ragscope 0.1.0
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 +183 -0
- package/README.md +135 -0
- package/dist/bin/ragscope.d.ts +3 -0
- package/dist/bin/ragscope.d.ts.map +1 -0
- package/dist/bin/ragscope.js +135 -0
- package/dist/bin/ragscope.js.map +1 -0
- package/dist/src/app.d.ts +5 -0
- package/dist/src/app.d.ts.map +1 -0
- package/dist/src/app.js +88 -0
- package/dist/src/app.js.map +1 -0
- package/dist/src/app.test.d.ts +2 -0
- package/dist/src/app.test.d.ts.map +1 -0
- package/dist/src/app.test.js +59 -0
- package/dist/src/app.test.js.map +1 -0
- package/dist/src/audit/scorer.d.ts +17 -0
- package/dist/src/audit/scorer.d.ts.map +1 -0
- package/dist/src/audit/scorer.js +85 -0
- package/dist/src/audit/scorer.js.map +1 -0
- package/dist/src/audit/scorer.test.d.ts +2 -0
- package/dist/src/audit/scorer.test.d.ts.map +1 -0
- package/dist/src/audit/scorer.test.js +101 -0
- package/dist/src/audit/scorer.test.js.map +1 -0
- package/dist/src/db/index.d.ts +731 -0
- package/dist/src/db/index.d.ts.map +1 -0
- package/dist/src/db/index.js +59 -0
- package/dist/src/db/index.js.map +1 -0
- package/dist/src/db/queries.d.ts +12 -0
- package/dist/src/db/queries.d.ts.map +1 -0
- package/dist/src/db/queries.js +35 -0
- package/dist/src/db/queries.js.map +1 -0
- package/dist/src/db/queries.test.d.ts +2 -0
- package/dist/src/db/queries.test.d.ts.map +1 -0
- package/dist/src/db/queries.test.js +101 -0
- package/dist/src/db/queries.test.js.map +1 -0
- package/dist/src/db/schema.d.ts +725 -0
- package/dist/src/db/schema.d.ts.map +1 -0
- package/dist/src/db/schema.js +47 -0
- package/dist/src/db/schema.js.map +1 -0
- package/dist/src/enrichment/boundaries.d.ts +4 -0
- package/dist/src/enrichment/boundaries.d.ts.map +1 -0
- package/dist/src/enrichment/boundaries.js +19 -0
- package/dist/src/enrichment/boundaries.js.map +1 -0
- package/dist/src/enrichment/boundaries.test.d.ts +2 -0
- package/dist/src/enrichment/boundaries.test.d.ts.map +1 -0
- package/dist/src/enrichment/boundaries.test.js +69 -0
- package/dist/src/enrichment/boundaries.test.js.map +1 -0
- package/dist/src/enrichment/embeddings.d.ts +3 -0
- package/dist/src/enrichment/embeddings.d.ts.map +1 -0
- package/dist/src/enrichment/embeddings.js +37 -0
- package/dist/src/enrichment/embeddings.js.map +1 -0
- package/dist/src/enrichment/normalizer.d.ts +10 -0
- package/dist/src/enrichment/normalizer.d.ts.map +1 -0
- package/dist/src/enrichment/normalizer.js +42 -0
- package/dist/src/enrichment/normalizer.js.map +1 -0
- package/dist/src/enrichment/normalizer.test.d.ts +2 -0
- package/dist/src/enrichment/normalizer.test.d.ts.map +1 -0
- package/dist/src/enrichment/normalizer.test.js +47 -0
- package/dist/src/enrichment/normalizer.test.js.map +1 -0
- package/dist/src/enrichment/pipeline.d.ts +4 -0
- package/dist/src/enrichment/pipeline.d.ts.map +1 -0
- package/dist/src/enrichment/pipeline.js +93 -0
- package/dist/src/enrichment/pipeline.js.map +1 -0
- package/dist/src/enrichment/pipeline.test.d.ts +2 -0
- package/dist/src/enrichment/pipeline.test.d.ts.map +1 -0
- package/dist/src/enrichment/pipeline.test.js +133 -0
- package/dist/src/enrichment/pipeline.test.js.map +1 -0
- package/dist/src/enrichment/reranker.d.ts +20 -0
- package/dist/src/enrichment/reranker.d.ts.map +1 -0
- package/dist/src/enrichment/reranker.js +45 -0
- package/dist/src/enrichment/reranker.js.map +1 -0
