@devflow-tools/benchmark 0.17.8 → 0.18.1
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/LICENSE +21 -0
- package/dist/accuracy-command-options.d.ts +3 -0
- package/dist/accuracy-command-options.d.ts.map +1 -0
- package/dist/accuracy-command-options.js +17 -0
- package/dist/accuracy-command-options.js.map +1 -0
- package/dist/context-symbol-quality-command.d.ts +59 -0
- package/dist/context-symbol-quality-command.d.ts.map +1 -0
- package/dist/context-symbol-quality-command.js +482 -0
- package/dist/context-symbol-quality-command.js.map +1 -0
- package/dist/evaluation-bundle-builder.d.ts +48 -0
- package/dist/evaluation-bundle-builder.d.ts.map +1 -0
- package/dist/evaluation-bundle-builder.js +496 -0
- package/dist/evaluation-bundle-builder.js.map +1 -0
- package/dist/evaluation-bundle-command.d.ts +2 -0
- package/dist/evaluation-bundle-command.d.ts.map +1 -0
- package/dist/evaluation-bundle-command.js +15 -0
- package/dist/evaluation-bundle-command.js.map +1 -0
- package/dist/evaluation-bundle.d.ts +55 -0
- package/dist/evaluation-bundle.d.ts.map +1 -0
- package/dist/evaluation-bundle.js +272 -0
- package/dist/evaluation-bundle.js.map +1 -0
- package/dist/evaluation-run-state.d.ts +23 -0
- package/dist/evaluation-run-state.d.ts.map +1 -0
- package/dist/evaluation-run-state.js +115 -0
- package/dist/evaluation-run-state.js.map +1 -0
- package/dist/evidence-quality-metrics.d.ts +15 -0
- package/dist/evidence-quality-metrics.d.ts.map +1 -1
- package/dist/evidence-quality-metrics.js +51 -1
- package/dist/evidence-quality-metrics.js.map +1 -1
- package/dist/graders/deterministic.d.ts.map +1 -1
- package/dist/graders/deterministic.js +5 -3
- package/dist/graders/deterministic.js.map +1 -1
- package/dist/index.d.ts +15 -2
- package/dist/index.d.ts.map +1 -1
- package/dist/index.js +12 -1
- package/dist/index.js.map +1 -1
- package/dist/release-quality-bundle-command.d.ts +3 -0
- package/dist/release-quality-bundle-command.d.ts.map +1 -0
- package/dist/release-quality-bundle-command.js +72 -0
- package/dist/release-quality-bundle-command.js.map +1 -0
- package/dist/release-quality-bundle-evaluator.d.ts +61 -0
- package/dist/release-quality-bundle-evaluator.d.ts.map +1 -0
- package/dist/release-quality-bundle-evaluator.js +321 -0
- package/dist/release-quality-bundle-evaluator.js.map +1 -0
- package/dist/release-quality-bundle-materializer.d.ts +79 -0
- package/dist/release-quality-bundle-materializer.d.ts.map +1 -0
- package/dist/release-quality-bundle-materializer.js +395 -0
- package/dist/release-quality-bundle-materializer.js.map +1 -0
- package/dist/release-quality-comparison.d.ts +34 -0
- package/dist/release-quality-comparison.d.ts.map +1 -0
- package/dist/release-quality-comparison.js +60 -0
- package/dist/release-quality-comparison.js.map +1 -0
- package/dist/release-quality-evaluation-command.d.ts +42 -0
- package/dist/release-quality-evaluation-command.d.ts.map +1 -0
- package/dist/release-quality-evaluation-command.js +91 -0
- package/dist/release-quality-evaluation-command.js.map +1 -0
- package/dist/release-quality-host-proof.d.ts +30 -0
- package/dist/release-quality-host-proof.d.ts.map +1 -0
- package/dist/release-quality-host-proof.js +109 -0
- package/dist/release-quality-host-proof.js.map +1 -0
- package/dist/release-quality-output-cleanup.d.ts +2 -0
- package/dist/release-quality-output-cleanup.d.ts.map +1 -0
- package/dist/release-quality-output-cleanup.js +18 -0
- package/dist/release-quality-output-cleanup.js.map +1 -0
- package/dist/release-quality-policy.d.ts +87 -0
- package/dist/release-quality-policy.d.ts.map +1 -0
- package/dist/release-quality-policy.js +179 -0
- package/dist/release-quality-policy.js.map +1 -0
- package/dist/release-quality-query-sampling.d.ts +36 -0
- package/dist/release-quality-query-sampling.d.ts.map +1 -0
- package/dist/release-quality-query-sampling.js +87 -0
- package/dist/release-quality-query-sampling.js.map +1 -0
- package/dist/release-quality-source-snapshot.d.ts +18 -0
- package/dist/release-quality-source-snapshot.d.ts.map +1 -0
- package/dist/release-quality-source-snapshot.js +46 -0
- package/dist/release-quality-source-snapshot.js.map +1 -0
- package/dist/retrieval-quality-corpus.d.ts +0 -1
- package/dist/retrieval-quality-corpus.d.ts.map +1 -1
- package/dist/retrieval-quality-corpus.js +1 -1
- package/dist/retrieval-quality-corpus.js.map +1 -1
- package/dist/retrieval-quality.d.ts.map +1 -1
- package/dist/retrieval-quality.js +29 -26
- package/dist/retrieval-quality.js.map +1 -1
- package/dist/run-accuracy-tests.js +82 -2
- package/dist/run-accuracy-tests.js.map +1 -1
- package/dist/suite-runner.js +1 -1
- package/dist/suite-runner.js.map +1 -1
- package/dist/unified-runtime-corpus.d.ts +64 -0
- package/dist/unified-runtime-corpus.d.ts.map +1 -0
- package/dist/unified-runtime-corpus.js +161 -0
- package/dist/unified-runtime-corpus.js.map +1 -0
- package/dist/unified-runtime-evaluation.d.ts +133 -0
- package/dist/unified-runtime-evaluation.d.ts.map +1 -0
- package/dist/unified-runtime-evaluation.js +388 -0
- package/dist/unified-runtime-evaluation.js.map +1 -0
- package/dist/unified-runtime-observation.d.ts +101 -0
- package/dist/unified-runtime-observation.d.ts.map +1 -0
- package/dist/unified-runtime-observation.js +2 -0
- package/dist/unified-runtime-observation.js.map +1 -0
- package/dist/unified-runtime-runner.d.ts +73 -0
- package/dist/unified-runtime-runner.d.ts.map +1 -0
- package/dist/unified-runtime-runner.js +620 -0
- package/dist/unified-runtime-runner.js.map +1 -0
- package/fixtures/evidence-runtime-corpus.json +0 -100
- package/fixtures/release-quality/bundles.json +1 -0
- package/fixtures/release-quality/calibration.json +602 -0
- package/fixtures/release-quality/ecc/corpus.json +257 -0
- package/fixtures/release-quality/ecc/input.json +62 -0
- package/fixtures/release-quality/llm-wiki/corpus.json +263 -0
- package/fixtures/release-quality/llm-wiki/input.json +72 -0
- package/fixtures/release-quality/mem0/corpus.json +288 -0
- package/fixtures/release-quality/mem0/input.json +68 -0
- package/fixtures/release-quality/projects.json +29 -0
- package/fixtures/release-quality/release-quality-policy.json +232 -0
- package/fixtures/release-quality/rsbuild-react/corpus.json +263 -0
- package/fixtures/release-quality/rsbuild-react/input.json +62 -0
- package/fixtures/retrieval-quality-corpus.json +7 -7
- package/fixtures/semantic-regex-inventory.json +3 -10
- package/fixtures/unified-runtime/baselines/2026-08-08.json +37 -0
- package/fixtures/unified-runtime/context.json +206 -0
- package/fixtures/unified-runtime/knowledge.json +224 -0
- package/fixtures/unified-runtime/memory.json +224 -0
- package/fixtures/unified-runtime/real-projects/rsbuild-react-2026-08-07/context.json +32 -0
- package/fixtures/unified-runtime/real-projects/rsbuild-react-2026-08-07/knowledge.json +32 -0
- package/fixtures/unified-runtime/real-projects/rsbuild-react-2026-08-07/memory.json +32 -0
- package/fixtures/unified-runtime/real-projects/rsbuild-react-2026-08-07/semantic.json +409 -0
- package/fixtures/unified-runtime/runs/rsbuild-react-2026-08-05.json +54 -0
- package/fixtures/unified-runtime/semantic.json +409 -0
- package/package.json +24 -6
|
@@ -0,0 +1,55 @@
|
|
|
1
|
+
import { type EvaluationBundleAssetV1, type EvaluationBundleManifestV1, type EvaluationChannel, type EmbeddingProfileV1, type ProjectSnapshotIdentityV1 } from '@devflow-tools/sdk';
|
|
2
|
+
export interface EvaluationLabelCaseV1 {
|
|
3
|
+
caseId: string;
|
|
4
|
+
goldSets: string[][];
|
|
5
|
+
forbiddenRefs: string[];
|
|
6
|
+
reviewer: string;
|
|
7
|
+
labelledAt: string;
|
|
8
|
+
sourceEvidence: string[];
|
|
9
|
+
}
|
|
10
|
+
export interface EvaluationLabelSetV1 {
|
|
11
|
+
schemaVersion: 'devflow.evaluation-labels.v1';
|
|
12
|
+
projectSnapshotId: string;
|
|
13
|
+
channel: EvaluationChannel;
