundle = (input) => {
|
|
245
|
+
const artifact = input.artifact ?? "eval";
|
|
246
|
+
const inspection = input.inspection;
|
|
247
|
+
const rowMap = rowsByRecordId(inspection.primaryArtifactRows);
|
|
248
|
+
const acceptedRows = inspection.curationReport.acceptedRecordIds
|
|
249
|
+
.map((recordId) => rowMap.get(recordId))
|
|
250
|
+
.filter((row) => Boolean(row));
|
|
251
|
+
const replayFixtureRef = replayFixtureRefForInspection(inspection);
|
|
252
|
+
const failureLabels = allFailureLabels(inspection);
|
|
253
|
+
const idSeed = {
|
|
254
|
+
artifact,
|
|
255
|
+
manifestId: inspection.manifest.manifestId,
|
|
256
|
+
checksum: inspection.manifest.checksum,
|
|
257
|
+
acceptedRecordIds: inspection.curationReport.acceptedRecordIds,
|
|
258
|
+
failureLabels,
|
|
259
|
+
};
|
|
260
|
+
const evalFixtureId = stableId("eval-fixture", idSeed);
|
|
261
|
+
const replayFixtureId = stableId("replay-fixture", {
|
|
262
|
+
...idSeed,
|
|
263
|
+
replayFixtureContentHash: replayFixtureRef?.contentHash,
|
|
264
|
+
});
|
|
265
|
+
const candidateId = stableId("candidate", {
|
|
266
|
+
...idSeed,
|
|
267
|
+
evalFixtureId,
|
|
268
|
+
replayFixtureId,
|
|
269
|
+
});
|
|
270
|
+
const accepted = acceptedEvidence(inspection.curationReport.accepted, rowMap);
|
|
271
|
+
const rejected = rejectedEvidence(inspection.curationReport.rejected);
|
|
272
|
+
const replayRecords = acceptedRows.map(summarizeRecordRow);
|
|
273
|
+
const evalCases = replayRecords.map((record) => ({
|
|
274
|
+
caseId: stableId("eval-case", {
|
|
275
|
+
evalFixtureId,
|
|
276
|
+
recordId: record.recordId,
|
|
277
|
+
inputHash: record.inputRef?.contentHash,
|
|
278
|
+
outputHash: record.outputRef?.contentHash,
|
|
279
|
+
}),
|
|
280
|
+
sourceRecordId: record.recordId,
|
|
281
|
+
...(record.datasetKind ? { datasetKind: record.datasetKind } : {}),
|
|
282
|
+
expectedShape: record.expectedShape,
|
|
283
|
+
acceptedEvidenceRecordId: record.recordId,
|
|
284
|
+
rejectedEvidenceRecordIds: rejected
|
|
285
|
+
.filter((item) => item.recordId === record.recordId)
|
|
286
|
+
.map((item) => item.recordId)
|
|
287
|
+
.filter((recordId) => Boolean(recordId)),
|
|
288
|
+
failureLabels: rejected
|
|
289
|
+
.filter((item) => item.recordId === record.recordId)
|
|
290
|
+
.flatMap((item) => item.failureLabels),
|
|
291
|
+
objectRefs: {
|
|
292
|
+
...(record.inputRef ? { inputRef: record.inputRef } : {}),
|
|
293
|
+
...(record.outputRef ? { outputRef: record.outputRef } : {}),
|
|
294
|
+
evidenceRefs: record.evidenceRefs,
|
|
295
|
+
},
|
|
296
|
+
}));
|
|
297
|
+
const candidateStatus = acceptedRows.length > 0 &&
|
|
298
|
+
inspection.manifest.exportKind === "eval-replay" &&
|
|
299
|
+
inspection.curationReport.artifactReadAllowed &&
|
|
300
|
+
inspection.curationReport.lineageValid
|
|
301
|
+
? "proposed"
|
|
302
|
+
: "blocked";
|
|
303
|
+
return {
|
|
304
|
+
schemaVersion: CODALI_EVAL_REPLAY_CANDIDATE_SCHEMA_VERSION,
|
|
305
|
+
artifact,
|
|
306
|
+
source: {
|
|
307
|
+
exportId: inspection.exportId,
|
|
308
|
+
manifestId: inspection.manifest.manifestId,
|
|
309
|
+
manifestPath: inspection.manifestPath,
|
|
310
|
+
exportKind: inspection.manifest.exportKind,
|
|
311
|
+
checksum: inspection.manifest.checksum,
|
|
312
|
+
recordCount: inspection.manifest.recordCount,
|
|
313
|
+
primaryArtifactRef: summarizeObjectRef(primaryArtifactRefForInspection(inspection)),
|
|
314
|
+
...(replayFixtureRef ? { replayFixtureRef: summarizeObjectRef(replayFixtureRef) } : {}),
|
|
315
|
+
},
|
|
316
|
+
generationPolicy: {
|
|
317
|
+
deterministic: true,
|
|
318
|
+
modifiesRuntimePrompts: false,
|
|
319
|
+
modifiesRuntimeCode: false,
|
|
320
|
+
bodyPolicy: "object_refs_only",
|
|
321
|
+
},
|
|
322
|
+
expectedShape: expectedShape(artifact),
|
|
323
|
+
fixtureIds: {
|
|
324
|
+
evalFixtureId,
|
|
325
|
+
replayFixtureId,
|
|
326
|
+
},
|
|
327
|
+
acceptedEvidence: accepted,
|
|
328
|
+
rejectedEvidence: rejected,
|
|
329
|
+
failureLabels,
|
|
330
|
+
evalFixture: {
|
|
331
|
+
fixtureId: evalFixtureId,
|
|
332
|
+
sourceExportId: inspection.manifest.manifestId,
|
|
333
|
+
caseCount: evalCases.length,
|
|
334
|
+
cases: evalCases,
|
|
335
|
+
},
|
|
336
|
+
replayFixture: {
|
|
337
|
+
fixtureId: replayFixtureId,
|
|
338
|
+
schemaVersion: CODALI_DATASET_REPLAY_FIXTURE_SCHEMA_VERSION,
|
|
339
|
+
sourceExportId: inspection.manifest.manifestId,
|
|
340
|
+
exportKind: inspection.manifest.exportKind,
|
|
341
|
+
generatedAt: inspection.manifest.createdAt,
|
|
342
|
+
bodyStorage: replayFixtureRef ? "object_ref" : "record_refs_only",
|
|
343
|
+
...(replayFixtureRef ? { bodyRef: summarizeObjectRef(replayFixtureRef) } : {}),
|
|
344
|
+
recordCount: replayRecords.length,
|
|
345
|
+
records: replayRecords,
|
|
346
|
+
},
|
|
347
|
+
candidates: [{
|
|
348
|
+
candidateId,
|
|
349
|
+
candidateKind: "eval_replay",
|
|
350
|
+
status: candidateStatus,
|
|
351
|
+
fixtureIds: {
|
|
352
|
+
evalFixtureId,
|
|
353
|
+
replayFixtureId,
|
|
354
|
+
},
|
|
355
|
+
sourceExportIds: [inspection.manifest.manifestId],
|
|
356
|
+
sourceRecordIds: inspection.curationReport.acceptedRecordIds,
|
|
357
|
+
artifactIds: inspection.manifest.artifactRefs.map((ref) => ref.refId),
|
|
358
|
+
exampleCount: acceptedRows.length,
|
|
359
|
+
objectBytes: objectBytesForRecords(acceptedRows, replayFixtureRef),
|
|
360
|
+
failureLabels,
|
|
361
|
+
}],
|
|
362
|
+
};
|
|
363
|
+
};
|