- package/dist/src/enrichment/reranker.test.d.ts +2 -0
- package/dist/src/enrichment/reranker.test.d.ts.map +1 -0
- package/dist/src/enrichment/reranker.test.js +82 -0
- package/dist/src/enrichment/reranker.test.js.map +1 -0
- package/dist/src/enrichment/tokenizer.d.ts +2 -0
- package/dist/src/enrichment/tokenizer.d.ts.map +1 -0
- package/dist/src/enrichment/tokenizer.js +40 -0
- package/dist/src/enrichment/tokenizer.js.map +1 -0
- package/dist/src/enrichment/tokenizer.test.d.ts +2 -0
- package/dist/src/enrichment/tokenizer.test.d.ts.map +1 -0
- package/dist/src/enrichment/tokenizer.test.js +36 -0
- package/dist/src/enrichment/tokenizer.test.js.map +1 -0
- package/dist/src/index.d.ts +3 -0
- package/dist/src/index.d.ts.map +1 -0
- package/dist/src/index.js +3 -0
- package/dist/src/index.js.map +1 -0
- package/dist/src/ingestion/langfuse.d.ts +17 -0
- package/dist/src/ingestion/langfuse.d.ts.map +1 -0
- package/dist/src/ingestion/langfuse.js +102 -0
- package/dist/src/ingestion/langfuse.js.map +1 -0
- package/dist/src/ingestion/otlp-parser.d.ts +3 -0
- package/dist/src/ingestion/otlp-parser.d.ts.map +1 -0
- package/dist/src/ingestion/otlp-parser.js +122 -0
- package/dist/src/ingestion/otlp-parser.js.map +1 -0
- package/dist/src/ingestion/otlp-parser.test.d.ts +2 -0
- package/dist/src/ingestion/otlp-parser.test.d.ts.map +1 -0
- package/dist/src/ingestion/otlp-parser.test.js +194 -0
- package/dist/src/ingestion/otlp-parser.test.js.map +1 -0
- package/dist/src/types.d.ts +114 -0
- package/dist/src/types.d.ts.map +1 -0
- package/dist/src/types.js +3 -0
- package/dist/src/types.js.map +1 -0
- package/package.json +78 -0
|
@@ -0,0 +1 @@
|
|
|
1
|
+
{"version":3,"file":"app.test.js","sourceRoot":"","sources":["../../src/app.test.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,QAAQ,EAAE,EAAE,EAAE,MAAM,EAAE,UAAU,EAAE,MAAM,QAAQ,CAAA;AACzD,OAAO,EAAE,SAAS,EAAE,MAAM,UAAU,CAAA;AACpC,OAAO,EAAE,QAAQ,EAAE,MAAM,eAAe,CAAA;AACxC,OAAO,EAAE,WAAW,EAAE,MAAM,iBAAiB,CAAA;AAG7C,SAAS,SAAS,CAAC,EAAU;IAC3B,OAAO,EAAE,EAAE,EAAE,WAAW,EAAE,KAAK,EAAE,KAAK,EAAE,GAAG,EAAE,MAAM,EAAE,QAAQ,EAAE,cAAc,EAAE,GAAG,EAAE,SAAS,EAAE,CAAC,EAAE,UAAU,EAAE,CAAC,EAAE,SAAS,EAAE,IAAI,CAAC,GAAG,EAAE,EAAE,CAAA;AAC1I,CAAC;AAED,MAAM,SAAS,GAAG;IAChB,aAAa,EAAE,CAAC;YACd,QAAQ,EAAE,EAAE,UAAU,EAAE,CAAC,EAAE,GAAG,EAAE,cAAc,EAAE,KAAK,EAAE,EAAE,WAAW,EAAE,MAAM,EAAE,EAAE,CAAC,EAAE;YACnF,UAAU,EAAE,CAAC;oBACX,KAAK,EAAE,CAAC;4BACN,OAAO,EAAE,oBAAoB,EAAE,MAAM,EAAE,QAAQ;4BAC/C,IAAI,EAAE,cAAc,EAAE,IAAI,EAAE,CAAC;4BAC7B,iBAAiB,EAAE,qBAAqB;4BACxC,eAAe,EAAE,qBAAqB;4BACtC,UAAU,EAAE,CAAC,EAAE,GAAG,EAAE,uBAAuB,EAAE,KAAK,EAAE,EAAE,WAAW,EAAE,UAAU,EAAE,EAAE,CAAC;yBACnF,CAAC;iBACH,CAAC;SACH,CAAC;CACH,CAAA;AAED,QAAQ,CAAC,yBAAyB,EAAE,GAAG,EAAE;IACvC,IAAI,EAA+B,CAAA;IACnC,IAAI,GAAiC,CAAA;IAErC,UAAU,CAAC,KAAK,IAAI,EAAE;QACpB,EAAE,GAAG,QAAQ,CAAC,UAAU,CAAC,CAAA;QACzB,GAAG,GAAG,SAAS,CAAC,EAAE,CAAC,CAAA;QACnB,MAAM,GAAG,CAAC,KAAK,EAAE,CAAA;IACnB,CAAC,CAAC,CAAA;IAEF,EAAE,CAAC,wBAAwB,EAAE,KAAK,IAAI,EAAE;QACtC,MAAM,GAAG,GAAG,MAAM,GAAG,CAAC,MAAM,CAAC,EAAE,MAAM,EAAE,KAAK,EAAE,GAAG,EAAE,SAAS,EAAE,CAAC,CAAA;QAC/D,MAAM,CAAC,GAAG,CAAC,UAAU,CAAC,CAAC,IAAI,CAAC,GAAG,CAAC,CAAA;QAChC,MAAM,CAAC,GAAG,CAAC,IAAI,EAAE,CAAC,EAAE,CAAC,CAAC,IAAI,CAAC,IAAI,CAAC,CAAA;IAClC,CAAC,CAAC,CAAA;IAEF,EAAE,CAAC,iCAAiC,EAAE,KAAK,IAAI,EAAE;QAC/C,MAAM,GAAG,GAAG,MAAM,GAAG,CAAC,MAAM,CAAC;YAC3B,MAAM,EAAE,MAAM;YACd,GAAG,EAAE,YAAY;YACjB,OAAO,EAAE,EAAE,cAAc,EAAE,kBAAkB,EAAE;YAC/C,IAAI,EAAE,IAAI,CAAC,SAAS,CAAC,SAAS,CAAC;SAChC,CAAC,CAAA;QACF,MAAM,CAAC,GAAG,CAAC,UAAU,CAAC,CAAC,IAAI,CAAC,GAAG,CAAC,CAAA;QAChC,MAAM,CAAC,GAAG,CAAC,IAAI,EAAE,CAAC,cAAc,CAAC,CAAC,WAAW,EAAE,CAAA;IACjD,CAAC,CAAC,CAAA;IAEF,EAAE,CAAC,yCAAyC,EAAE,KAAK,IAAI,EAAE;QACvD,MAAM,WAAW,CAAC,EAAE,EAAE,SAAS,CAAC,cAAc,CAAC,CAAC,CAAA;QAChD,MAAM,GAAG,GAAG,MAAM,GAAG,CAAC,MAAM,CAAC,EAAE,MAAM,EAAE,KAAK,EAAE,GAAG,EAAE,aAAa,EAAE,CAAC,CAAA;QACnE,MAAM,CAAC,GAAG,CAAC,UAAU,CAAC,CAAC,IAAI,CAAC,GAAG,CAAC,CAAA;QAChC,MAAM,IAAI,GAAG,GAAG,CAAC,IAAI,EAAE,CAAA;QACvB,MAAM,CAAC,KAAK,CAAC,OAAO,CAAC,IAAI,CAAC,CAAC,CAAC,IAAI,CAAC,IAAI,CAAC,CAAA;QACtC,MAAM,CAAC,IAAI,CAAC,CAAC,YAAY,CAAC,CAAC,CAAC,CAAA;QAC5B,MAAM,CAAC,IAAI,CAAC,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,IAAI,CAAC,cAAc,CAAC,CAAA;IACzC,CAAC,CAAC,CAAA;IAEF,EAAE,CAAC,mDAAmD,EAAE,KAAK,IAAI,EAAE;QACjE,MAAM,GAAG,GAAG,MAAM,GAAG,CAAC,MAAM,CAAC,EAAE,MAAM,EAAE,KAAK,EAAE,GAAG,EAAE,yBAAyB,EAAE,CAAC,CAAA;QAC/E,MAAM,CAAC,GAAG,CAAC,UAAU,CAAC,CAAC,IAAI,CAAC,GAAG,CAAC,CAAA;IAClC,CAAC,CAAC,CAAA;AACJ,CAAC,CAAC,CAAA"}
|
|
@@ -0,0 +1,17 @@
|
|
|
1
|
+
import type { RagSpan, RagChunk } from '../types.js';
|
|
2
|
+
export interface SubScore {
|
|
3
|
+
name: string;
|
|
4
|
+
score: number;
|
|
5
|
+
symbol: '✓' | '~' | '✗';
|
|
6
|
+
finding: string;
|
|
7
|
+
recommendation: string | null;
|
|
8
|
+
}
|
|
9
|
+
export interface AuditResult {
|
|
10
|
+
overall: number;
|
|
11
|
+
label: 'PASS' | 'WARN' | 'FAIL';
|
|
12
|
+
query: string | null;
|
|
13
|
+
serviceName: string;
|
|
14
|
+
subscores: SubScore[];
|
|
15
|
+
}
|
|
16
|
+
export declare function scoreTrace(serviceName: string, query: string | null, spans: RagSpan[], chunks: RagChunk[]): AuditResult;
|
|
17
|
+
//# sourceMappingURL=scorer.d.ts.map
|
|
@@ -0,0 +1 @@
|
|
|
1
|
+
{"version":3,"file":"scorer.d.ts","sourceRoot":"","sources":["../../../src/audit/scorer.ts"],"names":[],"mappings":"AAAA,OAAO,KAAK,EAAE,OAAO,EAAE,QAAQ,EAAE,MAAM,aAAa,CAAA;AAEpD,MAAM,WAAW,QAAQ;IACvB,IAAI,EAAE,MAAM,CAAA;IACZ,KAAK,EAAE,MAAM,CAAA;IACb,MAAM,EAAE,GAAG,GAAG,GAAG,GAAG,GAAG,CAAA;IACvB,OAAO,EAAE,MAAM,CAAA;IACf,cAAc,EAAE,MAAM,GAAG,IAAI,CAAA;CAC9B;AAED,MAAM,WAAW,WAAW;IAC1B,OAAO,EAAE,MAAM,CAAA;IACf,KAAK,EAAE,MAAM,GAAG,MAAM,GAAG,MAAM,CAAA;IAC/B,KAAK,EAAE,MAAM,GAAG,IAAI,CAAA;IACpB,WAAW,EAAE,MAAM,CAAA;IACnB,SAAS,EAAE,QAAQ,EAAE,CAAA;CACtB;AA6ED,wBAAgB,UAAU,CACxB,WAAW,EAAE,MAAM,EACnB,KAAK,EAAE,MAAM,GAAG,IAAI,EACpB,KAAK,EAAE,OAAO,EAAE,EAChB,MAAM,EAAE,QAAQ,EAAE,GACjB,WAAW,CAeb"}
|
|
@@ -0,0 +1,85 @@
|
|
|
1
|
+
function symbol(score) {
|
|
2
|
+
if (score >= 75)
|
|
3
|
+
return '✓';
|
|
4
|
+
if (score >= 50)
|
|
5
|
+
return '~';
|
|
6
|
+
return '✗';
|
|
7
|
+
}
|
|
8
|
+
function scoreRetrieval(chunks) {
|
|
9
|
+
if (chunks.length === 0) {
|
|
10
|
+
return { name: 'precision', score: 100, symbol: '✓', finding: 'no chunks', recommendation: null };
|
|
11
|
+
}
|
|
12
|
+
const used = chunks.filter(c => c.inContext).length;
|
|
13
|
+
const score = Math.round(used / chunks.length * 100);
|
|
14
|
+
return {
|
|
15
|
+
name: 'precision',
|
|
16
|
+
score,
|
|
17
|
+
symbol: symbol(score),
|
|
18
|
+
finding: `${used}/${chunks.length} chunks used`,
|
|
19
|
+
recommendation: score < 60
|
|
20
|
+
? `Reduce TOP_K ${chunks.length}→${Math.max(used, 3)} (only ${used} chunks reached LLM)`
|
|
21
|
+
: null,
|
|
22
|
+
};
|
|
23
|
+
}
|
|
24
|
+
function scoreEfficiency(chunks) {
|