|
|
14
|
+
channelSnapshotHash: string;
|
|
15
|
+
datasetHash: string;
|
|
16
|
+
labels: EvaluationLabelCaseV1[];
|
|
17
|
+
}
|
|
18
|
+
export interface EvaluationCorpusV1 {
|
|
19
|
+
schemaVersion: 'devflow.evaluation-corpus.v1';
|
|
20
|
+
cases: Array<{
|
|
21
|
+
caseId: string;
|
|
22
|
+
channel: EvaluationChannel;
|
|
23
|
+
[key: string]: unknown;
|
|
24
|
+
}>;
|
|
25
|
+
}
|
|
26
|
+
export interface EvaluationProfilesV1 {
|
|
27
|
+
context: string;
|
|
28
|
+
memory: string;
|
|
29
|
+
knowledge: string;
|
|
30
|
+
embeddingProfiles: {
|
|
31
|
+
code_retrieval: EmbeddingProfileV1;
|
|
32
|
+
};
|
|
33
|
+
}
|
|
34
|
+
export type EvaluationBundleValidation = {
|
|
35
|
+
status: 'valid';
|
|
36
|
+
manifest: EvaluationBundleManifestV1;
|
|
37
|
+
} | {
|
|
38
|
+
status: 'invalid';
|
|
39
|
+
reasonCode: 'manifest_missing' | 'manifest_invalid' | 'asset_hash_mismatch' | 'bundle_identity_mismatch' | 'channel_snapshot_mismatch';
|
|
40
|
+
asset?: string;
|
|
41
|
+
};
|
|
42
|
+
export declare function readProjectSnapshotIdentity(projectRoot: string): ProjectSnapshotIdentityV1;
|
|
43
|
+
export declare function assertCleanGitProject(projectRoot: string): void;
|
|
44
|
+
export declare function logicalDatabaseHash(databasePath: string): string;
|
|
45
|
+
export declare function collectBundleAssets(bundleRoot: string): EvaluationBundleAssetV1[];
|
|
46
|
+
export declare function extractEvaluationStableRefs(input: {
|
|
47
|
+
channel: EvaluationChannel;
|
|
48
|
+
databasePath: string;
|
|
49
|
+
projectRoot: string;
|
|
50
|
+
projectSnapshotId: string;
|
|
51
|
+
channelSnapshotHash: string;
|
|
52
|
+
}): string[];
|
|
53
|
+
export declare function validateEvaluationBundle(bundleRoot: string): EvaluationBundleValidation;
|
|
54
|
+
export declare function vectorAssetIdentity(bundleRoot: string): EvaluationBundleAssetV1[];
|
|
55
|
+
//# sourceMappingURL=evaluation-bundle.d.ts.map
|
|
@@ -0,0 +1 @@
|
|
|
1
|
+
{"version":3,"file":"evaluation-bundle.d.ts","sourceRoot":"","sources":["../src/evaluation-bundle.ts"],"names":[],"mappings":"AAMA,OAAO,EAYL,KAAK,uBAAuB,EAC5B,KAAK,0BAA0B,EAC/B,KAAK,iBAAiB,EACtB,KAAK,kBAAkB,EACvB,KAAK,yBAAyB,EAC/B,MAAM,oBAAoB,CAAC;AAE5B,MAAM,WAAW,qBAAqB;IACpC,MAAM,EAAE,MAAM,CAAC;IACf,QAAQ,EAAE,MAAM,EAAE,EAAE,CAAC;IACrB,aAAa,EAAE,MAAM,EAAE,CAAC;IACxB,QAAQ,EAAE,MAAM,CAAC;IACjB,UAAU,EAAE,MAAM,CAAC;IACnB,cAAc,EAAE,MAAM,EAAE,CAAC;CAC1B;AAED,MAAM,WAAW,oBAAoB;IACnC,aAAa,EAAE,8BAA8B,CAAC;IAC9C,iBAAiB,EAAE,MAAM,CAAC;IAC1B,OAAO,EAAE,iBAAiB,CAAC;IAC3B,mBAAmB,EAAE,MAAM,CAAC;IAC5B,WAAW,EAAE,MAAM,CAAC;IACpB,MAAM,EAAE,qBAAqB,EAAE,CAAC;CACjC;AAED,MAAM,WAAW,kBAAkB;IACjC,aAAa,EAAE,8BAA8B,CAAC;IAC9C,KAAK,EAAE,KAAK,CAAC;QAAE,MAAM,EAAE,MAAM,CAAC;QAAC,OAAO,EAAE,iBAAiB,CAAC;QAAC,CAAC,GAAG,EAAE,MAAM,GAAG,OAAO,CAAA;KAAE,CAAC,CAAC;CACtF;AAED,MAAM,WAAW,oBAAoB;IACnC,OAAO,EAAE,MAAM,CAAC;IAChB,MAAM,EAAE,MAAM,CAAC;IACf,SAAS,EAAE,MAAM,CAAC;IAClB,iBAAiB,EAAE;QAAE,cAAc,EAAE,kBAAkB,CAAA;KAAE,CAAC;CAC3D;AAED,MAAM,MAAM,0BAA0B,GAAG;IACvC,MAAM,EAAE,OAAO,CAAC;IAChB,QAAQ,EAAE,0BAA0B,CAAC;CACtC,GAAG;IACF,MAAM,EAAE,SAAS,CAAC;IAClB,UAAU,EAAE,kBAAkB,GAAG,kBAAkB,GAAG,qBAAqB,GACvE,0BAA0B,GAAG,2BAA2B,CAAC;IAC7D,KAAK,CAAC,EAAE,MAAM,CAAC;CAChB,CAAC;AAEF,wBAAgB,2BAA2B,CAAC,WAAW,EAAE,MAAM,GAAG,yBAAyB,CAY1F;AAED,wBAAgB,qBAAqB,CAAC,WAAW,EAAE,MAAM,GAAG,IAAI,CAI/D;AAED,wBAAgB,mBAAmB,CAAC,YAAY,EAAE,MAAM,GAAG,MAAM,CAmBhE;AAED,wBAAgB,mBAAmB,CAAC,UAAU,EAAE,MAAM,GAAG,uBAAuB,EAAE,CAYjF;AAED,wBAAgB,2BAA2B,CAAC,KAAK,EAAE;IACjD,OAAO,EAAE,iBAAiB,CAAC;IAC3B,YAAY,EAAE,MAAM,CAAC;IACrB,WAAW,EAAE,MAAM,CAAC;IACpB,iBAAiB,EAAE,MAAM,CAAC;IAC1B,mBAAmB,EAAE,MAAM,CAAC;CAC7B,GAAG,MAAM,EAAE,CAwDX;AAED,wBAAgB,wBAAwB,CAAC,UAAU,EAAE,MAAM,GAAG,0BAA0B,CAqEvF;AA4BD,wBAAgB,mBAAmB,CAAC,UAAU,EAAE,MAAM,GAAG,uBAAuB,EAAE,CAQjF"}
|
|
@@ -0,0 +1,272 @@
|
|
|
1
|
+
import { execFileSync } from 'node:child_process';
|
|
2
|
+
import { createHash } from 'node:crypto';
|
|
3
|
+
import { existsSync, lstatSync, readFileSync, readlinkSync, readdirSync } from 'node:fs';
|
|
4
|
+
import { join, relative, resolve } from 'node:path';
|
|
5
|
+
import Database from 'better-sqlite3';
|
|
6
|
+
import * as sqliteVec from 'sqlite-vec';
|
|
7
|
+
import { EVALUATION_BUNDLE_SCHEMA, assetTreeHash, channelSnapshotHash, createEmbeddingProfile, datasetHash, evaluationBundleId, evaluationContextStableRef, evaluationKnowledgeStableRef, evaluationMemoryStableRef, projectSnapshotId, stableRuntimeJson, } from '@devflow-tools/sdk';
|
|
8
|
+
export function readProjectSnapshotIdentity(projectRoot) {
|
|
9
|
+
const snapshotPath = join(projectRoot, '.devflow/source-snapshot.json');
|
|
10
|
+
if (existsSync(snapshotPath))
|
|
11
|
+
return JSON.parse(readFileSync(snapshotPath, 'utf8'));
|
|
12
|
+
const commit = git(projectRoot, ['rev-parse', 'HEAD']).trim();
|
|
13
|
+
const tree = git(projectRoot, ['rev-parse', 'HEAD^{tree}']).trim();
|
|
14
|
+
const output = execFileSync('git', ['-C', projectRoot, 'ls-tree', '-r', '-z', '--full-tree', 'HEAD']);
|
|
15
|
+
const trackedFiles = output.toString('utf8').split('\0').filter(Boolean).map(entry => {
|
|
16
|
+
const match = /^(\d+)\s+(?:blob|commit)\s+([a-f0-9]+)\t(.+)$/u.exec(entry);
|
|
17
|
+
if (!match)
|
|
18
|
+
throw new Error(`EVALUATION_GIT_TREE_ENTRY_INVALID:${entry}`);
|
|
19
|
+
return { mode: match[1], blob: match[2], path: match[3] };
|
|
20
|
+
});
|
|
21
|
+
return { commit, tree, trackedFiles };
|
|
22
|
+
}
|
|
23
|
+
export function assertCleanGitProject(projectRoot) {
|
|
24
|
+
if (existsSync(join(projectRoot, '.devflow/source-snapshot.json')))
|
|
25
|
+
return;
|
|
26
|
+
const status = git(projectRoot, ['status', '--porcelain', '--untracked-files=all']).trim();
|
|
27
|
+
if (status)
|
|
28
|
+
throw new Error(`EVALUATION_PROJECT_NOT_CLEAN\n${status}`);
|
|
29
|
+
}
|
|
30
|
+
export function logicalDatabaseHash(databasePath) {
|
|
31
|
+
const database = new Database(databasePath, { readonly: true, fileMustExist: true });
|
|
32
|
+
try {
|
|
33
|
+
sqliteVec.load(database);
|
|
34
|
+
const tables = database.prepare(`SELECT name, sql FROM sqlite_master
|
|
35
|
+
WHERE type = 'table' AND name NOT LIKE 'sqlite_%' ORDER BY name`).all();
|
|
36
|
+
const body = tables.map(table => {
|
|
37
|
+
const rows = database.prepare(`SELECT * FROM ${quoteIdentifier(table.name)}`).all()
|
|
38
|
+
.map(row => typedSqliteValue(row))
|
|
39
|
+
.sort((left, right) => stableRuntimeJson(left).localeCompare(stableRuntimeJson(right)));
|
|
40
|
+
return { name: table.name, sql: table.sql ?? '', rows };
|
|
41
|
+
});
|
|
42
|
+
return channelSnapshotHash({ userVersion: database.pragma('user_version', { simple: true }), tables: body });
|
|
43
|
+
}
|
|
44
|
+
finally {
|
|
45
|
+
database.close();
|
|
46
|
+
}
|
|
47
|
+
}
|
|
48
|
+
export function collectBundleAssets(bundleRoot) {
|
|
49
|
+
return walkAssets(bundleRoot).filter(path => path !== 'manifest.json').map((path) => {
|
|
50
|
+
const absolute = join(bundleRoot, path);
|
|
51
|
+
const stat = lstatSync(absolute);
|
|
52
|
+
const content = stat.isSymbolicLink() ? Buffer.from(readlinkSync(absolute)) : readFileSync(absolute);
|
|
53
|
+
return {
|
|
54
|
+
path,
|
|
55
|
+
kind: stat.isSymbolicLink() ? 'symlink' : 'file',
|
|
56
|
+
sha256: createHash('sha256').update(content).digest('hex'),
|
|
57
|
+
size: content.length,
|
|
58
|
+
};
|
|
59
|
+
}).sort((left, right) => left.path.localeCompare(right.path));
|
|
60
|
+
}
|
|
61
|
+
export function extractEvaluationStableRefs(input) {
|
|
62
|
+
const database = new Database(input.databasePath, { readonly: true, fileMustExist: true });
|
|
63
|
+
try {
|
|
64
|
+
if (input.channel === 'context') {
|
|
65
|
+
requireColumns(database, 'nodes', ['id', 'file_path', 'start_line', 'end_line']);
|
|
66
|
+
const rows = database.prepare(`SELECT id, file_path, start_line, end_line FROM nodes ORDER BY id`).all();
|
|
67
|
+
