@opensearch-project/agent-health 0.5.2 → 0.6.0
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/README.md +1 -0
- package/cli/dist/index.js +1592 -620
- package/dist/assets/index-BfxtxmKc.css +1 -0
- package/dist/assets/index-CrjAfDHu.js +243 -0
- package/dist/index.html +2 -2
- package/docs/CLI.md +45 -2
- package/docs/CONFIGURATION.md +1 -1
- package/docs/INSTRUMENT_WITH_OTEL.md +7 -0
- package/docs/SDK.md +52 -3
- package/docs/STORAGE_INDEX_FIELD_LIMITS.md +216 -0
- package/docs/skills/AGENT_HEALTH.md +55 -1
- package/examples/eval-files/demo.eval.js +1 -1
- package/examples/eval-files/ops-rca-classification.eval.js +71 -0
- package/examples/eval-files/ops-rca-evaluator.json +15 -0
- package/examples/eval-files/sdk-demo.eval.js +72 -0
- package/examples/eval-files/sdk-describe-demo.eval.js +50 -0
- package/examples/eval-files/sdk-hooks-demo.eval.js +1 -1
- package/lib/dist/lib/bedrockCompat.d.ts +27 -0
- package/lib/dist/lib/bedrockCompat.d.ts.map +1 -0
- package/lib/dist/lib/bedrockCompat.js +83 -0
- package/lib/dist/lib/bedrockCompat.js.map +1 -0
- package/lib/dist/lib/benchmarkImage.d.ts +52 -0
- package/lib/dist/lib/benchmarkImage.d.ts.map +1 -0
- package/lib/dist/lib/benchmarkImage.js +113 -0
- package/lib/dist/lib/benchmarkImage.js.map +1 -0
- package/lib/dist/lib/benchmarkVersionUtils.d.ts +13 -0
- package/lib/dist/lib/benchmarkVersionUtils.d.ts.map +1 -1
- package/lib/dist/lib/benchmarkVersionUtils.js +21 -0
- package/lib/dist/lib/benchmarkVersionUtils.js.map +1 -1
- package/lib/dist/lib/chunkedFetch.d.ts +18 -0
- package/lib/dist/lib/chunkedFetch.d.ts.map +1 -0
- package/lib/dist/lib/chunkedFetch.js +40 -0
- package/lib/dist/lib/chunkedFetch.js.map +1 -0
- package/lib/dist/lib/comparisonInsights.d.ts +104 -0
- package/lib/dist/lib/comparisonInsights.d.ts.map +1 -0
- package/lib/dist/lib/comparisonInsights.js +212 -0
- package/lib/dist/lib/comparisonInsights.js.map +1 -0
- package/lib/dist/lib/config/loader.d.ts.map +1 -1
- package/lib/dist/lib/config/loader.js +5 -0
- package/lib/dist/lib/config/loader.js.map +1 -1
- package/lib/dist/lib/config/types.d.ts +14 -0
- package/lib/dist/lib/config/types.d.ts.map +1 -1
- package/lib/dist/lib/constants.d.ts +11 -0
- package/lib/dist/lib/constants.d.ts.map +1 -1
- package/lib/dist/lib/constants.js +10 -1
- package/lib/dist/lib/constants.js.map +1 -1
- package/lib/dist/lib/contextFormat.d.ts +26 -0
- package/lib/dist/lib/contextFormat.d.ts.map +1 -0
- package/lib/dist/lib/contextFormat.js +28 -0
- package/lib/dist/lib/contextFormat.js.map +1 -0
- package/lib/dist/lib/envCompat.d.ts.map +1 -1
- package/lib/dist/lib/envCompat.js +14 -5
- package/lib/dist/lib/envCompat.js.map +1 -1
- package/lib/dist/lib/evaluationRerun.d.ts +63 -0
- package/lib/dist/lib/evaluationRerun.d.ts.map +1 -0
- package/lib/dist/lib/evaluationRerun.js +85 -0
- package/lib/dist/lib/evaluationRerun.js.map +1 -0
- package/lib/dist/lib/matchers/traces.d.ts +17 -2
- package/lib/dist/lib/matchers/traces.d.ts.map +1 -1
- package/lib/dist/lib/matchers/traces.js +136 -16