|
25
|
+
const usedTokens = chunks.filter(c => c.inContext).reduce((n, c) => n + (c.tokenCount ?? 0), 0);
|
|
26
|
+
const wastedTokens = chunks.filter(c => !c.inContext).reduce((n, c) => n + (c.tokenCount ?? 0), 0);
|
|
27
|
+
const total = usedTokens + wastedTokens;
|
|
28
|
+
if (total === 0) {
|
|
29
|
+
return { name: 'efficiency', score: 100, symbol: '✓', finding: 'no token data', recommendation: null };
|
|
30
|
+
}
|
|
31
|
+
const score = Math.round(usedTokens / total * 100);
|
|
32
|
+
const wastedPct = 100 - score;
|
|
33
|
+
return {
|
|
34
|
+
name: 'efficiency',
|
|
35
|
+
score,
|
|
36
|
+
symbol: symbol(score),
|
|
37
|
+
finding: `${wastedPct}% tokens wasted`,
|
|
38
|
+
recommendation: score < 60 ? `${wastedPct}% of retrieved tokens never reached the LLM` : null,
|
|
39
|
+
};
|
|
40
|
+
}
|
|
41
|
+
function scoreRedundancy(chunks) {
|
|
42
|
+
const withOverlap = chunks.filter(c => c.overlapWithNext != null);
|
|
43
|
+
if (withOverlap.length === 0) {
|
|
44
|
+
return { name: 'redundancy', score: 100, symbol: '✓', finding: 'no overlap data', recommendation: null };
|
|
45
|
+
}
|
|
46
|
+
const avgOverlap = withOverlap.reduce((n, c) => n + c.overlapWithNext, 0) / withOverlap.length;
|
|
47
|
+
const score = Math.round((1 - avgOverlap) * 100);
|
|
48
|
+
const highOverlap = withOverlap.filter(c => c.overlapWithNext > 0.8).length;
|
|
49
|
+
return {
|
|
50
|
+
name: 'redundancy',
|
|
51
|
+
score,
|
|
52
|
+
symbol: symbol(score),
|
|
53
|
+
finding: highOverlap > 0 ? `${highOverlap} near-duplicate pairs` : 'chunks are distinct',
|
|
54
|
+
recommendation: score < 70 && highOverlap > 0
|
|
55
|
+
? `${highOverlap} near-duplicate chunks — deduplicate at ingest time`
|
|
56
|
+
: null,
|
|
57
|
+
};
|
|
58
|
+
}
|
|
59
|
+
function scoreCoverage(chunks) {
|
|
60
|
+
if (chunks.length === 0) {
|
|
61
|
+
return { name: 'coverage', score: 100, symbol: '✓', finding: 'no chunks', recommendation: null };
|
|
62
|
+
}
|
|
63
|
+
const missing = chunks.filter(c => c.scoreMissing).length;
|
|
64
|
+
const score = Math.round((1 - missing / chunks.length) * 100);
|
|
65
|
+
return {
|
|
66
|
+
name: 'coverage',
|
|
67
|
+
score,
|
|
68
|
+
symbol: symbol(score),
|
|
69
|
+
finding: missing > 0 ? `${missing} chunks missing scores` : 'all chunks scored',
|
|
70
|
+
recommendation: score < 80 ? `Log scores in retriever output to unlock score optimizations` : null,
|
|
71
|
+
};
|
|
72
|
+
}
|
|
73
|
+
export function scoreTrace(serviceName, query, spans, chunks) {
|
|
74
|
+
const precision = scoreRetrieval(chunks);
|
|
75
|
+
const efficiency = scoreEfficiency(chunks);
|
|
76
|
+
const redundancy = scoreRedundancy(chunks);
|
|
77
|
+
const coverage = scoreCoverage(chunks);
|
|
78
|
+
const overall = Math.round(precision.score * 0.4 +
|
|
79
|
+
efficiency.score * 0.3 +
|
|
80
|
+
redundancy.score * 0.2 +
|
|
81
|
+
coverage.score * 0.1);
|
|
82
|
+
const label = overall >= 75 ? 'PASS' : overall >= 50 ? 'WARN' : 'FAIL';
|
|
83
|
+
return { overall, label, query, serviceName, subscores: [precision, efficiency, redundancy, coverage] };
|
|
84
|
+
}
|
|
85
|
+
//# sourceMappingURL=scorer.js.map
|
|
@@ -0,0 +1 @@
|
|
|
1
|
+