return unique(rows.map(row => evaluationContextStableRef({
|
|
68
|
+
projectSnapshotId: input.projectSnapshotId,
|
|
69
|
+
normalizedPath: projectRelativePath(input.projectRoot, row.file_path),
|
|
70
|
+
symbolId: row.id,
|
|
71
|
+
range: [row.start_line ?? 0, row.end_line ?? 0],
|
|
72
|
+
})));
|
|
73
|
+
}
|
|
74
|
+
if (input.channel === 'memory') {
|
|
75
|
+
requireColumns(database, 'memories', ['id', 'version']);
|
|
76
|
+
const facts = database.prepare('SELECT id, version FROM memories ORDER BY id').all().map(row => evaluationMemoryStableRef({
|
|
77
|
+
channelSnapshotHash: input.channelSnapshotHash,
|
|
78
|
+
kind: 'fact', immutableId: row.id, version: row.version,
|
|
79
|
+
}));
|
|
80
|
+
const episodes = tableExists(database, 'memory_episodes_v1')
|
|
81
|
+
? database.prepare(`SELECT episode_id, episode_hash FROM memory_episodes_v1
|
|
82
|
+
ORDER BY episode_id`).all().map(row => evaluationMemoryStableRef({
|
|
83
|
+
channelSnapshotHash: input.channelSnapshotHash,
|
|
84
|
+
kind: 'episode', immutableId: row.episode_id, version: row.episode_hash,
|
|
85
|
+
}))
|
|
86
|
+
: [];
|
|
87
|
+
return unique([...facts, ...episodes]);
|
|
88
|
+
}
|
|
89
|
+
requireColumns(database, 'knowledge_docs', ['id', 'source', 'version', 'status']);
|
|
90
|
+
requireColumns(database, 'knowledge_chunks', ['id', 'doc_id', 'content_hash']);
|
|
91
|
+
const rows = database.prepare(`SELECT c.id, c.content_hash, d.source, d.version
|
|
92
|
+
FROM knowledge_chunks c JOIN knowledge_docs d ON d.id = c.doc_id
|
|
93
|
+
WHERE d.status = 'active' ORDER BY c.id`).all();
|
|
94
|
+
return unique(rows.map(row => evaluationKnowledgeStableRef({
|
|
95
|
+
sourceId: row.source,
|
|
96
|
+
sourceVersion: required(row.version, 'EVALUATION_KNOWLEDGE_VERSION_REQUIRED'),
|
|
97
|
+
contentHash: row.content_hash,
|
|
98
|
+
chunkId: row.id,
|
|
99
|
+
})));
|
|
100
|
+
}
|
|
101
|
+
finally {
|
|
102
|
+
database.close();
|
|
103
|
+
}
|
|
104
|
+
}
|
|
105
|
+
export function validateEvaluationBundle(bundleRoot) {
|
|
106
|
+
const manifestPath = join(bundleRoot, 'manifest.json');
|
|
107
|
+
if (!existsSync(manifestPath))
|
|
108
|
+
return { status: 'invalid', reasonCode: 'manifest_missing' };
|
|
109
|
+
let manifest;
|
|
110
|
+
let project;
|
|
111
|
+
let profiles;
|
|
112
|
+
try {
|
|
113
|
+
manifest = JSON.parse(readFileSync(manifestPath, 'utf8'));
|
|
114
|
+
project = JSON.parse(readFileSync(join(bundleRoot, 'project.json'), 'utf8'));
|
|
115
|
+
profiles = JSON.parse(readFileSync(join(bundleRoot, 'profiles.json'), 'utf8'));
|
|
116
|
+
if (manifest.schemaVersion !== EVALUATION_BUNDLE_SCHEMA
|
|
117
|
+
|| typeof manifest.projectId !== 'string' || !manifest.projectId.trim()
|
|
118
|
+
|| manifest.projectSnapshotId !== projectSnapshotId(project)
|
|
119
|
+
|| !isEvaluationProfiles(profiles)
|
|
120
|
+
|| datasetHash(profiles) !== manifest.profilesHash) {
|
|
121
|
+
return { status: 'invalid', reasonCode: 'manifest_invalid' };
|
|
122
|
+
}
|
|
123
|
+
}
|
|
124
|
+
catch {
|
|
125
|
+
return { status: 'invalid', reasonCode: 'manifest_invalid' };
|
|
126
|
+
}
|
|
127
|
+
const actualAssets = collectBundleAssets(bundleRoot);
|
|
128
|
+
const expected = new Map(manifest.assets.map(asset => [asset.path, asset]));
|
|
129
|
+
for (const asset of actualAssets) {
|
|
130
|
+
const recorded = expected.get(asset.path);
|
|
131
|
+
if (!recorded || recorded.sha256 !== asset.sha256 || recorded.size !== asset.size) {
|
|
132
|
+
return { status: 'invalid', reasonCode: 'asset_hash_mismatch', asset: asset.path };
|
|
133
|
+
}
|
|
134
|
+
expected.delete(asset.path);
|
|
135
|
+
}
|
|
136
|
+
if (expected.size > 0) {
|
|
137
|
+
return { status: 'invalid', reasonCode: 'asset_hash_mismatch', asset: [...expected.keys()].sort()[0] };
|
|
138
|
+
}
|
|
139
|
+
const channelPaths = {
|
|
140
|
+
context: join(bundleRoot, 'channels/codegraph.db'),
|
|
141
|
+
contextVectors: join(bundleRoot, 'channels/context-symbols.db'),
|
|
142
|
+
memory: join(bundleRoot, 'channels/memory.db'),
|
|
143
|
+
knowledge: join(bundleRoot, 'channels/knowledge.db'),
|
|
144
|
+
};
|
|
145
|
+
for (const channel of ['context', 'memory', 'knowledge']) {
|
|
146
|
+
const databaseHash = logicalDatabaseHash(channelPaths[channel]);
|
|
147
|
+
const logicalHash = channel === 'context'
|
|
148
|
+
? channelSnapshotHash({
|
|
149
|
+
databaseHash,
|
|
150
|
+
vectorDatabaseHash: logicalDatabaseHash(channelPaths.contextVectors),
|
|
151
|
+
})
|
|
152
|
+
: channel === 'knowledge'
|
|
153
|
+
? channelSnapshotHash({ databaseHash, vectorAssets: vectorAssetIdentity(bundleRoot) })
|
|
154
|
+
: databaseHash;
|
|
155
|
+
if (logicalHash !== manifest.channelSnapshotHashes[channel]) {
|
|
156
|
+
return { status: 'invalid', reasonCode: 'channel_snapshot_mismatch', asset: channel };
|
|
157
|
+
}
|
|
158
|
+
}
|
|
159
|
+
const bundleId = evaluationBundleId({
|
|
160
|
+
schemaVersion: manifest.schemaVersion,
|
|
161
|
+
projectId: manifest.projectId,
|
|
162
|
+
project,
|
|
163
|
+
channels: manifest.channelSnapshotHashes,
|
|
164
|
+
corpusHash: manifest.corpusHash,
|
|
165
|
+
labelHashes: manifest.labelHashes,
|
|
166
|
+
assetTreeHash: assetTreeHash(manifest.assets),
|
|
167
|
+
builderIdentity: manifest.builderIdentity,
|
|
168
|
+
profilesHash: manifest.profilesHash,
|
|
169
|
+
});
|
|
170
|
+
return bundleId === manifest.bundleId
|
|
171
|
+
? { status: 'valid', manifest }
|
|
172
|
+
: { status: 'invalid', reasonCode: 'bundle_identity_mismatch' };
|
|
173
|
+
}
|
|
174
|
+
function isEvaluationProfiles(value) {
|
|
175
|
+
if (!value || typeof value !== 'object'
|
|
176
|
+
|| stableRuntimeJson(Object.keys(value).sort()) !== stableRuntimeJson([
|
|
177
|
+
'context', 'embeddingProfiles', 'knowledge', 'memory',
|
|
178
|
+
])
|
|
179
|
+
|| typeof value.context !== 'string' || !value.context.trim()
|
|
180
|
+
|| typeof value.memory !== 'string' || !value.memory.trim()
|
|
181
|
+
|| typeof value.knowledge !== 'string' || !value.knowledge.trim()
|
|
182
|
+
|| !value.embeddingProfiles || typeof value.embeddingProfiles !== 'object'
|
|
183
|
+
|| stableRuntimeJson(Object.keys(value.embeddingProfiles).sort()) !== stableRuntimeJson(['code_retrieval'])) {
|
|
184
|
+
return false;
|
|
185
|
+
}
|
|
186
|
+
const profile = value.embeddingProfiles.code_retrieval;
|
|
187
|
+
if (!profile || profile.purpose !== 'code_retrieval'
|
|
188
|
+
|| stableRuntimeJson(Object.keys(profile).sort()) !== stableRuntimeJson([
|
|
189
|
+
'dimension', 'generation', 'modelId', 'modelRevision', 'normalization',
|
|
190
|
+
'profileHash', 'provider', 'purpose', 'tokenizerConfigHash',
|
|
191
|
+
]))
|
|
192
|
+
return false;
|
|
193
|
+
const { profileHash, ...body } = profile;
|
|
194
|
+
try {
|
|
195
|
+
return createEmbeddingProfile(body).profileHash === profileHash;
|
|
196
|
+
}
|
|
197
|
+
catch {
|
|
198
|
+
return false;
|
|
199
|
+
}
|
|
200
|
+
}
|
|
201
|
+
export function vectorAssetIdentity(bundleRoot) {
|
|
202
|
+
const vectorRoot = join(bundleRoot, 'channels/knowledge-assets');
|
|
203
|
+
if (!existsSync(vectorRoot))
|
|
204
|
+
return [];
|
|
205
|
+
return walkAssets(vectorRoot).map((path) => {
|
|
206
|
+
const absolute = join(vectorRoot, path);
|
|
207
|
+
const content = readFileSync(absolute);
|
|
208
|
+
return { path, kind: 'file', sha256: createHash('sha256').update(content).digest('hex'), size: content.length };
|
|
209
|
+
}).sort((left, right) => left.path.localeCompare(right.path));
|
|
210
|
+
}
|
|
211
|
+
function typedSqliteValue(value) {
|
|
212
|
+
if (Buffer.isBuffer(value))
|
|
213
|
+
return { type: 'blob', hex: value.toString('hex') };
|
|
214
|
+
if (Array.isArray(value))
|
|
215
|
+
return value.map(typedSqliteValue);
|
|
216
|
+
if (value && typeof value === 'object') {
|
|
217
|
+
return Object.fromEntries(Object.entries(value).sort(([left], [right]) => left.localeCompare(right))