- package/lib/dist/lib/matchers/traces.js.map +1 -1
- package/lib/dist/lib/matchers/tracesPricing.d.ts +38 -0
- package/lib/dist/lib/matchers/tracesPricing.d.ts.map +1 -0
- package/lib/dist/lib/matchers/tracesPricing.js +64 -0
- package/lib/dist/lib/matchers/tracesPricing.js.map +1 -0
- package/lib/dist/lib/runStats.d.ts +24 -0
- package/lib/dist/lib/runStats.d.ts.map +1 -1
- package/lib/dist/lib/runStats.js +32 -0
- package/lib/dist/lib/runStats.js.map +1 -1
- package/lib/dist/lib/testCases/loader.d.ts +11 -1
- package/lib/dist/lib/testCases/loader.d.ts.map +1 -1
- package/lib/dist/lib/testCases/loader.js +35 -5
- package/lib/dist/lib/testCases/loader.js.map +1 -1
- package/lib/dist/lib/utils.d.ts +15 -0
- package/lib/dist/lib/utils.d.ts.map +1 -1
- package/lib/dist/lib/utils.js +22 -0
- package/lib/dist/lib/utils.js.map +1 -1
- package/lib/dist/services/connectors/claude-code/ClaudeCodeConnector.d.ts +10 -0
- package/lib/dist/services/connectors/claude-code/ClaudeCodeConnector.d.ts.map +1 -1
- package/lib/dist/services/connectors/claude-code/ClaudeCodeConnector.js +17 -3
- package/lib/dist/services/connectors/claude-code/ClaudeCodeConnector.js.map +1 -1
- package/lib/dist/services/connectors/subprocess/SubprocessConnector.d.ts.map +1 -1
- package/lib/dist/services/connectors/subprocess/SubprocessConnector.js +8 -0
- package/lib/dist/services/connectors/subprocess/SubprocessConnector.js.map +1 -1
- package/lib/dist/services/evaluation/bedrockJudge.d.ts +7 -0
- package/lib/dist/services/evaluation/bedrockJudge.d.ts.map +1 -1
- package/lib/dist/services/evaluation/bedrockJudge.js +2 -0
- package/lib/dist/services/evaluation/bedrockJudge.js.map +1 -1
- package/lib/dist/services/storage/asyncBenchmarkStorage.d.ts +1 -1
- package/lib/dist/services/storage/asyncBenchmarkStorage.d.ts.map +1 -1
- package/lib/dist/services/storage/asyncBenchmarkStorage.js +14 -3
- package/lib/dist/services/storage/asyncBenchmarkStorage.js.map +1 -1
- package/lib/dist/services/storage/asyncRunStorage.d.ts +9 -1
- package/lib/dist/services/storage/asyncRunStorage.d.ts.map +1 -1
- package/lib/dist/services/storage/asyncRunStorage.js +77 -1
- package/lib/dist/services/storage/asyncRunStorage.js.map +1 -1
- package/lib/dist/services/storage/asyncTestCaseStorage.d.ts +18 -4
- package/lib/dist/services/storage/asyncTestCaseStorage.d.ts.map +1 -1
- package/lib/dist/services/storage/asyncTestCaseStorage.js +20 -4
- package/lib/dist/services/storage/asyncTestCaseStorage.js.map +1 -1
- package/lib/dist/services/storage/opensearchClient.d.ts +35 -12
- package/lib/dist/services/storage/opensearchClient.d.ts.map +1 -1
- package/lib/dist/services/storage/opensearchClient.js +12 -5
- package/lib/dist/services/storage/opensearchClient.js.map +1 -1
- package/lib/dist/services/traces/browserRecovery.d.ts +12 -1
- package/lib/dist/services/traces/browserRecovery.d.ts.map +1 -1
- package/lib/dist/services/traces/browserRecovery.js +32 -5
- package/lib/dist/services/traces/browserRecovery.js.map +1 -1