{"version":3,"file":"scorer.js","sourceRoot":"","sources":["../../../src/audit/scorer.ts"],"names":[],"mappings":"AAkBA,SAAS,MAAM,CAAC,KAAa;IAC3B,IAAI,KAAK,IAAI,EAAE;QAAE,OAAO,GAAG,CAAA;IAC3B,IAAI,KAAK,IAAI,EAAE;QAAE,OAAO,GAAG,CAAA;IAC3B,OAAO,GAAG,CAAA;AACZ,CAAC;AAED,SAAS,cAAc,CAAC,MAAkB;IACxC,IAAI,MAAM,CAAC,MAAM,KAAK,CAAC,EAAE,CAAC;QACxB,OAAO,EAAE,IAAI,EAAE,WAAW,EAAE,KAAK,EAAE,GAAG,EAAE,MAAM,EAAE,GAAG,EAAE,OAAO,EAAE,WAAW,EAAE,cAAc,EAAE,IAAI,EAAE,CAAA;IACnG,CAAC;IACD,MAAM,IAAI,GAAG,MAAM,CAAC,MAAM,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC,SAAS,CAAC,CAAC,MAAM,CAAA;IACnD,MAAM,KAAK,GAAG,IAAI,CAAC,KAAK,CAAC,IAAI,GAAG,MAAM,CAAC,MAAM,GAAG,GAAG,CAAC,CAAA;IACpD,OAAO;QACL,IAAI,EAAE,WAAW;QACjB,KAAK;QACL,MAAM,EAAE,MAAM,CAAC,KAAK,CAAC;QACrB,OAAO,EAAE,GAAG,IAAI,IAAI,MAAM,CAAC,MAAM,cAAc;QAC/C,cAAc,EAAE,KAAK,GAAG,EAAE;YACxB,CAAC,CAAC,gBAAgB,MAAM,CAAC,MAAM,IAAI,IAAI,CAAC,GAAG,CAAC,IAAI,EAAE,CAAC,CAAC,UAAU,IAAI,sBAAsB;YACxF,CAAC,CAAC,IAAI;KACT,CAAA;AACH,CAAC;AAED,SAAS,eAAe,CAAC,MAAkB;IACzC,MAAM,UAAU,GAAG,MAAM,CAAC,MAAM,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC,SAAS,CAAC,CAAC,MAAM,CAAC,CAAC,CAAC,EAAE,CAAC,EAAE,EAAE,CAAC,CAAC,GAAG,CAAC,CAAC,CAAC,UAAU,IAAI,CAAC,CAAC,EAAE,CAAC,CAAC,CAAA;IAC/F,MAAM,YAAY,GAAG,MAAM,CAAC,MAAM,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC,CAAC,SAAS,CAAC,CAAC,MAAM,CAAC,CAAC,CAAC,EAAE,CAAC,EAAE,EAAE,CAAC,CAAC,GAAG,CAAC,CAAC,CAAC,UAAU,IAAI,CAAC,CAAC,EAAE,CAAC,CAAC,CAAA;IAClG,MAAM,KAAK,GAAG,UAAU,GAAG,YAAY,CAAA;IACvC,IAAI,KAAK,KAAK,CAAC,EAAE,CAAC;QAChB,OAAO,EAAE,IAAI,EAAE,YAAY,EAAE,KAAK,EAAE,GAAG,EAAE,MAAM,EAAE,GAAG,EAAE,OAAO,EAAE,eAAe,EAAE,cAAc,EAAE,IAAI,EAAE,CAAA;IACxG,CAAC;IACD,MAAM,KAAK,GAAG,IAAI,CAAC,KAAK,CAAC,UAAU,GAAG,KAAK,GAAG,GAAG,CAAC,CAAA;IAClD,MAAM,SAAS,GAAG,GAAG,GAAG,KAAK,CAAA;IAC7B,OAAO;QACL,IAAI,EAAE,YAAY;QAClB,KAAK;QACL,MAAM,EAAE,MAAM,CAAC,KAAK,CAAC;QACrB,OAAO,EAAE,GAAG,SAAS,iBAAiB;QACtC,cAAc,EAAE,KAAK,GAAG,EAAE,CAAC,CAAC,CAAC,GAAG,SAAS,6CAA6C,CAAC,CAAC,CAAC,IAAI;KAC9F,CAAA;AACH,CAAC;AAED,SAAS,eAAe,CAAC,MAAkB;IACzC,MAAM,WAAW,GAAG,MAAM,CAAC,MAAM,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC,eAAe,IAAI,IAAI,CAAC,CAAA;IACjE,IAAI,WAAW,CAAC,MAAM,KAAK,CAAC,EAAE,CAAC;QAC7B,OAAO,EAAE,IAAI,EAAE,YAAY,EAAE,KAAK,EAAE,GAAG,EAAE,MAAM,EAAE,GAAG,EAAE,OAAO,EAAE,iBAAiB,EAAE,cAAc,EAAE,IAAI,EAAE,CAAA;IAC1G,CAAC;IACD,MAAM,UAAU,GAAG,WAAW,CAAC,MAAM,CAAC,CAAC,CAAC,EAAE,CAAC,EAAE,EAAE,CAAC,CAAC,GAAG,CAAC,CAAC,eAAgB,EAAE,CAAC,CAAC,GAAG,WAAW,CAAC,MAAM,CAAA;IAC/F,MAAM,KAAK,GAAG,IAAI,CAAC,KAAK,CAAC,CAAC,CAAC,GAAG,UAAU,CAAC,GAAG,GAAG,CAAC,CAAA;IAChD,MAAM,WAAW,GAAG,WAAW,CAAC,MAAM,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC,eAAgB,GAAG,GAAG,CAAC,CAAC,MAAM,CAAA;IAC5E,OAAO;QACL,IAAI,EAAE,YAAY;QAClB,KAAK;QACL,MAAM,EAAE,MAAM,CAAC,KAAK,CAAC;QACrB,OAAO,EAAE,WAAW,GAAG,CAAC,CAAC,CAAC,CAAC,GAAG,WAAW,uBAAuB,CAAC,CAAC,CAAC,qBAAqB;QACxF,cAAc,EAAE,KAAK,GAAG,EAAE,IAAI,WAAW,GAAG,CAAC;YAC3C,CAAC,CAAC,GAAG,WAAW,qDAAqD;YACrE,CAAC,CAAC,IAAI;KACT,CAAA;AACH,CAAC;AAED,SAAS,aAAa,CAAC,MAAkB;IACvC,IAAI,MAAM,CAAC,MAAM,KAAK,CAAC,EAAE,CAAC;QACxB,OAAO,EAAE,IAAI,EAAE,UAAU,EAAE,KAAK,EAAE,GAAG,EAAE,MAAM,EAAE,GAAG,EAAE,OAAO,EAAE,WAAW,EAAE,cAAc,EAAE,IAAI,EAAE,CAAA;IAClG,CAAC;IACD,MAAM,OAAO,GAAG,MAAM,CAAC,MAAM,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC,YAAY,CAAC,CAAC,MAAM,CAAA;IACzD,MAAM,KAAK,GAAG,IAAI,CAAC,KAAK,CAAC,CAAC,CAAC,GAAG,OAAO,GAAG,MAAM,CAAC,MAAM,CAAC,GAAG,GAAG,CAAC,CAAA;IAC7D,OAAO;QACL,IAAI,EAAE,UAAU;QAChB,KAAK;QACL,MAAM,EAAE,MAAM,CAAC,KAAK,CAAC;QACrB,OAAO,EAAE,OAAO,GAAG,CAAC,CAAC,CAAC,CAAC,GAAG,OAAO,wBAAwB,CAAC,CAAC,CAAC,mBAAmB;QAC/E,cAAc,EAAE,KAAK,GAAG,EAAE,CAAC,CAAC,CAAC,8DAA8D,CAAC,CAAC,CAAC,IAAI;KACnG