|
|
218
|
+
.map(([key, child]) => [key, typedSqliteValue(child)]));
|
|
219
|
+
}
|
|
220
|
+
if (typeof value === 'bigint')
|
|
221
|
+
return { type: 'integer', value: value.toString() };
|
|
222
|
+
return value;
|
|
223
|
+
}
|
|
224
|
+
function walkAssets(root, prefix = '') {
|
|
225
|
+
if (!existsSync(root))
|
|
226
|
+
return [];
|
|
227
|
+
return readdirSync(join(root, prefix), { withFileTypes: true }).flatMap(entry => {
|
|
228
|
+
const path = prefix ? `${prefix}/${entry.name}` : entry.name;
|
|
229
|
+
if (entry.isDirectory())
|
|
230
|
+
return walkAssets(root, path);
|
|
231
|
+
if (entry.isFile() || entry.isSymbolicLink())
|
|
232
|
+
return [path];
|
|
233
|
+
throw new Error(`EVALUATION_ASSET_TYPE_UNSUPPORTED:${path}`);
|
|
234
|
+
}).sort();
|
|
235
|
+
}
|
|
236
|
+
function git(root, args) {
|
|
237
|
+
return execFileSync('git', ['-C', root, ...args], { encoding: 'utf8' });
|
|
238
|
+
}
|
|
239
|
+
function quoteIdentifier(value) {
|
|
240
|
+
return `"${value.replaceAll('"', '""')}"`;
|
|
241
|
+
}
|
|
242
|
+
function tableExists(database, table) {
|
|
243
|
+
return Boolean(database.prepare(`SELECT 1 FROM sqlite_master WHERE type = 'table' AND name = ?`).get(table));
|
|
244
|
+
}
|
|
245
|
+
function requireColumns(database, table, columns) {
|
|
246
|
+
if (!tableExists(database, table))
|
|
247
|
+
throw new Error(`EVALUATION_CHANNEL_TABLE_REQUIRED:${table}`);
|
|
248
|
+
const actual = new Set(database.pragma(`table_info(${quoteIdentifier(table)})`)
|
|
249
|
+
.map(column => column.name));
|
|
250
|
+
for (const column of columns) {
|
|
251
|
+
if (!actual.has(column))
|
|
252
|
+
throw new Error(`EVALUATION_CHANNEL_COLUMN_REQUIRED:${table}.${column}`);
|
|
253
|
+
}
|
|
254
|
+
}
|
|
255
|
+
function projectRelativePath(projectRoot, value) {
|
|
256
|
+
const normalized = value.replaceAll('\\', '/');
|
|
257
|
+
if (!normalized.startsWith('/') && !/^[A-Za-z]:\//u.test(normalized))
|
|
258
|
+
return normalized;
|
|
259
|
+
const path = relative(resolve(projectRoot), resolve(value)).replaceAll('\\', '/');
|
|
260
|
+
if (!path || path === '..' || path.startsWith('../'))
|
|
261
|
+
throw new Error('EVALUATION_CHANNEL_PATH_OUTSIDE_SOURCE');
|
|
262
|
+
return path;
|
|
263
|
+
}
|
|
264
|
+
function unique(values) {
|
|
265
|
+
return [...new Set(values)].sort();
|
|
266
|
+
}
|
|
267
|
+
function required(value, error) {
|
|
268
|
+
if (!value?.trim())
|
|
269
|
+
throw new Error(error);
|
|
270
|
+
return value;
|
|
271
|
+
}
|
|
272
|
+
//# sourceMappingURL=evaluation-bundle.js.map
|
|
@@ -0,0 +1 @@
|
|
|
1
|
+
{"version":3,"file":"evaluation-bundle.js","sourceRoot":"","sources":["../src/evaluation-bundle.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,YAAY,EAAE,MAAM,oBAAoB,CAAC;AAClD,OAAO,EAAE,UAAU,EAAE,MAAM,aAAa,CAAC;AACzC,OAAO,EAAE,UAAU,EAAE,SAAS,EAAE,YAAY,EAAE,YAAY,EAAE,WAAW,EAAE,MAAM,SAAS,CAAC;AACzF,OAAO,EAAE,IAAI,EAAE,QAAQ,EAAE,OAAO,EAAE,MAAM,WAAW,CAAC;AACpD,OAAO,QAAQ,MAAM,gBAAgB,CAAC;AACtC,OAAO,KAAK,SAAS,MAAM,YAAY,CAAC;AACxC,OAAO,EACL,wBAAwB,EACxB,aAAa,EACb,mBAAmB,EACnB,sBAAsB,EACtB,WAAW,EACX,kBAAkB,EAClB,0BAA0B,EAC1B,4BAA4B,EAC5B,yBAAyB,EACzB,iBAAiB,EACjB,iBAAiB,GAMlB,MAAM,oBAAoB,CAAC;AA0C5B,MAAM,UAAU,2BAA2B,CAAC,WAAmB;IAC7D,MAAM,YAAY,GAAG,IAAI,CAAC,WAAW,EAAE,+BAA+B,CAAC,CAAC;IACxE,IAAI,UAAU,CAAC,YAAY,CAAC;QAAE,OAAO,IAAI,CAAC,KAAK,CAAC,YAAY,CAAC,YAAY,EAAE,MAAM,CAAC,CAA8B,CAAC;IACjH,MAAM,MAAM,GAAG,GAAG,CAAC,WAAW,EAAE,CAAC,WAAW,EAAE,MAAM,CAAC,CAAC,CAAC,IAAI,EAAE,CAAC;IAC9D,MAAM,IAAI,GAAG,GAAG,CAAC,WAAW,EAAE,CAAC,WAAW,EAAE,aAAa,CAAC,CAAC,CAAC,IAAI,EAAE,CAAC;IACnE,MAAM,MAAM,GAAG,YAAY,CAAC,KAAK,EAAE,CAAC,IAAI,EAAE,WAAW,EAAE,SAAS,EAAE,IAAI,EAAE,IAAI,EAAE,aAAa,EAAE,MAAM,CAAC,CAAC,CAAC;IACtG,MAAM,YAAY,GAAG,MAAM,CAAC,QAAQ,CAAC,MAAM,CAAC,CAAC,KAAK,CAAC,IAAI,CAAC,CAAC,MAAM,CAAC,OAAO,CAAC,CAAC,GAAG,CAAC,KAAK,CAAC,EAAE;QACnF,MAAM,KAAK,GAAG,gDAAgD,CAAC,IAAI,CAAC,KAAK,CAAC,CAAC;QAC3E,IAAI,CAAC,KAAK;YAAE,MAAM,IAAI,KAAK,CAAC,qCAAqC,KAAK,EAAE,CAAC,CAAC;QAC1E,OAAO,EAAE,IAAI,EAAE,KAAK,CAAC,CAAC,CAAC,EAAE,IAAI,EAAE,KAAK,CAAC,CAAC,CAAC,EAAE,IAAI,EAAE,KAAK,CAAC,CAAC,CAAC,EAAE,CAAC;IAC5D,CAAC,CAAC,CAAC;IACH,OAAO,EAAE,MAAM,EAAE,IAAI,EAAE,YAAY,EAAE,CAAC;AACxC,CAAC;AAED,MAAM,UAAU,qBAAqB,CAAC,WAAmB;IACvD,IAAI,UAAU,CAAC,IAAI,CAAC,WAAW,EAAE,+BAA+B,CAAC,CAAC;QAAE,OAAO;IAC3E,MAAM,MAAM,GAAG,GAAG,CAAC,WAAW,EAAE,CAAC,QAAQ,EAAE,aAAa,EAAE,uBAAuB,CAAC,CAAC,CAAC,IAAI,EAAE,CAAC;IAC3F,IAAI,MAAM;QAAE,MAAM,IAAI,KAAK,CAAC,iCAAiC,MAAM,EAAE,CAAC,CAAC;AACzE,CAAC;AAED,MAAM,UAAU,mBAAmB,CAAC,YAAoB;IACtD,MAAM,QAAQ,GAAG,IAAI,QAAQ,CAAC,YAAY,EAAE,EAAE,QAAQ,EAAE,IAAI,EAAE,aAAa,EAAE,IAAI,EAAE,CAAC,CAAC;IACrF,IAAI,CAAC;QACH,SAAS,CAAC,IAAI,CAAC,QAAQ,CAAC,CAAC;QACzB,MAAM,MAAM,GAAG,QAAQ,CAAC,OAAO,CAAC;sEACkC,CAAC,CAAC,GAAG,EAGnE,CAAC;QACL,MAAM,IAAI,GAAG,MAAM,CAAC,GAAG,CAAC,KAAK,CAAC,EAAE;YAC9B,MAAM,IAAI,GAAG,QAAQ,CAAC,OAAO,CAAC,iBAAiB,eAAe,CAAC,KAAK,CAAC,IAAI,CAAC,EAAE,CAAC,CAAC,GAAG,EAAE;iBAChF,GAAG,CAAC,GAAG,CAAC,EAAE,CAAC,gBAAgB,CAAC,GAAG,CAAC,CAAC;iBACjC,IAAI,CAAC,CAAC,IAAI,EAAE,KAAK,EAAE,EAAE,CAAC,iBAAiB,CAAC,IAAI,CAAC,CAAC,aAAa,CAAC,iBAAiB,CAAC,KAAK,CAAC,CAAC,CAAC,CAAC;YAC1F,OAAO,EAAE,IAAI,EAAE,KAAK,CAAC,IAAI,EAAE,GAAG,EAAE,KAAK,CAAC,GAAG,IAAI,EAAE,EAAE,IAAI,EAAE,CAAC;QAC1D,CAAC,CAAC,CAAC;QACH,OAAO,mBAAmB,CAAC,EAAE,WAAW,EAAE,QAAQ,CAAC,MAAM,CAAC,cAAc,EAAE,EAAE,MAAM,EAAE,IAAI,EAAE,CAAC,EAAE,MAAM,EAAE,IAAI,EAAE,CAAC,CAAC;IAC/G,CAAC;YAAS,CAAC;QACT,QAAQ,CAAC,KAAK,EAAE,CAAC;IACnB,CAAC;AACH,CAAC;AAED,MAAM,UAAU,mBAAmB,CAAC,UAAkB;IACpD,OAAO,UAAU,CAAC,UAAU,CAAC,CAAC,MAAM,CAAC,IAAI,CAAC,EAAE,CAAC,IAAI,KAAK,eAAe,CAAC,CAAC,GAAG,CAAC,CAAC,IAAI,EAA2B,EAAE;QAC3G,MAAM,QAAQ,GAAG,IAAI,CAAC,UAAU,EAAE,IAAI,CAAC,CAAC;QACxC,MAAM,IAAI,GAAG,SAAS,CAAC,QAAQ,CAAC,CAAC;QACjC,MAAM,OAAO,GAAG,IAAI,CAAC,cAAc,EAAE,CAAC,CAAC,CAAC,MAAM,CAAC,IAAI,CAAC,YAAY,CAAC,QAAQ,CAAC,CAAC,CAAC,CAAC,CAAC,YAAY,CAAC,QAAQ,CAAC,CAAC;QACrG,OAAO;YACL,IAAI;YACJ,IAAI,EAAE,IAAI,CAAC,cAAc,EAAE,CAAC,CAAC,CAAC,SAAS,CAAC,CAAC,CAAC,MAAM;YAChD,MAAM,EAAE,UAAU,CAAC,QAAQ,CAAC,CAAC,MAAM,CAAC,OAAO,CAAC,CAAC,MAAM,CAAC,KAAK,CAAC;YAC1D,IAAI,EAAE,OAAO,CAAC,MAAM;SACrB,CAAC;IACJ,CAAC,CAAC,CAAC,IAAI,CAAC,CAAC,IAAI,EAAE,KAAK,EAAE,EAAE,CAAC,IAAI,CAAC,IAAI,CAAC,aAAa,CAAC,KAAK,CAAC,IAAI,CAAC,CAAC,CAAC;AAChE,CAAC;AAED,MAAM,UAAU,2BAA2B,CAAC,KAM3C;IACC,MAAM,QAAQ,GAAG,IAAI,QAAQ,CAAC,KAAK,CAAC,YAAY,EAAE,EAAE,QAAQ,EAAE,IAAI,EAAE,aAAa,EAAE,IAAI,EAAE,CAAC,CAAC;IAC3F,IAAI,CAAC;QACH,IAAI,KAAK,CAAC,OAAO,KAAK,SAAS,EAAE,CAAC;YAChC,cAAc,CAAC,QAAQ,EAAE,OAAO,EAAE,CAAC,IAAI,EAAE,WAAW,EAAE,YAAY,EAAE,UAAU,CAAC,CAAC,CAAC;YACjF,MAAM,IAAI,GAAG,QAAQ,CAAC,OAAO,CAAC,mEAAmE,CAAC,CAAC,GAAG,EAKpG,CAAC;YACH,OAAO,MAAM,CAAC,IAAI,CAAC,GAAG,CAAC,GAAG,CAAC,EAAE,CAAC,0BAA0B,CAAC;gBACvD,iBAAiB,EAAE,KAAK,CAAC,iBAAiB;gBAC1C,cAAc,EAAE,mBAAmB,CAAC,KAAK,CAAC,WAAW,EAAE,GAAG,CAAC,SAAS,CAAC;gBACrE,QAAQ,EAAE,GAAG,CAAC,EAAE;gBAChB,KAAK,EAAE,CAAC,GAAG,CAAC,UAAU,IAAI,CAAC,EAAE,GAAG,CAAC,QAAQ,IAAI,CAAC,CAAC;aAChD,CAAC,CAAC,CAAC,CAAC;QACP,CAAC;QACD,IAAI,KAAK,CAAC,OAAO,KAAK,QAAQ,EAAE,CAAC;YAC/B,cAAc,CAAC,QAAQ,EAAE,UAAU,EAAE,CAAC,IAAI,EAAE,SAAS,CAAC,CAAC,CAAC;YACxD,MAAM,KAAK,GAAI,QAAQ,CAAC,OAAO,CAAC,8CAA8C,CAAC,CAAC,GAAG,EAGhF,CAAC,GAAG,CAAC,GAAG,CAAC,EAAE,CAAC,yBAAyB,CAAC;gBACvC,mBAAmB,EAAE,KAAK,CAAC,mBAAmB;gBAC9C,IAAI,EAAE,MAAM,EAAE,WAAW,EAAE,GAAG,CAAC,EAAE,EAAE,OAAO,EAAE,GAAG,CAAC,OAAO;aACxD,CAAC,CAAC,CAAC;YACJ,MAAM,QAAQ,GAAG,WAAW,CAAC,QAAQ,EAAE,oBAAoB,CAAC;gBAC1D,CAAC,CAAE,QAAQ,CAAC,OAAO,CAAC;gCACI,CAAC,CAAC,GAAG,EAA0D,CAAC,GAAG,CAAC,GAAG,CAAC,EAAE,CAC9F,yBAAyB,CAAC;oBACxB,mBAAmB,EAAE,KAAK,CAAC,mBAAmB;oBAC9C,IAAI,EAAE,SAAS,EAAE,WAAW,EAAE,GAAG,CAAC,UAAU,EAAE,OAAO,EAAE,GAAG,CAAC,YAAY;iBACxE,CAAC,CAAC;gBACP,CAAC,CAAC,EAAE,CAAC;YACP,OAAO,MAAM,CAAC,CAAC,GAAG,KAAK,EAAE,GAAG,QAAQ,CAAC,CAAC,CAAC;QACzC,CAAC;QACD,cAAc,CAAC,QAAQ,EAAE,gBAAgB,EAAE,CAAC,IAAI,EAAE,QAAQ,EAAE,SAAS,EAAE,QAAQ,CAAC,CAAC,CAAC;QAClF,cAAc,CAAC,QAAQ,EAAE,kBAAkB,EAAE,CAAC,IAAI,EAAE,QAAQ,EAAE,cAAc,CAAC,CAAC,CAAC;QAC/E,MAAM,IAAI,GAAG,QAAQ,CAAC,OAAO,CAAC;;8CAEY,CAAC,CAAC,GAAG,EAK3C,CAAC;QACL,OAAO,MAAM,CAAC,IAAI,CAAC,GAAG,CAAC,GAAG,CAAC,EAAE,CAAC,4BAA4B,CAAC;YACzD,QAAQ,EAAE,GAAG,CAAC,MAAM;YACpB,aAAa,EAAE,QAAQ,CAAC,GAAG,CAAC,OAAO,EAAE,uCAAuC,CAAC;YAC7E,WAAW,EAAE,GAAG,CAAC,YAAY;YAC7B,OAAO,EAAE,GAAG,CAAC,EAAE;SAChB,CAAC,CAAC,CAAC,CAAC;IACP,CAAC;YAAS,CAAC;QACT,QAAQ,CAAC,KAAK,EAAE,CAAC;IACnB,CAAC;AACH,CAAC;AAED,MAAM,UAAU,wBAAwB,CAAC,UAAkB;IACzD,MAAM,YAAY,GAAG,IAAI,CAAC,UAAU,EAAE,eAAe,CAAC,CAAC;IACvD,IAAI,CAAC,UAAU,CAAC,YAAY,CAAC;QAAE,OAAO,EAAE,MAAM,EAAE,SAAS,EAAE,UAAU,EAAE,kBAAkB,EAAE,CAAC;IAC5F,IAAI,QAAoC,CAAC;IACzC,IAAI,OAAkC,CAAC;IACvC,IAAI,QAA8B,CAAC;IACnC,IAAI,CAAC;QACH,QAAQ,GAAG,IAAI,CAAC,KAAK,CAAC,YAAY,CAAC,YAAY,EAAE,MAAM,CAAC,CAA+B,CAAC;QACxF,OAAO,GAAG,IAAI,CAAC,KAAK,CAAC,YAAY,CAAC,IAAI,CAAC,UAAU,EAAE,cAAc,CAAC,EAAE,MAAM,CAAC,CAA8B,CAAC;QAC1G,QAAQ,GAAG,IAAI,CAAC,KAAK,CAAC,YAAY,CAAC,IAAI,CAAC,UAAU,EAAE,eAAe,CAAC,EAAE,MAAM,CAAC,CAAyB,CAAC;QACvG,IAAI,QAAQ,CAAC,aAAa,KAAK,wBAAwB;eAClD,OAAO,QAAQ,CAAC,SAAS,KAAK,QAAQ,IAAI,CAAC,QAAQ,CAAC,SAAS,CAAC,IAAI,EAAE;eACpE,QAAQ,CAAC,iBAAiB,KAAK,iBAAiB,CAAC,OAAO,CAAC;eACzD,CAAC,oBAAoB,CAAC,QAAQ,CAAC;eAC/B,WAAW,CAAC,QAAQ,CAAC,KAAK,QAAQ,CAAC,YAAY,EAAE,CAAC;YACrD,OAAO,EAAE,MAAM,EAAE,SAAS,EAAE,UAAU,EAAE,kBAAkB,EAAE,CAAC;QAC/D,CAAC;IACH,CAAC;IAAC,MAAM,CAAC;QACP,OAAO,EAAE,MAAM,EAAE,SAAS,EAAE,UAAU,EAAE,kBAAkB,EAAE,CAAC;IAC/D,CAAC;IAED,MAAM,YAAY,GAAG,mBAAmB,CAAC,UAAU,CAAC,CAAC;IACrD,MAAM,QAAQ,GAAG,IAAI,GAAG,CAAC,QAAQ,CAAC,MAAM,CAAC,GAAG,CAAC,KAAK,CAAC,EAAE,CAAC,CAAC,KAAK,CAAC,IAAI,EAAE,KAAK,CAAC,CAAC,CAAC,CAAC;IAC5E,KAAK,MAAM,KAAK,IAAI,YAAY,EAAE,CAAC;QACjC,MAAM,QAAQ,GAAG,QAAQ,CAAC,GAAG,CAAC,KAAK,CAAC,IAAI,CAAC,CAAC;QAC1C,IAAI,CAAC,QAAQ,IAAI,QAAQ,CAAC,MAAM,KAAK,KAAK,CAAC,MAAM,IAAI,QAAQ,CAAC,IAAI,KAAK,KAAK,CAAC,IAAI,EAAE,CAAC;YAClF,OAAO,EAAE,MAAM,EAAE,SAAS,EAAE,UAAU,EAAE,qBAAqB,EAAE,KAAK,EAAE,KAAK,CAAC,IAAI,EAAE,CAAC;QACrF,CAAC;QACD,QAAQ,CAAC,MAAM,CAAC,KAAK,CAAC,IAAI,CAAC,CAAC;IAC9B,CAAC;IACD,IAAI,QAAQ,CAAC,IAAI,GAAG,CAAC,EAAE,CAAC;QACtB,OAAO,EAAE,MAAM,EAAE,SAAS,EAAE,UAAU,EAAE,qBAAqB,EAAE,KAAK,EAAE,CAAC,GAAG,QAAQ,CAAC,IAAI,EAAE,CAAC,CAAC,IAAI,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC;IACzG,CAAC;IAED,MAAM,YAAY,GAAG;QACnB,OAAO,EAAE,IAAI,CAAC,UAAU,EAAE,uBAAuB,CAAC;QAClD,cAAc,EAAE,IAAI,CAAC,UAAU,EAAE,6BAA6B,CAAC;QAC/D,MAAM,EAAE,IAAI,CAAC,UAAU,EAAE,oBAAoB,CAAC;QAC9C,SAAS,EAAE,IAAI,CAAC,UAAU,EAAE,uBAAuB,CAAC;KACrD,CAAC;IACF,KAAK,MAAM,OAAO,IAAI,CAAC,SAAS,EAAE,QAAQ,EAAE,WAAW,CAAU,EAAE,CAAC;QAClE,MAAM,YAAY,GAAG,mBAAmB,CAAC,YAAY,CAAC,OAAO,CAAC,CAAC,CAAC;QAChE,MAAM,WAAW,GAAG,OAAO,KAAK,SAAS;YACvC,CAAC,CAAC,mBAAmB,CAAC;gBACpB,YAAY;gBACZ,kBAAkB,EAAE,mBAAmB,CAAC,YAAY,CAAC,cAAc,CAAC;aACrE,CAAC;YACF,CAAC,CAAC,OAAO,KAAK,WAAW;gBACvB,CAAC,CAAC,mBAAmB,CAAC,EAAE,YAAY,EAAE,YAAY,EAAE,mBAAmB,CAAC,UAAU,CAAC,EAAE,CAAC;gBACtF,CAAC,CAAC,YAAY,CAAC;QACnB,IAAI,WAAW,KAAK,QAAQ,CAAC,qBAAqB,CAAC,OAAO,CAAC,EAAE,CAAC;YAC5D,OAAO,EAAE,MAAM,EAAE,SAAS,EAAE,UAAU,EAAE,2BAA2B,EAAE,KAAK,EAAE,OAAO,EAAE,CAAC;QACxF,CAAC;IACH,CAAC;IAED,MAAM,QAAQ,GAAG,kBAAkB,CAAC;QAClC,aAAa,EAAE,QAAQ,CAAC,aAAa;QACrC,SAAS,EAAE,QAAQ,CAAC,SAAS;QAC7B,OAAO;QACP,QAAQ,EAAE,QAAQ,CAAC,qBAAqB;QACxC,UAAU,EAAE,QAAQ,CAAC,UAAU;QAC/B,WAAW,EAAE,QAAQ,CAAC,WAAW;QACjC,aAAa,EAAE,aAAa,CAAC,QAAQ,CAAC,MAAM,CAAC;QAC7C,eAAe,EAAE,QAAQ,CAAC,eAAe;QACzC,YAAY,EAAE,QAAQ,CAAC,YAAY;KACpC,CAAC,CAAC;IACH,OAAO,QAAQ,KAAK,QAAQ,CAAC,QAAQ;QACnC,CAAC,CAAC,EAAE,MAAM,EAAE,OAAO,EAAE,QAAQ,EAAE;QAC/B,CAAC,CAAC,EAAE,MAAM,EAAE,SAAS,EAAE,UAAU,EAAE,0BAA0B,EAAE,CAAC;AACpE,CAAC;AAED,SAAS,oBAAoB,CAAC,KAA2B;IACvD,IAAI,CAAC,KAAK,IAAI,OAAO,KAAK,KAAK,QAAQ;WAClC,iBAAiB,CAAC,MAAM,CAAC,IAAI,CAAC,KAAK,CAAC,CAAC,IAAI,EAAE,CAAC,KAAK,iBAAiB,CAAC;YACpE,SAAS,EAAE,mBAAmB,EAAE,WAAW,EAAE,QAAQ;SACtD,CAAC;WACC,OAAO,KAAK,CAAC,OAAO,KAAK,QAAQ,IAAI,CAAC,KAAK,CAAC,OAAO,CAAC,IAAI,EAAE;WAC1D,OAAO,KAAK,CAAC,MAAM,KAAK,QAAQ,IAAI,CAAC,KAAK,CAAC,MAAM,CAAC,IAAI,EAAE;WACxD,OAAO,KAAK,CAAC,SAAS,KAAK,QAAQ,IAAI,CAAC,KAAK,CAAC,SAAS,CAAC,IAAI,EAAE;WAC9D,CAAC,KAAK,CAAC,iBAAiB,IAAI,OAAO,KAAK,CAAC,iBAAiB,KAAK,QAAQ;WACvE,iBAAiB,CAAC,MAAM,CAAC,IAAI,CAAC,KAAK,CAAC,iBAAiB,CAAC,CAAC,IAAI,EAAE,CAAC,KAAK,iBAAiB,CAAC,CAAC,gBAAgB,CAAC,CAAC,EAAE,CAAC;QAC9G,OAAO,KAAK,CAAC;IACf,CAAC;IACD,MAAM,OAAO,GAAG,KAAK,CAAC,iBAAiB,CAAC,cAAc,CAAC;IACvD,IAAI,CAAC,OAAO,IAAI,OAAO,CAAC,OAAO,KAAK,gBAAgB;WAC/C,iBAAiB,CAAC,MAAM,CAAC,IAAI,CAAC,OAAO,CAAC,CAAC,IAAI,EAAE,CAAC,KAAK,iBAAiB,CAAC;YACtE,WAAW,EAAE,YAAY,EAAE,SAAS,EAAE,eAAe,EAAE,eAAe;YACtE,aAAa,EAAE,UAAU,EAAE,SAAS,EAAE,qBAAqB;SAC5D,CAAC;QAAE,OAAO,KAAK,CAAC;IACnB,MAAM,EAAE,WAAW,EAAE,GAAG,IAAI,EAAE,GAAG,OAAO,CAAC;IACzC,IAAI,CAAC;QACH,OAAO,sBAAsB,CAAC,IAAI,CAAC,CAAC,WAAW,KAAK,WAAW,CAAC;IAClE,CAAC;IAAC,MAAM,CAAC;QACP,OAAO,KAAK,CAAC;IACf,CAAC;AACH,CAAC;AAED,MAAM,UAAU,mBAAmB,CAAC,UAAkB;IACpD,MAAM,UAAU,GAAG,IAAI,CAAC,UAAU,EAAE,2BAA2B,CAAC,CAAC;IACjE,IAAI,CAAC,UAAU,CAAC,UAAU,CAAC;QAAE,OAAO,EAAE,CAAC;IACvC,OAAO,UAAU,CAAC,UAAU,CAAC,CAAC,GAAG,CAAC,CAAC,IAAI,EAA2B,EAAE;QAClE,MAAM,QAAQ,GAAG,IAAI,CAAC,UAAU,EAAE,IAAI,CAAC,CAAC;QACxC,MAAM,OAAO,GAAG,YAAY,CAAC,QAAQ,CAAC,CAAC;QACvC,OAAO,EAAE,IAAI,EAAE,IAAI,EAAE,MAAM,EAAE,MAAM,EAAE,UAAU,CAAC,QAAQ,CAAC,CAAC,MAAM,CAAC,OAAO,CAAC,CAAC,MAAM,CAAC,KAAK,CAAC,EAAE,IAAI,EAAE,OAAO,CAAC,MAAM,EAAE,CAAC;IAClH,CAAC,CAAC,CAAC,IAAI,CAAC,CAAC,IAAI,EAAE,KAAK,EAAE,EAAE,CAAC,IAAI,CAAC,IAAI,CAAC,aAAa,CAAC,KAAK,CAAC,IAAI,CAAC,CAAC,CAAC;AAChE,CAAC;AAED,SAAS,gBAAgB,CAAC,KAAc;IACtC,IAAI,MAAM,CAAC,QAAQ,CAAC,KAAK,CAAC;QAAE,OAAO,EAAE,IAAI,EAAE,MAAM,EAAE,GAAG,EAAE,KAAK,CAAC,QAAQ,CAAC,KAAK,CAAC,EAAE,CAAC;IAChF,IAAI,KAAK,CAAC,OAAO,CAAC,KAAK,CAAC;QAAE,OAAO,KAAK,CAAC,GAAG,CAAC,gBAAgB,CAAC,CAAC;IAC7D,IAAI,KAAK,IAAI,OAAO,KAAK,KAAK,QAAQ,EAAE,CAAC;QACvC,OAAO,MAAM,CAAC,WAAW,CAAC,MAAM,CAAC,OAAO,CAAC,KAAK,CAAC,CAAC,IAAI,CAAC,CAAC,CAAC,IAAI,CAAC,EAAE,CAAC,KAAK,CAAC,EAAE,EAAE,CAAC,IAAI,CAAC,aAAa,CAAC,KAAK,CAAC,CAAC;aACjG,GAAG,CAAC,CAAC,CAAC,GAAG,EAAE,KAAK,CAAC,EAAE,EAAE,CAAC,CAAC,GAAG,EAAE,gBAAgB,CAAC,KAAK,CAAC,CAAC,CAAC,CAAC,CAAC;IAC5D,CAAC;IACD,IAAI,OAAO,KAAK,KAAK,QAAQ;QAAE,OAAO,EAAE,IAAI,EAAE,SAAS,EAAE,KAAK,EAAE,KAAK,CAAC,QAAQ,EAAE,EAAE,CAAC;IACnF,OAAO,KAAK,CAAC;AACf,CAAC;AAED,SAAS,UAAU,CAAC,IAAY,EAAE,MAAM,GAAG,EAAE;IAC3C,IAAI,CAAC,UAAU,CAAC,IAAI,CAAC;QAAE,OAAO,EAAE,CAAC;IACjC,OAAO,WAAW,CAAC,IAAI,CAAC,IAAI,EAAE,MAAM,CAAC,EAAE,EAAE,aAAa,EAAE,IAAI,EAAE,CAAC,CAAC,OAAO,CAAC,KAAK,CAAC,EAAE;QAC9E,MAAM,IAAI,GAAG,MAAM,CAAC,CAAC,CAAC,GAAG,MAAM,IAAI,KAAK,CAAC,IAAI,EAAE,CAAC,CAAC,CAAC,KAAK,CAAC,IAAI,CAAC;QAC7D,IAAI,KAAK,CAAC,WAAW,EAAE;YAAE,OAAO,UAAU,CAAC,IAAI,EAAE,IAAI,CAAC,CAAC;QACvD,IAAI,KAAK,CAAC,MAAM,EAAE,IAAI,KAAK,CAAC,cAAc,EAAE;YAAE,OAAO,CAAC,IAAI,CAAC,CAAC;QAC5D,MAAM,IAAI,KAAK,CAAC,qCAAqC,IAAI,EAAE,CAAC,CAAC;IAC/D,CAAC,CAAC,CAAC,IAAI,EAAE,CAAC;AACZ,CAAC;AAED,SAAS,GAAG,CAAC,IAAY,EAAE,IAAc;IACvC,OAAO,YAAY,CAAC,KAAK,EAAE,CAAC,IAAI,EAAE,IAAI,EAAE,GAAG,IAAI,CAAC,EAAE,EAAE,QAAQ,EAAE,MAAM,EAAE,CAAC,CAAC;AAC1E,CAAC;AAED,SAAS,eAAe,CAAC,KAAa;IACpC,OAAO,IAAI,KAAK,CAAC,UAAU,CAAC,GAAG,EAAE,IAAI,CAAC,GAAG,CAAC;AAC5C,CAAC;AAED,SAAS,WAAW,CAAC,QAA2B,EAAE,KAAa;IAC7D,OAAO,OAAO,CAAC,QAAQ,CAAC,OAAO,CAAC,+DAA+D,CAAC,CAAC,GAAG,CAAC,KAAK,CAAC,CAAC,CAAC;AAC/G,CAAC;AAED,SAAS,cAAc,CAAC,QAA2B,EAAE,KAAa,EAAE,OAAiB;IACnF,IAAI,CAAC,WAAW,CAAC,QAAQ,EAAE,KAAK,CAAC;QAAE,MAAM,IAAI,KAAK,CAAC,qCAAqC,KAAK,EAAE,CAAC,CAAC;IACjG,MAAM,MAAM,GAAG,IAAI,GAAG,CAAE,QAAQ,CAAC,MAAM,CAAC,cAAc,eAAe,CAAC,KAAK,CAAC,GAAG,CAA6B;SACzG,GAAG,CAAC,MAAM,CAAC,EAAE,CAAC,MAAM,CAAC,IAAI,CAAC,CAAC,CAAC;IAC/B,KAAK,MAAM,MAAM,IAAI,OAAO,EAAE,CAAC;QAC7B,IAAI,CAAC,MAAM,CAAC,GAAG,CAAC,MAAM,CAAC;YAAE,MAAM,IAAI,KAAK,CAAC,sCAAsC,KAAK,IAAI,MAAM,EAAE,CAAC,CAAC;IACpG,CAAC;AACH,CAAC;AAED,SAAS,mBAAmB,CAAC,WAAmB,EAAE,KAAa;IAC7D,MAAM,UAAU,GAAG,KAAK,CAAC,UAAU,CAAC,IAAI,EAAE,GAAG,CAAC,CAAC;IAC/C,IAAI,CAAC,UAAU,CAAC,UAAU,CAAC,GAAG,CAAC,IAAI,CAAC,eAAe,CAAC,IAAI,CAAC,UAAU,CAAC;QAAE,OAAO,UAAU,CAAC;IACxF,MAAM,IAAI,GAAG,QAAQ,CAAC,OAAO,CAAC,WAAW,CAAC,EAAE,OAAO,CAAC,KAAK,CAAC,CAAC,CAAC,UAAU,CAAC,IAAI,EAAE,GAAG,CAAC,CAAC;IAClF,IAAI,CAAC,IAAI,IAAI,IAAI,KAAK,IAAI,IAAI,IAAI,CAAC,UAAU,CAAC,KAAK,CAAC;QAAE,MAAM,IAAI,KAAK,CAAC,wCAAwC,CAAC,CAAC;IAChH,OAAO,IAAI,CAAC;AACd,CAAC;AAED,SAAS,MAAM,CAAC,MAAgB;IAC9B,OAAO,CAAC,GAAG,IAAI,GAAG,CAAC,MAAM,CAAC,CAAC,CAAC,IAAI,EAAE,CAAC;AACrC,CAAC;AAED,SAAS,QAAQ,CAAC,KAAoB,EAAE,KAAa;IACnD,IAAI,CAAC,KAAK,EAAE,IAAI,EAAE;QAAE,MAAM,IAAI,KAAK,CAAC,KAAK,CAAC,CAAC;IAC3C,OAAO,KAAK,CAAC;AACf,CAAC"}