- package/lib/dist/services/traces/messageExtraction.d.ts.map +1 -1
- package/lib/dist/services/traces/messageExtraction.js +95 -33
- package/lib/dist/services/traces/messageExtraction.js.map +1 -1
- package/lib/dist/services/traces/spansToTrajectory.d.ts.map +1 -1
- package/lib/dist/services/traces/spansToTrajectory.js +47 -9
- package/lib/dist/services/traces/spansToTrajectory.js.map +1 -1
- package/lib/dist/services/traces/tracePoller.d.ts +27 -3
- package/lib/dist/services/traces/tracePoller.d.ts.map +1 -1
- package/lib/dist/services/traces/tracePoller.js +199 -33
- package/lib/dist/services/traces/tracePoller.js.map +1 -1
- package/lib/dist/types/index.d.ts +55 -1
- package/lib/dist/types/index.d.ts.map +1 -1
- package/lib/dist/types/index.js.map +1 -1
- package/package.json +3 -1
- package/server/dist/app.js +2238 -777
- package/server/dist/index.js +2238 -777
- package/dist/assets/index-CCQRDlO0.js +0 -243
- package/dist/assets/index-CNHQVbcj.css +0 -1
|
@@ -0,0 +1,85 @@
|
|
|
1
|
+
/*
|
|
2
|
+
* Copyright OpenSearch Contributors
|
|
3
|
+
* SPDX-License-Identifier: Apache-2.0
|
|
4
|
+
*/
|
|
5
|
+
const DEFAULT_RUN_NAME = 'Evaluation Run';
|
|
6
|
+
/**
|
|
7
|
+
* Matches a trailing "(re-run)" or "(re-run N)" suffix, case-insensitively,
|
|
8
|
+
* capturing the base name and the optional numeric suffix.
|
|
9
|
+
*/
|
|
10
|
+
const RERUN_SUFFIX_RE = /^(.*?)\s*\(re-run(?:\s+(\d+))?\)$/i;
|
|
11
|
+
/**
|
|
12
|
+
* Compute the name for a new run being created as a re-run of `sourceName`.
|
|
13
|
+
*
|
|
14
|
+
* - "My Run" -> "My Run (re-run)"
|
|
15
|
+
* - "My Run (re-run)" -> "My Run (re-run 2)"
|
|
16
|
+
* - "My Run (re-run 2)" -> "My Run (re-run 3)"
|
|
17
|
+
* - undefined/empty -> "Evaluation Run (re-run)"
|
|
18
|
+
*/
|
|
19
|
+
export function computeRerunName(sourceName) {
|
|
20
|
+
const trimmed = (sourceName || '').trim();
|
|
21
|
+
const base = trimmed || DEFAULT_RUN_NAME;
|
|
22
|
+
const match = base.match(RERUN_SUFFIX_RE);
|
|
23
|
+
if (!match) {
|
|
24
|
+
return `${base} (re-run)`;
|
|
25
|
+
}
|
|
26
|
+
const [, rawBaseName, suffixNumber] = match;
|
|
27
|
+
// A source named literally "(re-run)" / "(re-run 2)" (no prefix) captures
|
|
28
|
+
// an empty baseName — fall back to the default rather than emitting a
|
|
29
|
+
// leading-space name like " (re-run 2)".
|
|
30
|
+
const baseName = rawBaseName.trim() || DEFAULT_RUN_NAME;
|
|
31
|
+
const nextNumber = suffixNumber ? parseInt(suffixNumber, 10) + 1 : 2;
|
|
32
|
+
return `${baseName} (re-run ${nextNumber})`;
|
|
33
|
+
}
|
|
34
|
+
/**
|
|
35
|
+
* Duplicate a source run's execution config for a re-run, applying explicit,
|
|
36
|
+
* reported defaults for fields missing on legacy run documents. Pure
|
|
37
|
+
* function — no storage access (benchmark-existence checks live in
|
|
38
|
+
* services/evaluationRerun.ts, which needs the storage module).
|
|
39
|
+
*
|
|
40
|
+
* Returns `{ error }` only when the source run is missing a field with no
|
|
41
|
+
* safe default (nothing to run, or no agent to run it against).