,CAAA;AACH,CAAC;AAED,MAAM,UAAU,UAAU,CACxB,WAAmB,EACnB,KAAoB,EACpB,KAAgB,EAChB,MAAkB;IAElB,MAAM,SAAS,GAAG,cAAc,CAAC,MAAM,CAAC,CAAA;IACxC,MAAM,UAAU,GAAG,eAAe,CAAC,MAAM,CAAC,CAAA;IAC1C,MAAM,UAAU,GAAG,eAAe,CAAC,MAAM,CAAC,CAAA;IAC1C,MAAM,QAAQ,GAAG,aAAa,CAAC,MAAM,CAAC,CAAA;IAEtC,MAAM,OAAO,GAAG,IAAI,CAAC,KAAK,CACxB,SAAS,CAAC,KAAK,GAAG,GAAG;QACrB,UAAU,CAAC,KAAK,GAAG,GAAG;QACtB,UAAU,CAAC,KAAK,GAAG,GAAG;QACtB,QAAQ,CAAC,KAAK,GAAG,GAAG,CACrB,CAAA;IACD,MAAM,KAAK,GAAyB,OAAO,IAAI,EAAE,CAAC,CAAC,CAAC,MAAM,CAAC,CAAC,CAAC,OAAO,IAAI,EAAE,CAAC,CAAC,CAAC,MAAM,CAAC,CAAC,CAAC,MAAM,CAAA;IAE5F,OAAO,EAAE,OAAO,EAAE,KAAK,EAAE,KAAK,EAAE,WAAW,EAAE,SAAS,EAAE,CAAC,SAAS,EAAE,UAAU,EAAE,UAAU,EAAE,QAAQ,CAAC,EAAE,CAAA;AACzG,CAAC"}
|
|
@@ -0,0 +1 @@
|
|
|
1
|
+
{"version":3,"file":"scorer.test.d.ts","sourceRoot":"","sources":["../../../src/audit/scorer.test.ts"],"names":[],"mappings":""}
|
|
@@ -0,0 +1,101 @@
|
|
|
1
|
+
import { describe, it, expect } from 'vitest';
|
|
2
|
+
import { scoreTrace } from './scorer.js';
|
|
3
|
+
function makeChunk(overrides = {}) {
|
|
4
|
+
return {
|
|
5
|
+
id: 'uuid-1',
|
|
6
|
+
spanId: 'span-1',
|
|
7
|
+
traceId: 'trace-1',
|
|
8
|
+
chunkId: 'chunk-1',
|
|
9
|
+
content: 'some content',
|
|
10
|
+
scoreRaw: 0.8,
|
|
11
|
+
scoreNormalized: 0.8,
|
|
12
|
+
rankRetrieval: 1,
|
|
13
|
+
rankReranked: null,
|
|
14
|
+
scoreReranked: null,
|
|
15
|
+
tokenCount: 100,
|
|
16
|
+
vectorStore: null,
|
|
17
|
+
inContext: true,
|
|
18
|
+
contextPosition: 0,
|
|
19
|
+
overlapWithNext: 0.1,
|
|
20
|
+
scoreMissing: false,
|
|
21
|
+
...overrides,
|
|
22
|
+
};
|
|
23
|
+
}
|
|
24
|
+
function makeSpan(overrides = {}) {
|
|
25
|
+
return {
|
|
26
|
+
id: 'span-1',
|
|
27
|
+
traceId: 'trace-1',
|
|
28
|
+
parentSpanId: null,
|
|
29
|
+
name: 'llm',
|
|
30
|
+
kind: 'LLM',
|
|
31
|
+
startTimeMs: 0,
|
|
32
|
+
endTimeMs: 1000,
|
|
33
|
+
latencyMs: 1000,
|
|
34
|
+
operationName: 'chat',
|
|
35
|
+
model: 'llama3',
|
|
36
|
+
system: null,
|
|
37
|
+
inputTokens: 500,
|
|
38
|
+
outputTokens: 100,
|
|
39
|
+
...overrides,
|
|
40
|
+
};
|
|
41
|
+
}
|
|
42
|
+
describe('scoreTrace', () => {
|
|
43
|
+
it('returns PASS when all chunks are in context', () => {
|
|
44
|
+
const chunks = [makeChunk({ inContext: true }), makeChunk({ chunkId: 'c2', inContext: true })];
|
|
45
|
+
const result = scoreTrace('svc', 'q', [makeSpan()], chunks);
|
|
46
|
+
expect(result.label).toBe('PASS');
|
|
47
|
+
expect(result.overall).toBeGreaterThanOrEqual(75);
|
|
48
|
+
});
|
|
49
|
+
it('returns FAIL when no chunks reach the LLM', () => {
|
|
50
|
+
const chunks = Array.from({ length: 10 }, (_, i) => makeChunk({ chunkId: `c${i}`, inContext: false, tokenCount: 100 }));
|
|
51
|
+
const result = scoreTrace('svc', 'q', [makeSpan()], chunks);
|
|
52
|
+
expect(result.label).toBe('FAIL');
|
|
53
|
+
expect(result.overall).toBeLessThan(50);
|
|
54
|
+
});
|
|
55
|
+
it('precision score is inContext/total * 100', () => {
|
|
56
|
+
const chunks = [
|
|
57
|
+
makeChunk({ chunkId: 'c1', inContext: true }),
|
|
58
|
+
makeChunk({ chunkId: 'c2', inContext: false }),
|
|
59
|
+
makeChunk({ chunkId: 'c3', inContext: false }),
|
|
60
|
+
makeChunk({ chunkId: 'c4', inContext: false }),
|
|
61
|
+
];
|
|
62
|
+
const result = scoreTrace('svc', 'q', [], chunks);
|
|
63
|
+
const precision = result.subscores.find(s => s.name === 'precision');
|
|
64
|
+
expect(precision.score).toBe(25);