|
|
@@ -0,0 +1,23 @@
|
|
|
1
|
+
export interface RuntimeDataSources {
|
|
2
|
+
sourceRoot: string;
|
|
3
|
+
codegraphDatabasePath: string;
|
|
4
|
+
contextVectorDatabasePath: string;
|
|
5
|
+
memoryDatabasePath: string;
|
|
6
|
+
knowledgeDatabasePath: string;
|
|
7
|
+
knowledgeVectorPath: string;
|
|
8
|
+
operationalDatabasePath: string;
|
|
9
|
+
}
|
|
10
|
+
export interface EvaluationRunState {
|
|
11
|
+
root: string;
|
|
12
|
+
kind: 'retrieval' | 'lifecycle';
|
|
13
|
+
caseId: string;
|
|
14
|
+
dataSources: RuntimeDataSources;
|
|
15
|
+
verifyBundleUnchanged(): Promise<void>;
|
|
16
|
+
dispose(): void;
|
|
17
|
+
}
|
|
18
|
+
export declare function createEvaluationRunState(input: {
|
|
19
|
+
bundlePath: string;
|
|
20
|
+
kind: EvaluationRunState['kind'];
|
|
21
|
+
caseId: string;
|
|
22
|
+
}): EvaluationRunState;
|
|
23
|
+
//# sourceMappingURL=evaluation-run-state.d.ts.map
|
|
@@ -0,0 +1 @@
|
|
|
1
|
+
{"version":3,"file":"evaluation-run-state.d.ts","sourceRoot":"","sources":["../src/evaluation-run-state.ts"],"names":[],"mappings":"AAiBA,MAAM,WAAW,kBAAkB;IACjC,UAAU,EAAE,MAAM,CAAC;IACnB,qBAAqB,EAAE,MAAM,CAAC;IAC9B,yBAAyB,EAAE,MAAM,CAAC;IAClC,kBAAkB,EAAE,MAAM,CAAC;IAC3B,qBAAqB,EAAE,MAAM,CAAC;IAC9B,mBAAmB,EAAE,MAAM,CAAC;IAC5B,uBAAuB,EAAE,MAAM,CAAC;CACjC;AAED,MAAM,WAAW,kBAAkB;IACjC,IAAI,EAAE,MAAM,CAAC;IACb,IAAI,EAAE,WAAW,GAAG,WAAW,CAAC;IAChC,MAAM,EAAE,MAAM,CAAC;IACf,WAAW,EAAE,kBAAkB,CAAC;IAChC,qBAAqB,IAAI,OAAO,CAAC,IAAI,CAAC,CAAC;IACvC,OAAO,IAAI,IAAI,CAAC;CACjB;AAED,wBAAgB,wBAAwB,CAAC,KAAK,EAAE;IAC9C,UAAU,EAAE,MAAM,CAAC;IACnB,IAAI,EAAE,kBAAkB,CAAC,MAAM,CAAC,CAAC;IACjC,MAAM,EAAE,MAAM,CAAC;CAChB,GAAG,kBAAkB,CAgErB"}
|
|
@@ -0,0 +1,115 @@
|
|
|
1
|
+
import { chmodSync, cpSync, existsSync, mkdirSync, mkdtempSync, readdirSync, rmSync, lstatSync, writeFileSync, } from 'node:fs';
|
|
2
|
+
import { tmpdir } from 'node:os';
|
|
3
|
+
import { dirname, join } from 'node:path';
|
|
4
|
+
import Database from 'better-sqlite3';
|
|
5
|
+
import * as sqliteVec from 'sqlite-vec';
|
|
6
|
+
import { validateEvaluationBundle } from './evaluation-bundle.js';
|
|
7
|
+
export function createEvaluationRunState(input) {
|
|
8
|
+
if (!input.caseId.trim() || input.caseId.includes('/') || input.caseId.includes('\\')) {
|
|
9
|
+
throw new Error('EVALUATION_RUN_CASE_ID_INVALID');
|
|
10
|
+
}
|
|
11
|
+
const validation = validateEvaluationBundle(input.bundlePath);
|
|
12
|
+
if (validation.status !== 'valid')
|
|
13
|
+
throw new Error(`EVALUATION_BUNDLE_INVALID:${validation.reasonCode}`);
|
|
14
|
+
const bundleId = validation.manifest.bundleId;
|
|
15
|
+
const root = mkdtempSync(join(tmpdir(), `devflow-evaluation-${input.kind}-`));
|
|
16
|
+
const sourceRoot = join(root, 'source');
|
|
17
|
+
const codegraphDatabasePath = join(sourceRoot, '.codegraph', 'codegraph.db');
|
|
18
|
+
const contextVectorDatabasePath = join(sourceRoot, '.devflow', 'symbols-vec.db');
|
|
19
|
+
const memoryDatabasePath = join(sourceRoot, '.devflow', 'memory', 'memories.db');
|
|
20
|
+
const knowledgeDatabasePath = join(sourceRoot, '.devflow', 'knowledge', 'knowledge.db');
|
|
21
|
+
const knowledgeVectorPath = join(sourceRoot, '.devflow', 'knowledge', 'assets');
|
|
22
|
+
const operationalDatabasePath = join(root, 'operational', `${input.caseId}.db`);
|
|
23
|
+
try {
|
|
24
|
+
cpSync(join(input.bundlePath, 'source'), sourceRoot, { recursive: true, dereference: false });
|
|
25
|
+
copyFile(join(input.bundlePath, 'channels/codegraph.db'), codegraphDatabasePath);
|
|
26
|
+
copyFile(join(input.bundlePath, 'channels/context-symbols.db'), contextVectorDatabasePath);
|
|
27
|
+
copyFile(join(input.bundlePath, 'channels/memory.db'), memoryDatabasePath);
|
|
28
|
+
copyFile(join(input.bundlePath, 'channels/knowledge.db'), knowledgeDatabasePath);
|
|
29
|
+
cpSync(join(input.bundlePath, 'channels/knowledge-assets'), knowledgeVectorPath, {
|
|
30
|
+
recursive: true,
|
|
31
|
+
dereference: false,
|
|
32
|
+
});
|
|
33
|
+
makeWritable(root);
|
|
34
|
+
mkdirSync(join(sourceRoot, '.devflow'), { recursive: true });
|
|
35
|
+
writeFileSync(join(sourceRoot, '.devflow', 'config.json'), `${JSON.stringify({
|
|
36
|
+
projectId: validation.manifest.projectId,
|
|
37
|
+
memoryNamespace: `memory:${validation.manifest.projectId}`,
|
|
38
|
+
})}\n`);
|
|
39
|
+
bindPortableProjectScope(memoryDatabasePath, sourceRoot);
|
|
40
|
+
bindPortableProjectScope(knowledgeDatabasePath, sourceRoot);
|
|
41
|
+
mkdirSync(join(root, 'operational'), { recursive: true });
|
|
42
|
+
}
|
|
43
|
+
catch (error) {
|
|
44
|
+
makeWritable(root);
|
|
45
|
+
rmSync(root, { recursive: true, force: true });
|
|
46
|
+
throw error;
|
|
47
|
+
}
|
|
48
|
+
return {
|
|
49
|
+
root,
|
|
50
|
+
kind: input.kind,
|
|
51
|
+
caseId: input.caseId,
|
|
52
|
+
dataSources: {
|
|
53
|
+
sourceRoot,
|
|
54
|
+
codegraphDatabasePath,
|
|
55
|
+
contextVectorDatabasePath,
|
|
56
|
+
memoryDatabasePath,
|
|
57
|
+
knowledgeDatabasePath,
|
|
58
|
+
knowledgeVectorPath,
|
|
59
|
+
operationalDatabasePath,
|
|
60
|
+
},
|
|
61
|
+
async verifyBundleUnchanged() {
|
|
62
|
+
const current = validateEvaluationBundle(input.bundlePath);
|
|
63
|
+
if (current.status !== 'valid' || current.manifest.bundleId !== bundleId) {
|
|
64
|
+
throw new Error(`EVALUATION_BUNDLE_CHANGED:${current.status === 'invalid' ? current.reasonCode : 'bundle_id'}`);
|
|
65
|
+
}
|
|
66
|
+
},
|
|
67
|
+
dispose() {
|
|
68
|
+
makeWritable(root);
|
|
69
|
+
rmSync(root, { recursive: true, force: true });
|
|
70
|
+
},
|
|
71
|
+
};
|
|
72
|
+
}
|
|
73
|
+
function bindPortableProjectScope(databasePath, sourceRoot) {
|
|
74
|
+
const database = new Database(databasePath);
|
|
75
|
+
try {
|
|
76
|
+
sqliteVec.load(database);
|
|
77
|
+
database.transaction(() => {
|
|
78
|
+
const tables = database.prepare(`SELECT name FROM sqlite_master
|
|
79
|
+
WHERE type = 'table' AND name NOT LIKE 'sqlite_%' ORDER BY name`).all();
|
|
80
|
+
for (const { name } of tables) {
|
|
81
|
+
const columns = new Set(database.pragma(`table_info(${quoteIdentifier(name)})`)
|
|
82
|
+
.map(column => column.name));
|
|
83
|
+
if (columns.has('project_root')) {
|
|
84
|
+
database.prepare(`UPDATE ${quoteIdentifier(name)} SET project_root = ? WHERE project_root = '.'`)
|
|
85
|
+
.run(sourceRoot);
|
|
86
|
+
}
|
|
87
|
+
}
|
|
88
|
+
})();
|
|
89
|
+
}
|
|
90
|
+
finally {
|
|
91
|
+
database.close();
|
|
92
|
+
}
|
|
93
|
+
}
|
|
94
|
+
function quoteIdentifier(value) {
|
|
95
|
+
return `"${value.replaceAll('"', '""')}"`;
|
|
96
|
+
}
|
|
97
|
+
function copyFile(source, destination) {
|
|
98
|
+
mkdirSync(dirname(destination), { recursive: true });
|
|
99
|
+
cpSync(source, destination, { force: false, errorOnExist: true });
|
|
100
|
+
}
|
|
101
|
+
function makeWritable(path) {
|
|
102
|
+
if (!existsSync(path))
|
|
103
|
+
return;
|
|
104
|
+
const stat = lstatSync(path);
|
|
105
|
+
if (stat.isSymbolicLink())
|
|
106
|
+
return;
|
|
107
|
+
if (!stat.isDirectory()) {
|
|
108
|
+
chmodSync(path, 0o644);
|
|
109
|
+
return;
|
|
110
|
+
}
|
|
111
|
+
chmodSync(path, 0o755);
|
|
112
|
+
for (const name of readdirSync(path))
|
|
113
|
+
makeWritable(join(path, name));
|
|
114
|
+
}
|
|
115
|
+
//# sourceMappingURL=evaluation-run-state.js.map
|
|
@@ -0,0 +1 @@
|
|
|
1
|
+