|
|
42
|
+
*/
|
|
43
|
+
export function buildRerunConfig(sourceRun) {
|
|
44
|
+
const defaultsApplied = [];
|
|
45
|
+
if (!sourceRun.agentKey) {
|
|
46
|
+
return { error: 'Source run is missing agentKey; cannot determine which agent to re-run.' };
|
|
47
|
+
}
|
|
48
|
+
let sources = sourceRun.sources;
|
|
49
|
+
if (!sources || sources.length === 0) {
|
|
50
|
+
const snapshotIds = (sourceRun.testCaseSnapshots || []).map(s => s.id);
|
|
51
|
+
if (snapshotIds.length === 0) {
|
|
52
|
+
return {
|
|
53
|
+
error: 'Source run has no test cases to re-run (missing both sources and testCaseSnapshots).',
|
|
54
|
+
};
|
|
55
|
+
}
|
|
56
|
+
sources = [{ type: 'test-case-ids', ids: snapshotIds }];
|
|
57
|
+
defaultsApplied.push(`sources -> derived from testCaseSnapshots (${snapshotIds.length} test case id(s); legacy run had no sources recorded)`);
|
|
58
|
+
}
|
|
59
|
+
let concurrency = sourceRun.concurrency;
|
|
60
|
+
if (concurrency == null) {
|
|
61
|
+
concurrency = 1;
|
|
62
|
+
defaultsApplied.push('concurrency -> 1 (default; not set on source run)');
|
|
63
|
+
}
|
|
64
|
+
const config = {
|
|
65
|
+
sources,
|
|
66
|
+
agentKey: sourceRun.agentKey,
|
|
67
|
+
modelId: sourceRun.modelId || '',
|
|
68
|
+
judgeModelId: sourceRun.judgeModelId,
|
|
69
|
+
evaluatorId: sourceRun.evaluatorId,
|
|
70
|
+
headers: sourceRun.headers,
|
|
71
|
+
concurrency,
|
|
72
|
+
agentEndpoint: sourceRun.agentEndpoint,
|
|
73
|
+
description: sourceRun.description,
|
|
74
|
+
benchmarkId: sourceRun.benchmarkId,
|
|
75
|
+
benchmarkVersion: sourceRun.benchmarkVersion,
|
|
76
|
+
};
|
|
77
|
+
// NOTE: evaluatorId/judgeModelId/agentEndpoint/headers/description are
|
|
78
|
+
// legitimately optional on ANY run (not just legacy ones), so leaving them
|
|
79
|
+
// `undefined` is normal behavior, not a legacy-config gap — they are NOT
|
|
80
|
+
// reported in `defaultsApplied` (that list is reserved for cases where we
|
|
81
|
+
// had to fabricate a value because the source run's config was
|
|
82
|
+
// incomplete).
|
|
83
|
+
return { config, defaultsApplied };
|
|
84
|
+
}
|
|
85
|
+
//# sourceMappingURL=evaluationRerun.js.map
|
|
@@ -0,0 +1 @@
|
|
|
1
|
+