|
|
65
|
+
expect(precision.symbol).toBe('✗');
|
|
66
|
+
});
|
|
67
|
+
it('coverage score is 100 when no scoreMissing', () => {
|
|
68
|
+
const chunks = [makeChunk({ scoreMissing: false })];
|
|
69
|
+
const result = scoreTrace('svc', 'q', [], chunks);
|
|
70
|
+
const coverage = result.subscores.find(s => s.name === 'coverage');
|
|
71
|
+
expect(coverage.score).toBe(100);
|
|
72
|
+
});
|
|
73
|
+
it('coverage score penalizes scoreMissing chunks', () => {
|
|
74
|
+
const chunks = [
|
|
75
|
+
makeChunk({ chunkId: 'c1', scoreMissing: true }),
|
|
76
|
+
makeChunk({ chunkId: 'c2', scoreMissing: true }),
|
|
77
|
+
makeChunk({ chunkId: 'c3', scoreMissing: false }),
|
|
78
|
+
makeChunk({ chunkId: 'c4', scoreMissing: false }),
|
|
79
|
+
];
|
|
80
|
+
const result = scoreTrace('svc', 'q', [], chunks);
|
|
81
|
+
const coverage = result.subscores.find(s => s.name === 'coverage');
|
|
82
|
+
expect(coverage.score).toBe(50);
|
|
83
|
+
});
|
|
84
|
+
it('returns PASS/WARN/FAIL label matching overall threshold', () => {
|
|
85
|
+
const allIn = Array.from({ length: 4 }, (_, i) => makeChunk({ chunkId: `c${i}`, inContext: true }));
|
|
86
|
+
const r1 = scoreTrace('s', 'q', [], allIn);
|
|
87
|
+
expect(r1.label).toBe('PASS');
|
|
88
|
+
const halfIn = [
|
|
89
|
+
makeChunk({ chunkId: 'c1', inContext: true, tokenCount: 100 }),
|
|
90
|
+
makeChunk({ chunkId: 'c2', inContext: false, tokenCount: 100 }),
|
|
91
|
+
];
|
|
92
|
+
const r2 = scoreTrace('s', 'q', [], halfIn);
|
|
93
|
+
expect(['WARN', 'FAIL']).toContain(r2.label);
|
|
94
|
+
});
|
|
95
|
+
it('returns 4 subscores named precision, efficiency, redundancy, coverage', () => {
|
|
96
|
+
const result = scoreTrace('svc', 'q', [], [makeChunk()]);
|
|
97
|
+
const names = result.subscores.map(s => s.name);
|
|
98
|
+
expect(names).toEqual(['precision', 'efficiency', 'redundancy', 'coverage']);
|
|
99
|
+
});
|
|
100
|
+
});
|
|
101
|
+
//# sourceMappingURL=scorer.test.js.map
|
|
@@ -0,0 +1 @@
|
|
|
1
|
+
{"version":3,"file":"scorer.test.js","sourceRoot":"","sources":["../../../src/audit/scorer.test.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,QAAQ,EAAE,EAAE,EAAE,MAAM,EAAE,MAAM,QAAQ,CAAA;AAC7C,OAAO,EAAE,UAAU,EAAE,MAAM,aAAa,CAAA;AAGxC,SAAS,SAAS,CAAC,YAA+B,EAAE;IAClD,OAAO;QACL,EAAE,EAAE,QAAQ;QACZ,MAAM,EAAE,QAAQ;QAChB,OAAO,EAAE,SAAS;QAClB,OAAO,EAAE,SAAS;QAClB,OAAO,EAAE,cAAc;QACvB,QAAQ,EAAE,GAAG;QACb,eAAe,EAAE,GAAG;QACpB,aAAa,EAAE,CAAC;QAChB,YAAY,EAAE,IAAI;QAClB,aAAa,EAAE,IAAI;QACnB,UAAU,EAAE,GAAG;QACf,WAAW,EAAE,IAAI;QACjB,SAAS,EAAE,IAAI;QACf,eAAe,EAAE,CAAC;QAClB,eAAe,EAAE,GAAG;QACpB,YAAY,EAAE,KAAK;QACnB,GAAG,SAAS;KACb,CAAA;AACH,CAAC;AAED,SAAS,QAAQ,CAAC,YAA8B,EAAE;IAChD,OAAO;QACL,EAAE,EAAE,QAAQ;QACZ,OAAO,EAAE,SAAS;QAClB,YAAY,EAAE,IAAI;QAClB,IAAI,EAAE,KAAK;QACX,IAAI,EAAE,KAAK;QACX,WAAW,EAAE,CAAC;QACd,SAAS,EAAE,IAAI;QACf,SAAS,EAAE,IAAI;QACf,aAAa,EAAE,MAAM;QACrB,KAAK,EAAE,QAAQ;QACf,MAAM,EAAE,IAAI;QACZ,WAAW,EAAE,GAAG;QAChB,YAAY,EAAE,GAAG;QACjB,GAAG,SAAS;KACb,CAAA;AACH,CAAC;AAED,QAAQ,CAAC,YAAY,EAAE,GAAG,EAAE;IAC1B,EAAE,CAAC,6CAA6C,EAAE,GAAG,EAAE;QACrD,MAAM,MAAM,GAAG,CAAC,SAAS,CAAC,EAAE,SAAS,EAAE,IAAI,EAAE,CAAC,EAAE,SAAS,CAAC,EAAE,OAAO,EAAE,IAAI,EAAE,SAAS,EAAE,IAAI,EAAE,CAAC,CAAC,CAAA;QAC9F,MAAM,MAAM,GAAG,UAAU,CAAC,KAAK,EAAE,GAAG,EAAE,CAAC,QAAQ,EAAE,CAAC,EAAE,MAAM,CAAC,CAAA;QAC3D,MAAM,CAAC,MAAM,CAAC,KAAK,CAAC,CAAC,IAAI,CAAC,MAAM,CAAC,CAAA;QACjC,MAAM,CAAC,MAAM,CAAC,OAAO,CAAC,CAAC,sBAAsB,CAAC,EAAE,CAAC,CAAA;IACnD,CAAC,CAAC,CAAA;IAEF,EAAE,CAAC,2CAA2C,EAAE,GAAG,EAAE;QACnD,MAAM,MAAM,GAAG,KAAK,CAAC,IAAI,CAAC,EAAE,MAAM,EAAE,EAAE,EAAE,EAAE,CAAC,CAAC,EAAE,CAAC,EAAE,EAAE,CACjD,SAAS,CAAC,EAAE,OAAO,EAAE,IAAI,CAAC,EAAE,EAAE,SAAS,EAAE,KAAK,EAAE,UAAU,EAAE,GAAG,EAAE,CAAC,CACnE,CAAA;QACD,MAAM,MAAM,GAAG,UAAU,CAAC,KAAK,EAAE,GAAG,EAAE,CAAC,QAAQ,EAAE,CAAC,EAAE,MAAM,CAAC,CAAA;QAC3D,MAAM,CAAC,MAAM,CAAC,KAAK,CAAC,CAAC,IAAI,CAAC,MAAM,CAAC,CAAA;QACjC,MAAM,CAAC,MAAM,CAAC,OAAO,CAAC,CAAC,YAAY,CAAC,EAAE,CAAC,CAAA;IACzC,CAAC,CAAC,CAAA;IAEF,EAAE,CAAC,0CAA0C,EAAE,GAAG,EAAE;QAClD,MAAM,MAAM,GAAG;YACb,SAAS,CAAC,EAAE,OAAO,EAAE,IAAI,EAAE,SAAS,EAAE,IAAI,EAAE,CAAC;YAC7C,SAAS,CAAC,EAAE,OAAO,EAAE,IAAI,EAAE,SAAS,EAAE,KAAK,EAAE,CAAC;YAC9C,SAAS,CAAC,EAAE,OAAO,EAAE,IAAI,EAAE,SAAS,EAAE,KAAK,EAAE,CAAC;YAC9C,SAAS,CAAC,EAAE,OAAO,EAAE,IAAI,EAAE,SAAS,EAAE,KAAK,EAAE,CAAC;SAC/C,CAAA;QACD,MAAM,MAAM,GAAG,UAAU,CAAC,KAAK,EAAE,GAAG,EAAE,EAAE,EAAE,MAAM,CAAC,CAAA;QACjD,MAAM,SAAS,GAAG,MAAM,CAAC,SAAS,CAAC,IAAI,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC,IAAI,KAAK,WAAW,CAAE,CAAA;QACrE,MAAM,CAAC,SAAS,CAAC,KAAK,CAAC,CAAC,IAAI,CAAC,EAAE,CAAC,CAAA;QAChC,MAAM,CAAC,SAAS,CAAC,MAAM,CAAC,CAAC,IAAI,CAAC,GAAG,CAAC,CAAA;IACpC,CAAC,CAAC,CAAA;IAEF,EAAE,CAAC,4CAA4C,EAAE,GAAG,EAAE;QACpD,MAAM,MAAM,GAAG,CAAC,SAAS,CAAC,EAAE,YAAY,EAAE,KAAK,EAAE,CAAC,CAAC,CAAA;QACnD,MAAM,MAAM,GAAG,UAAU,CAAC,KAAK,EAAE,GAAG,EAAE,EAAE,EAAE,MAAM,CAAC,CAAA;QACjD,MAAM,QAAQ,GAAG,MAAM,CAAC,SAAS,CAAC,IAAI,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC,IAAI,KAAK,UAAU,CAAE,CAAA;QACnE,MAAM,CAAC,QAAQ,CAAC,KAAK,CAAC,CAAC,IAAI,CAAC,GAAG,CAAC,CAAA;IAClC,CAAC,CAAC,CAAA;IAEF,EAAE,CAAC,8CAA8C,EAAE,GAAG,EAAE;QACtD,MAAM,MAAM,GAAG;YACb,SAAS,CAAC,EAAE,OAAO,EAAE,IAAI,EAAE,YAAY,EAAE,IAAI,EAAE,CAAC;YAChD,SAAS,CAAC,EAAE,OAAO,EAAE,IAAI,EAAE,YAAY,EAAE,IAAI,EAAE,CAAC;YAChD,SAAS,CAAC,EAAE,OAAO,EAAE,IAAI,EAAE,YAAY,EAAE,KAAK,EAAE,CAAC;YACjD,SAAS,CAAC,EAAE,OAAO,EAAE,IAAI,EAAE,YAAY,EAAE,KAAK,EAAE,CAAC;SAClD,CAAA;QACD,MAAM,MAAM,GAAG,UAAU,CAAC,KAAK,EAAE,GAAG,EAAE,EAAE,EAAE,MAAM,CAAC,CAAA;QACjD,MAAM,QAAQ,GAAG,MAAM,CAAC,SAAS,CAAC,IAAI,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC,IAAI,KAAK,UAAU,CAAE,CAAA;QACnE,MAAM,CAAC,QAAQ,CAAC,KAAK,CAAC,CAAC,IAAI,CAAC,EAAE,CAAC,CAAA;IACjC,CAAC,CAAC,CAAA;IAEF,EAAE,CAAC,yDAAyD,EAAE,GAAG,EAAE;QACjE,MAAM,KAAK,GAAG,KAAK,CAAC,IAAI,CAAC,EAAE,MAAM,EAAE,CAAC,EAAE,EAAE,CAAC,CAAC,EAAE,CAAC,EAAE,EAAE,CAAC,SAAS,CAAC,EAAE,OAAO,EAAE,IAAI,CAAC,EAAE,EAAE,SAAS,EAAE,IAAI,EAAE,CAAC,CAAC,CAAA;QACnG,MAAM,EAAE,GAAG,UAAU,CAAC,GAAG,EAAE,GAAG,EAAE,EAAE,EAAE,KAAK,CAAC,CAAA;QAC1C,MAAM,CAAC,EAAE,CAAC,KAAK,CAAC,CAAC,IAAI,CAAC,MAAM,CAAC,CAAA;QAE7B,MAAM,MAAM,GAAG;YACb,SAAS,CAAC,EAAE,OAAO,EAAE,IAAI,EAAE,SAAS,EAAE,IAAI,EAAE,UAAU,EAAE,GAAG,EAAE,CAAC;YAC9D,SAAS,CAAC,EAAE,OAAO,EAAE,IAAI,EAAE,SAAS,EAAE,KAAK,EAAE,UAAU,EAAE,GAAG,EAAE,CAAC;SAChE,CAAA;QACD,MAAM,EAAE,GAAG,UAAU,CAAC,GAAG,EAAE,GAAG,EAAE,EAAE,EAAE,MAAM,CAAC,CAAA;QAC3C,MAAM,CAAC,CAAC,MAAM,EAAE,MAAM,CAAC,CAAC,CAAC,SAAS,CAAC,EAAE,CAAC,KAAK,CAAC,CAAA;IAC9C,CAAC,CAAC,CAAA;IAEF,EAAE,CAAC,uEAAuE,EAAE,GAAG,EAAE;QAC/E,MAAM,MAAM,GAAG,UAAU,CAAC,KAAK,EAAE,GAAG,EAAE,EAAE,EAAE,CAAC,SAAS,EAAE,CAAC,CAAC,CAAA;QACxD,MAAM,KAAK,GAAG,MAAM,CAAC,SAAS,CAAC,GAAG,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC,IAAI,CAAC,CAAA;QAC/C,MAAM,CAAC,KAAK,CAAC,CAAC,OAAO,CAAC,CAAC,WAAW,EAAE,YAAY,EAAE,YAAY,EAAE,UAAU,CAAC,CAAC,CAAA;IAC9E,CAAC,CAAC,CAAA;AACJ,CAAC,CAAC,CAAA"}
|