{"version":3,"file":"evaluation-run-state.js","sourceRoot":"","sources":["../src/evaluation-run-state.ts"],"names":[],"mappings":"AAAA,OAAO,EACL,SAAS,EACT,MAAM,EACN,UAAU,EACV,SAAS,EACT,WAAW,EACX,WAAW,EACX,MAAM,EACN,SAAS,EACT,aAAa,GACd,MAAM,SAAS,CAAC;AACjB,OAAO,EAAE,MAAM,EAAE,MAAM,SAAS,CAAC;AACjC,OAAO,EAAE,OAAO,EAAE,IAAI,EAAE,MAAM,WAAW,CAAC;AAC1C,OAAO,QAAQ,MAAM,gBAAgB,CAAC;AACtC,OAAO,KAAK,SAAS,MAAM,YAAY,CAAC;AACxC,OAAO,EAAE,wBAAwB,EAAE,MAAM,wBAAwB,CAAC;AAqBlE,MAAM,UAAU,wBAAwB,CAAC,KAIxC;IACC,IAAI,CAAC,KAAK,CAAC,MAAM,CAAC,IAAI,EAAE,IAAI,KAAK,CAAC,MAAM,CAAC,QAAQ,CAAC,GAAG,CAAC,IAAI,KAAK,CAAC,MAAM,CAAC,QAAQ,CAAC,IAAI,CAAC,EAAE,CAAC;QACtF,MAAM,IAAI,KAAK,CAAC,gCAAgC,CAAC,CAAC;IACpD,CAAC;IACD,MAAM,UAAU,GAAG,wBAAwB,CAAC,KAAK,CAAC,UAAU,CAAC,CAAC;IAC9D,IAAI,UAAU,CAAC,MAAM,KAAK,OAAO;QAAE,MAAM,IAAI,KAAK,CAAC,6BAA6B,UAAU,CAAC,UAAU,EAAE,CAAC,CAAC;IACzG,MAAM,QAAQ,GAAG,UAAU,CAAC,QAAQ,CAAC,QAAQ,CAAC;IAC9C,MAAM,IAAI,GAAG,WAAW,CAAC,IAAI,CAAC,MAAM,EAAE,EAAE,sBAAsB,KAAK,CAAC,IAAI,GAAG,CAAC,CAAC,CAAC;IAC9E,MAAM,UAAU,GAAG,IAAI,CAAC,IAAI,EAAE,QAAQ,CAAC,CAAC;IACxC,MAAM,qBAAqB,GAAG,IAAI,CAAC,UAAU,EAAE,YAAY,EAAE,cAAc,CAAC,CAAC;IAC7E,MAAM,yBAAyB,GAAG,IAAI,CAAC,UAAU,EAAE,UAAU,EAAE,gBAAgB,CAAC,CAAC;IACjF,MAAM,kBAAkB,GAAG,IAAI,CAAC,UAAU,EAAE,UAAU,EAAE,QAAQ,EAAE,aAAa,CAAC,CAAC;IACjF,MAAM,qBAAqB,GAAG,IAAI,CAAC,UAAU,EAAE,UAAU,EAAE,WAAW,EAAE,cAAc,CAAC,CAAC;IACxF,MAAM,mBAAmB,GAAG,IAAI,CAAC,UAAU,EAAE,UAAU,EAAE,WAAW,EAAE,QAAQ,CAAC,CAAC;IAChF,MAAM,uBAAuB,GAAG,IAAI,CAAC,IAAI,EAAE,aAAa,EAAE,GAAG,KAAK,CAAC,MAAM,KAAK,CAAC,CAAC;IAChF,IAAI,CAAC;QACH,MAAM,CAAC,IAAI,CAAC,KAAK,CAAC,UAAU,EAAE,QAAQ,CAAC,EAAE,UAAU,EAAE,EAAE,SAAS,EAAE,IAAI,EAAE,WAAW,EAAE,KAAK,EAAE,CAAC,CAAC;QAC9F,QAAQ,CAAC,IAAI,CAAC,KAAK,CAAC,UAAU,EAAE,uBAAuB,CAAC,EAAE,qBAAqB,CAAC,CAAC;QACjF,QAAQ,CAAC,IAAI,CAAC,KAAK,CAAC,UAAU,EAAE,6BAA6B,CAAC,EAAE,yBAAyB,CAAC,CAAC;QAC3F,QAAQ,CAAC,IAAI,CAAC,KAAK,CAAC,UAAU,EAAE,oBAAoB,CAAC,EAAE,kBAAkB,CAAC,CAAC;QAC3E,QAAQ,CAAC,IAAI,CAAC,KAAK,CAAC,UAAU,EAAE,uBAAuB,CAAC,EAAE,qBAAqB,CAAC,CAAC;QACjF,MAAM,CAAC,IAAI,CAAC,KAAK,CAAC,UAAU,EAAE,2BAA2B,CAAC,EAAE,mBAAmB,EAAE;YAC/E,SAAS,EAAE,IAAI;YACf,WAAW,EAAE,KAAK;SACnB,CAAC,CAAC;QACH,YAAY,CAAC,IAAI,CAAC,CAAC;QACnB,SAAS,CAAC,IAAI,CAAC,UAAU,EAAE,UAAU,CAAC,EAAE,EAAE,SAAS,EAAE,IAAI,EAAE,CAAC,CAAC;QAC7D,aAAa,CAAC,IAAI,CAAC,UAAU,EAAE,UAAU,EAAE,aAAa,CAAC,EAAE,GAAG,IAAI,CAAC,SAAS,CAAC;YAC3E,SAAS,EAAE,UAAU,CAAC,QAAQ,CAAC,SAAS;YACxC,eAAe,EAAE,UAAU,UAAU,CAAC,QAAQ,CAAC,SAAS,EAAE;SAC3D,CAAC,IAAI,CAAC,CAAC;QACR,wBAAwB,CAAC,kBAAkB,EAAE,UAAU,CAAC,CAAC;QACzD,wBAAwB,CAAC,qBAAqB,EAAE,UAAU,CAAC,CAAC;QAC5D,SAAS,CAAC,IAAI,CAAC,IAAI,EAAE,aAAa,CAAC,EAAE,EAAE,SAAS,EAAE,IAAI,EAAE,CAAC,CAAC;IAC5D,CAAC;IAAC,OAAO,KAAK,EAAE,CAAC;QACf,YAAY,CAAC,IAAI,CAAC,CAAC;QACnB,MAAM,CAAC,IAAI,EAAE,EAAE,SAAS,EAAE,IAAI,EAAE,KAAK,EAAE,IAAI,EAAE,CAAC,CAAC;QAC/C,MAAM,KAAK,CAAC;IACd,CAAC;IAED,OAAO;QACL,IAAI;QACJ,IAAI,EAAE,KAAK,CAAC,IAAI;QAChB,MAAM,EAAE,KAAK,CAAC,MAAM;QACpB,WAAW,EAAE;YACX,UAAU;YACV,qBAAqB;YACrB,yBAAyB;YACzB,kBAAkB;YAClB,qBAAqB;YACrB,mBAAmB;YACnB,uBAAuB;SACxB;QACD,KAAK,CAAC,qBAAqB;YACzB,MAAM,OAAO,GAAG,wBAAwB,CAAC,KAAK,CAAC,UAAU,CAAC,CAAC;YAC3D,IAAI,OAAO,CAAC,MAAM,KAAK,OAAO,IAAI,OAAO,CAAC,QAAQ,CAAC,QAAQ,KAAK,QAAQ,EAAE,CAAC;gBACzE,MAAM,IAAI,KAAK,CAAC,6BAA6B,OAAO,CAAC,MAAM,KAAK,SAAS,CAAC,CAAC,CAAC,OAAO,CAAC,UAAU,CAAC,CAAC,CAAC,WAAW,EAAE,CAAC,CAAC;YAClH,CAAC;QACH,CAAC;QACD,OAAO;YACL,YAAY,CAAC,IAAI,CAAC,CAAC;YACnB,MAAM,CAAC,IAAI,EAAE,EAAE,SAAS,EAAE,IAAI,EAAE,KAAK,EAAE,IAAI,EAAE,CAAC,CAAC;QACjD,CAAC;KACF,CAAC;AACJ,CAAC;AAED,SAAS,wBAAwB,CAAC,YAAoB,EAAE,UAAkB;IACxE,MAAM,QAAQ,GAAG,IAAI,QAAQ,CAAC,YAAY,CAAC,CAAC;IAC5C,IAAI,CAAC;QACH,SAAS,CAAC,IAAI,CAAC,QAAQ,CAAC,CAAC;QACzB,QAAQ,CAAC,WAAW,CAAC,GAAG,EAAE;YACxB,MAAM,MAAM,GAAG,QAAQ,CAAC,OAAO,CAAC;wEACkC,CAAC,CAAC,GAAG,EAA6B,CAAC;YACrG,KAAK,MAAM,EAAE,IAAI,EAAE,IAAI,MAAM,EAAE,CAAC;gBAC9B,MAAM,OAAO,GAAG,IAAI,GAAG,CAAE,QAAQ,CAAC,MAAM,CAAC,cAAc,eAAe,CAAC,IAAI,CAAC,GAAG,CAA6B;qBACzG,GAAG,CAAC,MAAM,CAAC,EAAE,CAAC,MAAM,CAAC,IAAI,CAAC,CAAC,CAAC;gBAC/B,IAAI,OAAO,CAAC,GAAG,CAAC,cAAc,CAAC,EAAE,CAAC;oBAChC,QAAQ,CAAC,OAAO,CAAC,UAAU,eAAe,CAAC,IAAI,CAAC,gDAAgD,CAAC;yBAC9F,GAAG,CAAC,UAAU,CAAC,CAAC;gBACrB,CAAC;YACH,CAAC;QACH,CAAC,CAAC,EAAE,CAAC;IACP,CAAC;YAAS,CAAC;QACT,QAAQ,CAAC,KAAK,EAAE,CAAC;IACnB,CAAC;AACH,CAAC;AAED,SAAS,eAAe,CAAC,KAAa;IACpC,OAAO,IAAI,KAAK,CAAC,UAAU,CAAC,GAAG,EAAE,IAAI,CAAC,GAAG,CAAC;AAC5C,CAAC;AAED,SAAS,QAAQ,CAAC,MAAc,EAAE,WAAmB;IACnD,SAAS,CAAC,OAAO,CAAC,WAAW,CAAC,EAAE,EAAE,SAAS,EAAE,IAAI,EAAE,CAAC,CAAC;IACrD,MAAM,CAAC,MAAM,EAAE,WAAW,EAAE,EAAE,KAAK,EAAE,KAAK,EAAE,YAAY,EAAE,IAAI,EAAE,CAAC,CAAC;AACpE,CAAC;AAED,SAAS,YAAY,CAAC,IAAY;IAChC,IAAI,CAAC,UAAU,CAAC,IAAI,CAAC;QAAE,OAAO;IAC9B,MAAM,IAAI,GAAG,SAAS,CAAC,IAAI,CAAC,CAAC;IAC7B,IAAI,IAAI,CAAC,cAAc,EAAE;QAAE,OAAO;IAClC,IAAI,CAAC,IAAI,CAAC,WAAW,EAAE,EAAE,CAAC;QACxB,SAAS,CAAC,IAAI,EAAE,KAAK,CAAC,CAAC;QACvB,OAAO;IACT,CAAC;IACD,SAAS,CAAC,IAAI,EAAE,KAAK,CAAC,CAAC;IACvB,KAAK,MAAM,IAAI,IAAI,WAAW,CAAC,IAAI,CAAC;QAAE,YAAY,CAAC,IAAI,CAAC,IAAI,EAAE,IAAI,CAAC,CAAC,CAAC;AACvE,CAAC"}
|
|
@@ -16,5 +16,20 @@ export interface EvidenceQualityMetadata {
|
|
|
16
16
|
latencyBudgetMs: number;
|
|
17
17
|
observedLatencyMs: number;
|
|
18
18
|
}
|
|
19
|
+
export interface AggregatedEvidenceQualityMetric {
|
|
20
|
+
caseMacro: number | null;
|
|
21
|
+
refMicro: EvidenceQualityMetric;
|
|
22
|
+
}
|
|
19
23
|
export declare function evidenceQualityMetric(numerator: number, denominator: number): EvidenceQualityMetric;
|
|
24
|
+
export declare function candidateRecallAt20(candidateRefs: string[], goldRefs: Set<string>): EvidenceQualityMetric;
|
|
25
|
+
export declare function eligibilitySurvival(candidateRefs: string[], eligibilityRejectedRefs: string[], goldRefs: Set<string>): EvidenceQualityMetric;
|
|
26
|
+
export declare function recallAt5(selectedRefs: string[], goldRefs: Set<string>): EvidenceQualityMetric;
|
|
27
|
+
export declare function precisionAt5(selectedRefs: string[], goldRefs: Set<string>): EvidenceQualityMetric;
|
|
28
|
+
export declare function forbiddenContaminationAt5(selectedRefs: string[], forbiddenRefs: Set<string>): EvidenceQualityMetric;
|
|
29
|
+
export declare function selectBestGoldSet(input: {
|
|
30
|
+
goldSets: string[][];
|
|
31
|
+
candidateRefs: string[];
|
|
32
|
+
selectedRefs: string[];
|
|
33
|
+
}): string[];
|
|
34
|
+
export declare function aggregateEvidenceQualityMetrics(metrics: EvidenceQualityMetric[]): AggregatedEvidenceQualityMetric;
|
|
20
35
|
//# sourceMappingURL=evidence-quality-metrics.d.ts.map
|
|
@@ -1 +1 @@
|
|
|
1
|
-
{"version":3,"file":"evidence-quality-metrics.d.ts","sourceRoot":"","sources":["../src/evidence-quality-metrics.ts"],"names":[],"mappings":"
|
|
1
|
+
{"version":3,"file":"evidence-quality-metrics.d.ts","sourceRoot":"","sources":["../src/evidence-quality-metrics.ts"],"names":[],"mappings":"AAEA,MAAM,WAAW,gBAAgB;IAC/B,KAAK,EAAE,MAAM,CAAC;IACd,KAAK,EAAE,MAAM,CAAC;CACf;AAED,MAAM,WAAW,qBAAqB;IACpC,SAAS,EAAE,MAAM,CAAC;IAClB,WAAW,EAAE,MAAM,CAAC;IACpB,KAAK,EAAE,MAAM,GAAG,IAAI,CAAC;IACrB,QAAQ,EAAE,gBAAgB,GAAG,IAAI,CAAC;CACnC;AAED,MAAM,WAAW,uBAAuB;IACtC,KAAK,EAAE,MAAM,CAAC;IACd,aAAa,EAAE,MAAM,CAAC;IACtB,KAAK,EAAE,MAAM,CAAC;IACd,WAAW,EAAE,MAAM,CAAC;IACpB,eAAe,EAAE,MAAM,CAAC;IACxB,iBAAiB,EAAE,MAAM,CAAC;CAC3B;AAED,MAAM,WAAW,+BAA+B;IAC9C,SAAS,EAAE,MAAM,GAAG,IAAI,CAAC;IACzB,QAAQ,EAAE,qBAAqB,CAAC;CACjC;AAED,wBAAgB,qBAAqB,CAAC,SAAS,EAAE,MAAM,EAAE,WAAW,EAAE,MAAM,GAAG,qBAAqB,CAuBnG;AAED,wBAAgB,mBAAmB,CAAC,aAAa,EAAE,MAAM,EAAE,EAAE,QAAQ,EAAE,GAAG,CAAC,MAAM,CAAC,GAAG,qBAAqB,CAGzG;AAED,wBAAgB,mBAAmB,CACjC,aAAa,EAAE,MAAM,EAAE,EACvB,uBAAuB,EAAE,MAAM,EAAE,EACjC,QAAQ,EAAE,GAAG,CAAC,MAAM,CAAC,GACpB,qBAAqB,CAIvB;AAED,wBAAgB,SAAS,CAAC,YAAY,EAAE,MAAM,EAAE,EAAE,QAAQ,EAAE,GAAG,CAAC,MAAM,CAAC,GAAG,qBAAqB,CAG9F;AAED,wBAAgB,YAAY,CAAC,YAAY,EAAE,MAAM,EAAE,EAAE,QAAQ,EAAE,GAAG,CAAC,MAAM,CAAC,GAAG,qBAAqB,CAEjG;AAED,wBAAgB,yBAAyB,CACvC,YAAY,EAAE,MAAM,EAAE,EACtB,aAAa,EAAE,GAAG,CAAC,MAAM,CAAC,GACzB,qBAAqB,CAEvB;AAED,wBAAgB,iBAAiB,CAAC,KAAK,EAAE;IACvC,QAAQ,EAAE,MAAM,EAAE,EAAE,CAAC;IACrB,aAAa,EAAE,MAAM,EAAE,CAAC;IACxB,YAAY,EAAE,MAAM,EAAE,CAAC;CACxB,GAAG,MAAM,EAAE,CAWX;AAED,wBAAgB,+BAA+B,CAC7C,OAAO,EAAE,qBAAqB,EAAE,GAC/B,+BAA+B,CAWjC"}
|
|
@@ -1,5 +1,6 @@
|
|
|
1
|
+
import { stableRuntimeHash } from '@devflow-tools/sdk';
|
|
1
2
|
export function evidenceQualityMetric(numerator, denominator) {
|
|
2
|
-
if (!Number.