{"version":3,"file":"evaluationRerun.js","sourceRoot":"","sources":["../../evaluationRerun.ts"],"names":[],"mappings":"AAAA;;;GAGG;AAcH,MAAM,gBAAgB,GAAG,gBAAgB,CAAC;AAE1C;;;GAGG;AACH,MAAM,eAAe,GAAG,oCAAoC,CAAC;AAE7D;;;;;;;GAOG;AACH,MAAM,UAAU,gBAAgB,CAAC,UAAqC;IACpE,MAAM,OAAO,GAAG,CAAC,UAAU,IAAI,EAAE,CAAC,CAAC,IAAI,EAAE,CAAC;IAC1C,MAAM,IAAI,GAAG,OAAO,IAAI,gBAAgB,CAAC;IAEzC,MAAM,KAAK,GAAG,IAAI,CAAC,KAAK,CAAC,eAAe,CAAC,CAAC;IAC1C,IAAI,CAAC,KAAK,EAAE,CAAC;QACX,OAAO,GAAG,IAAI,WAAW,CAAC;IAC5B,CAAC;IAED,MAAM,CAAC,EAAE,WAAW,EAAE,YAAY,CAAC,GAAG,KAAK,CAAC;IAC5C,0EAA0E;IAC1E,sEAAsE;IACtE,yCAAyC;IACzC,MAAM,QAAQ,GAAG,WAAW,CAAC,IAAI,EAAE,IAAI,gBAAgB,CAAC;IACxD,MAAM,UAAU,GAAG,YAAY,CAAC,CAAC,CAAC,QAAQ,CAAC,YAAY,EAAE,EAAE,CAAC,GAAG,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC;IACrE,OAAO,GAAG,QAAQ,YAAY,UAAU,GAAG,CAAC;AAC9C,CAAC;AAsCD;;;;;;;;GAQG;AACH,MAAM,UAAU,gBAAgB,CAC9B,SAAwB;IAExB,MAAM,eAAe,GAAa,EAAE,CAAC;IAErC,IAAI,CAAC,SAAS,CAAC,QAAQ,EAAE,CAAC;QACxB,OAAO,EAAE,KAAK,EAAE,yEAAyE,EAAE,CAAC;IAC9F,CAAC;IAED,IAAI,OAAO,GAAG,SAAS,CAAC,OAAO,CAAC;IAChC,IAAI,CAAC,OAAO,IAAI,OAAO,CAAC,MAAM,KAAK,CAAC,EAAE,CAAC;QACrC,MAAM,WAAW,GAAG,CAAC,SAAS,CAAC,iBAAiB,IAAI,EAAE,CAAC,CAAC,GAAG,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC;QACvE,IAAI,WAAW,CAAC,MAAM,KAAK,CAAC,EAAE,CAAC;YAC7B,OAAO;gBACL,KAAK,EAAE,sFAAsF;aAC9F,CAAC;QACJ,CAAC;QACD,OAAO,GAAG,CAAC,EAAE,IAAI,EAAE,eAAe,EAAE,GAAG,EAAE,WAAW,EAAE,CAAC,CAAC;QACxD,eAAe,CAAC,IAAI,CAClB,8CAA8C,WAAW,CAAC,MAAM,uDAAuD,CACxH,CAAC;IACJ,CAAC;IAED,IAAI,WAAW,GAAG,SAAS,CAAC,WAAW,CAAC;IACxC,IAAI,WAAW,IAAI,IAAI,EAAE,CAAC;QACxB,WAAW,GAAG,CAAC,CAAC;QAChB,eAAe,CAAC,IAAI,CAAC,mDAAmD,CAAC,CAAC;IAC5E,CAAC;IAED,MAAM,MAAM,GAAgB;QAC1B,OAAO;QACP,QAAQ,EAAE,SAAS,CAAC,QAAQ;QAC5B,OAAO,EAAE,SAAS,CAAC,OAAO,IAAI,EAAE;QAChC,YAAY,EAAE,SAAS,CAAC,YAAY;QACpC,WAAW,EAAE,SAAS,CAAC,WAAW;QAClC,OAAO,EAAE,SAAS,CAAC,OAAO;QAC1B,WAAW;QACX,aAAa,EAAE,SAAS,CAAC,aAAa;QACtC,WAAW,EAAE,SAAS,CAAC,WAAW;QAClC,WAAW,EAAE,SAAS,CAAC,WAAW;QAClC,gBAAgB,EAAE,SAAS,CAAC,gBAAgB;KAC7C,CAAC;IAEF,uEAAuE;IACvE,2EAA2E;IAC3E,yEAAyE;IACzE,0EAA0E;IAC1E,+DAA+D;IAC/D,eAAe;IAEf,OAAO,EAAE,MAAM,EAAE,eAAe,EAAE,CAAC;AACrC,CAAC"}
|
|
@@ -18,11 +18,26 @@
|
|
|
18
18
|
* accessor.
|
|
19
19
|
*/
|
|
20
20
|
import type { Span } from '../../types/index.js';
|
|
21
|
+
/**
|
|
22
|
+
* Where `totalCost` came from:
|
|
23
|
+
* - `'reported'` — every dollar came from a real cost attribute on a span.