|
|
3
|
+
if (!Number.isSafeInteger(numerator) || !Number.isSafeInteger(denominator)
|
|
3
4
|
|| numerator < 0 || denominator < 0 || numerator > denominator) {
|
|
4
5
|
throw new Error(`Invalid evidence quality metric: ${numerator}/${denominator}`);
|
|
5
6
|
}
|
|
@@ -22,4 +23,53 @@ export function evidenceQualityMetric(numerator, denominator) {
|
|
|
22
23
|
},
|
|
23
24
|
};
|
|
24
25
|
}
|
|
26
|
+
export function candidateRecallAt20(candidateRefs, goldRefs) {
|
|
27
|
+
const candidates = new Set(uniqueTopK(candidateRefs, 20));
|
|
28
|
+
return evidenceQualityMetric([...goldRefs].filter(ref => candidates.has(ref)).length, goldRefs.size);
|
|
29
|
+
}
|
|
30
|
+
export function eligibilitySurvival(candidateRefs, eligibilityRejectedRefs, goldRefs) {
|
|
31
|
+
const candidateGold = uniqueTopK(candidateRefs, 20).filter(ref => goldRefs.has(ref));
|
|
32
|
+
const rejected = new Set(eligibilityRejectedRefs);
|
|
33
|
+
return evidenceQualityMetric(candidateGold.filter(ref => !rejected.has(ref)).length, candidateGold.length);
|
|
34
|
+
}
|
|
35
|
+
export function recallAt5(selectedRefs, goldRefs) {
|
|
36
|
+
const selected = new Set(uniqueTopK(selectedRefs, 5));
|
|
37
|
+
return evidenceQualityMetric([...goldRefs].filter(ref => selected.has(ref)).length, goldRefs.size);
|
|
38
|
+
}
|
|
39
|
+
export function precisionAt5(selectedRefs, goldRefs) {
|
|
40
|
+
return evidenceQualityMetric(uniqueTopK(selectedRefs, 5).filter(ref => goldRefs.has(ref)).length, 5);
|
|
41
|
+
}
|
|
42
|
+
export function forbiddenContaminationAt5(selectedRefs, forbiddenRefs) {
|
|
43
|
+
return evidenceQualityMetric(uniqueTopK(selectedRefs, 5).filter(ref => forbiddenRefs.has(ref)).length, 5);
|
|
44
|
+
}
|
|
45
|
+
export function selectBestGoldSet(input) {
|
|
46
|
+
if (input.goldSets.length === 0)
|
|
47
|
+
return [];
|
|
48
|
+
const selected = new Set(uniqueTopK(input.selectedRefs, 5));
|
|
49
|
+
const candidates = new Set(uniqueTopK(input.candidateRefs, 20));
|
|
50
|
+
return input.goldSets.map(refs => [...new Set(refs)].sort()).sort((left, right) => {
|
|
51
|
+
const selectedOverlap = overlap(right, selected) - overlap(left, selected);
|
|
52
|
+
if (selectedOverlap !== 0)
|
|
53
|
+
return selectedOverlap;
|
|
54
|
+
const candidateOverlap = overlap(right, candidates) - overlap(left, candidates);
|
|
55
|
+
if (candidateOverlap !== 0)
|
|
56
|
+
return candidateOverlap;
|
|
57
|
+
return stableRuntimeHash({ refs: left }).localeCompare(stableRuntimeHash({ refs: right }));
|
|
58
|
+
})[0];
|
|
59
|
+
}
|
|
60
|
+
export function aggregateEvidenceQualityMetrics(metrics) {
|
|
61
|
+
const measurable = metrics.filter(metric => metric.denominator > 0);
|
|
62
|
+
return {
|
|
63
|
+
caseMacro: measurable.length === 0
|
|
64
|
+
? null
|
|
65
|
+
: measurable.reduce((total, metric) => total + metric.numerator / metric.denominator, 0) / measurable.length,
|
|
66
|
+
refMicro: evidenceQualityMetric(measurable.reduce((total, metric) => total + metric.numerator, 0), measurable.reduce((total, metric) => total + metric.denominator, 0)),
|
|
67
|
+
};
|
|
68
|
+
}
|
|
69
|
+
function uniqueTopK(refs, limit) {
|
|
70
|
+
return [...new Set(refs)].slice(0, limit);
|
|
71
|
+
}
|
|
72
|
+
function overlap(refs, selected) {
|
|
73
|
+
return refs.filter(ref => selected.has(ref)).length;
|
|
74
|
+
}
|
|
25
75
|
//# sourceMappingURL=evidence-quality-metrics.js.map
|
|
@@ -1 +1 @@
|
|
|
1
|
-
{"version":3,"file":"evidence-quality-metrics.js","sourceRoot":"","sources":["../src/evidence-quality-metrics.ts"],"names":[],"mappings":"
|
|
1
|
+
{"version":3,"file":"evidence-quality-metrics.js","sourceRoot":"","sources":["../src/evidence-quality-metrics.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,iBAAiB,EAAE,MAAM,oBAAoB,CAAC;AA4BvD,MAAM,UAAU,qBAAqB,CAAC,SAAiB,EAAE,WAAmB;IAC1E,IAAI,CAAC,MAAM,CAAC,aAAa,CAAC,SAAS,CAAC,IAAI,CAAC,MAAM,CAAC,aAAa,CAAC,WAAW,CAAC;WACrE,SAAS,GAAG,CAAC,IAAI,WAAW,GAAG,CAAC,IAAI,SAAS,GAAG,WAAW,EAAE,CAAC;QACjE,MAAM,IAAI,KAAK,CAAC,oCAAoC,SAAS,IAAI,WAAW,EAAE,CAAC,CAAC;IAClF,CAAC;IACD,IAAI,WAAW,KAAK,CAAC,EAAE,CAAC;QACtB,OAAO,EAAE,SAAS,EAAE,WAAW,EAAE,KAAK,EAAE,IAAI,EAAE,QAAQ,EAAE,IAAI,EAAE,CAAC;IACjE,CAAC;IACD,MAAM,KAAK,GAAG,SAAS,GAAG,WAAW,CAAC;IACtC,MAAM,CAAC,GAAG,iBAAiB,CAAC;IAC5B,MAAM,EAAE,GAAG,CAAC,GAAG,CAAC,CAAC;IACjB,MAAM,KAAK,GAAG,CAAC,GAAG,EAAE,GAAG,WAAW,CAAC;IACnC,MAAM,MAAM,GAAG,CAAC,KAAK,GAAG,EAAE,GAAG,CAAC,CAAC,GAAG,WAAW,CAAC,CAAC,GAAG,KAAK,CAAC;IACxD,MAAM,MAAM,GAAG,CAAC,GAAG,IAAI,CAAC,IAAI,CAAC,CAAC,KAAK,GAAG,CAAC,CAAC,GAAG,KAAK,CAAC,GAAG,EAAE,GAAG,CAAC,CAAC,GAAG,WAAW,CAAC,CAAC,GAAG,WAAW,CAAC,GAAG,KAAK,CAAC;IACnG,OAAO;QACL,SAAS;QACT,WAAW;QACX,KAAK;QACL,QAAQ,EAAE;YACR,KAAK,EAAE,IAAI,CAAC,GAAG,CAAC,CAAC,EAAE,MAAM,GAAG,MAAM,CAAC;YACnC,KAAK,EAAE,IAAI,CAAC,GAAG,CAAC,CAAC,EAAE,MAAM,GAAG,MAAM,CAAC;SACpC;KACF,CAAC;AACJ,CAAC;AAED,MAAM,UAAU,mBAAmB,CAAC,aAAuB,EAAE,QAAqB;IAChF,MAAM,UAAU,GAAG,IAAI,GAAG,CAAC,UAAU,CAAC,aAAa,EAAE,EAAE,CAAC,CAAC,CAAC;IAC1D,OAAO,qBAAqB,CAAC,CAAC,GAAG,QAAQ,CAAC,CAAC,MAAM,CAAC,GAAG,CAAC,EAAE,CAAC,UAAU,CAAC,GAAG,CAAC,GAAG,CAAC,CAAC,CAAC,MAAM,EAAE,QAAQ,CAAC,IAAI,CAAC,CAAC;AACvG,CAAC;AAED,MAAM,UAAU,mBAAmB,CACjC,aAAuB,EACvB,uBAAiC,EACjC,QAAqB;IAErB,MAAM,aAAa,GAAG,UAAU,CAAC,aAAa,EAAE,EAAE,CAAC,CAAC,MAAM,CAAC,GAAG,CAAC,EAAE,CAAC,QAAQ,CAAC,GAAG,CAAC,GAAG,CAAC,CAAC,CAAC;IACrF,MAAM,QAAQ,GAAG,IAAI,GAAG,CAAC,uBAAuB,CAAC,CAAC;IAClD,OAAO,qBAAqB,CAAC,aAAa,CAAC,MAAM,CAAC,GAAG,CAAC,EAAE,CAAC,CAAC,QAAQ,CAAC,GAAG,CAAC,GAAG,CAAC,CAAC,CAAC,MAAM,EAAE,aAAa,CAAC,MAAM,CAAC,CAAC;AAC7G,CAAC;AAED,MAAM,UAAU,SAAS,CAAC,YAAsB,EAAE,QAAqB;IACrE,MAAM,QAAQ,GAAG,IAAI,GAAG,CAAC,UAAU,CAAC,YAAY,EAAE,CAAC,CAAC,CAAC,CAAC;IACtD,OAAO,qBAAqB,CAAC,CAAC,GAAG,QAAQ,CAAC,CAAC,MAAM,CAAC,GAAG,CAAC,EAAE,CAAC,QAAQ,CAAC,GAAG,CAAC,GAAG,CAAC,CAAC,CAAC,MAAM,EAAE,QAAQ,CAAC,IAAI,CAAC,CAAC;AACrG,CAAC;AAED,MAAM,UAAU,YAAY,CAAC,YAAsB,EAAE,QAAqB;IACxE,OAAO,qBAAqB,CAAC,UAAU,CAAC,YAAY,EAAE,CAAC,CAAC,CAAC,MAAM,CAAC,GAAG,CAAC,EAAE,CAAC,QAAQ,CAAC,GAAG,CAAC,GAAG,CAAC,CAAC,CAAC,MAAM,EAAE,CAAC,CAAC,CAAC;AACvG,CAAC;AAED,MAAM,UAAU,yBAAyB,CACvC,YAAsB,EACtB,aAA0B;IAE1B,OAAO,qBAAqB,CAAC,UAAU,CAAC,YAAY,EAAE,CAAC,CAAC,CAAC,MAAM,CAAC,GAAG,CAAC,EAAE,CAAC,aAAa,CAAC,GAAG,CAAC,GAAG,CAAC,CAAC,CAAC,MAAM,EAAE,CAAC,CAAC,CAAC;AAC5G,CAAC;AAED,MAAM,UAAU,iBAAiB,CAAC,KAIjC;IACC,IAAI,KAAK,CAAC,QAAQ,CAAC,MAAM,KAAK,CAAC;QAAE,OAAO,EAAE,CAAC;IAC3C,MAAM,QAAQ,GAAG,IAAI,GAAG,CAAC,UAAU,CAAC,KAAK,CAAC,YAAY,EAAE,CAAC,CAAC,CAAC,CAAC;IAC5D,MAAM,UAAU,GAAG,IAAI,GAAG,CAAC,UAAU,CAAC,KAAK,CAAC,aAAa,EAAE,EAAE,CAAC,CAAC,CAAC;IAChE,OAAO,KAAK,CAAC,QAAQ,CAAC,GAAG,CAAC,IAAI,CAAC,EAAE,CAAC,CAAC,GAAG,IAAI,GAAG,CAAC,IAAI,CAAC,CAAC,CAAC,IAAI,EAAE,CAAC,CAAC,IAAI,CAAC,CAAC,IAAI,EAAE,KAAK,EAAE,EAAE;QAChF,MAAM,eAAe,GAAG,OAAO,CAAC,KAAK,EAAE,QAAQ,CAAC,GAAG,OAAO,CAAC,IAAI,EAAE,QAAQ,CAAC,CAAC;QAC3E,IAAI,eAAe,KAAK,CAAC;YAAE,OAAO,eAAe,CAAC;QAClD,MAAM,gBAAgB,GAAG,OAAO,CAAC,KAAK,EAAE,UAAU,CAAC,GAAG,OAAO,CAAC,IAAI,EAAE,UAAU,CAAC,CAAC;QAChF,IAAI,gBAAgB,KAAK,CAAC;YAAE,OAAO,gBAAgB,CAAC;QACpD,OAAO,iBAAiB,CAAC,EAAE,IAAI,EAAE,IAAI,EAAE,CAAC,CAAC,aAAa,CAAC,iBAAiB,CAAC,EAAE,IAAI,EAAE,KAAK,EAAE,CAAC,CAAC,CAAC;IAC7F,CAAC,CAAC,CAAC,CAAC,CAAE,CAAC;AACT,CAAC;AAED,MAAM,UAAU,+BAA+B,CAC7C,OAAgC;IAEhC,MAAM,UAAU,GAAG,OAAO,CAAC,MAAM,CAAC,MAAM,CAAC,EAAE,CAAC,MAAM,CAAC,WAAW,GAAG,CAAC,CAAC,CAAC;IACpE,OAAO;QACL,SAAS,EAAE,UAAU,CAAC,MAAM,KAAK,CAAC;YAChC,CAAC,CAAC,IAAI;YACN,CAAC,CAAC,UAAU,CAAC,MAAM,CAAC,CAAC,KAAK,EAAE,MAAM,EAAE,EAAE,CAAC,KAAK,GAAG,MAAM,CAAC,SAAS,GAAG,MAAM,CAAC,WAAW,EAAE,CAAC,CAAC,GAAG,UAAU,CAAC,MAAM;QAC9G,QAAQ,EAAE,qBAAqB,CAC7B,UAAU,CAAC,MAAM,CAAC,CAAC,KAAK,EAAE,MAAM,EAAE,EAAE,CAAC,KAAK,GAAG,MAAM,CAAC,SAAS,EAAE,CAAC,CAAC,EACjE,UAAU,CAAC,MAAM,CAAC,CAAC,KAAK,EAAE,MAAM,EAAE,EAAE,CAAC,KAAK,GAAG,MAAM,CAAC,WAAW,EAAE,CAAC,CAAC,CACpE;KACF,CAAC;AACJ,CAAC;AAED,SAAS,UAAU,CAAC,IAAc,EAAE,KAAa;IAC/C,OAAO,CAAC,GAAG,IAAI,GAAG,CAAC,IAAI,CAAC,CAAC,CAAC,KAAK,CAAC,CAAC,EAAE,KAAK,CAAC,CAAC;AAC5C,CAAC;AAED,SAAS,OAAO,CAAC,IAAc,EAAE,QAAqB;IACpD,OAAO,IAAI,CAAC,MAAM,CAAC,GAAG,CAAC,EAAE,CAAC,QAAQ,CAAC,GAAG,CAAC,GAAG,CAAC,CAAC,CAAC,MAAM,CAAC;AACtD,CAAC"}
|
|
@@ -1 +1 @@
|
|
|
1
|
-
{"version":3,"file":"deterministic.d.ts","sourceRoot":"","sources":["../../src/graders/deterministic.ts"],"names":[],"mappings":"AAEA,OAAO,KAAK,EAAiB,gBAAgB,EAAE,YAAY,EAAE,MAAM,YAAY,CAAC;AAKhF,eAAO,MAAM,qBAAqB,EAAE,gBAAgB,EAWnD,CAAC;AAEF,wBAAgB,uBAAuB,IAAI,gBAAgB,CAuC1D;AAED,wBAAsB,iBAAiB,CACrC,IAAI,EAAE,MAAM,EACZ,IAAI,EAAE,MAAM,EAAE,EACd,OAAO,EAAE;IAAE,GAAG,EAAE,MAAM,CAAC;IAAC,SAAS,CAAC,EAAE,MAAM,CAAC;IAAC,cAAc,CAAC,EAAE,MAAM,CAAA;CAAE,GACpE,OAAO,CAAC;IAAE,QAAQ,EAAE,MAAM,CAAC;IAAC,QAAQ,EAAE,OAAO,CAAC;IAAC,UAAU,EAAE,MAAM,CAAC;IAAC,SAAS,EAAE,OAAO,CAAA;CAAE,CAAC,CAgC1F;
|
|
1
|
+
{"version":3,"file":"deterministic.d.ts","sourceRoot":"","sources":["../../src/graders/deterministic.ts"],"names":[],"mappings":"AAEA,OAAO,KAAK,EAAiB,gBAAgB,EAAE,YAAY,EAAE,MAAM,YAAY,CAAC;AAKhF,eAAO,MAAM,qBAAqB,EAAE,gBAAgB,EAWnD,CAAC;AAEF,wBAAgB,uBAAuB,IAAI,gBAAgB,CAuC1D;AAED,wBAAsB,iBAAiB,CACrC,IAAI,EAAE,MAAM,EACZ,IAAI,EAAE,MAAM,EAAE,EACd,OAAO,EAAE;IAAE,GAAG,EAAE,MAAM,CAAC;IAAC,SAAS,CAAC,EAAE,MAAM,CAAC;IAAC,cAAc,CAAC,EAAE,MAAM,CAAA;CAAE,GACpE,OAAO,CAAC;IAAE,QAAQ,EAAE,MAAM,CAAC;IAAC,QAAQ,EAAE,OAAO,CAAC;IAAC,UAAU,EAAE,MAAM,CAAC;IAAC,SAAS,EAAE,OAAO,CAAA;CAAE,CAAC,CAgC1F;AAqHD,wBAAgB,MAAM,CACpB,QAAQ,EAAE,MAAM,EAChB,MAAM,EAAE,YAAY,CAAC,QAAQ,CAAC,EAC9B,MAAM,EAAE,MAAM,EACd,SAAS,EAAE,MAAM,EACjB,QAAQ,EAAE,YAAY,CAAC,UAAU,CAAC,EAClC,KAAK,CAAC,EAAE,MAAM,EACd,SAAS,CAAC,EAAE,MAAM,GACjB,YAAY,CAMd"}
|