|
|
24
|
+
* - `'computed'` — no span carried a cost attribute; the whole total was
|
|
25
|
+
* derived from token counts via the fallback pricing table.
|
|
26
|
+
* - `'mixed'` — some spans reported cost, others were computed.
|
|
27
|
+
* - `'none'` — no LLM spans, or no cost could be determined at all.
|
|
28
|
+
*/
|
|
29
|
+
export type CostSource = 'reported' | 'computed' | 'mixed' | 'none';
|
|
21
30
|
export interface TracesAccessor {
|
|
22
|
-
/** Sum of all
|
|
31
|
+
/** Sum of all prompt/completion/cache token-usage attrs across LLM spans. */
|
|
23
32
|
totalTokens: number;
|
|
24
|
-
/**
|
|
33
|
+
/**
|
|
34
|
+
* Sum of all reported cost attrs, falling back to a computed estimate
|
|
35
|
+
* (from token counts + the model pricing table) when no span reports
|
|
36
|
+
* cost directly. See `costSource` to tell which happened.
|
|
37
|
+
*/
|
|
25
38
|
totalCost: number;
|
|
39
|
+
/** Whether `totalCost` was reported by spans, computed, both, or neither. */
|
|
40
|
+
costSource: CostSource;
|
|
26
41
|
/** Tool invocation summary derived from spans. */
|
|
27
42
|
toolCalls: ReadonlyArray<{
|
|
28
43
|
name: string;
|
|
@@ -1 +1 @@
|
|
|
1
|
-
{"version":3,"file":"traces.d.ts","sourceRoot":"","sources":["../../../matchers/traces.ts"],"names":[],"mappings":"AAKA;;;;;;;;;;;;;;;;;;GAkBG;AAEH,OAAO,KAAK,EAAE,IAAI,EAAE,MAAM,sBAAsB,CAAC;
|
|
1
|
+
{"version":3,"file":"traces.d.ts","sourceRoot":"","sources":["../../../matchers/traces.ts"],"names":[],"mappings":"AAKA;;;;;;;;;;;;;;;;;;GAkBG;AAEH,OAAO,KAAK,EAAE,IAAI,EAAE,MAAM,sBAAsB,CAAC;AAGjD;;;;;;;GAOG;AACH,MAAM,MAAM,UAAU,GAAG,UAAU,GAAG,UAAU,GAAG,OAAO,GAAG,MAAM,CAAC;AAEpE,MAAM,WAAW,cAAc;IAC7B,6EAA6E;IAC7E,WAAW,EAAE,MAAM,CAAC;IACpB;;;;OAIG;IACH,SAAS,EAAE,MAAM,CAAC;IAClB,6EAA6E;IAC7E,UAAU,EAAE,UAAU,CAAC;IACvB,kDAAkD;IAClD,SAAS,EAAE,aAAa,CAAC;QAAE,IAAI,EAAE,MAAM,CAAC;QAAC,UAAU,EAAE,MAAM,CAAA;KAAE,CAAC,CAAC;IAC/D,+EAA+E;IAC/E,YAAY,CAAC,IAAI,EAAE,MAAM,GAAG,MAAM,CAAC;IACnC,6BAA6B;IAC7B,KAAK,EAAE,aAAa,CAAC,IAAI,CAAC,CAAC;CAC5B;AAED;;;;GAIG;AACH,wBAAgB,mBAAmB,IAAI,cAAc,CASpD;AAED;;;;;;;;;GASG;AACH,wBAAgB,yBAAyB,CAAC,MAAM,EAAE,MAAM,GAAG,cAAc,CAYxE;AAgCD,wDAAwD;AACxD,wBAAgB,mBAAmB,CAAC,KAAK,EAAE,IAAI,EAAE,GAAG,cAAc,CA+IjE"}
|
|
@@ -2,6 +2,7 @@
|
|
|
2
2
|
* Copyright OpenSearch Contributors
|
|
3
3
|
* SPDX-License-Identifier: Apache-2.0
|
|
4
4
|
*/
|
|
5
|
+
import { lookupModelRates, computeCost } from './tracesPricing.js';
|
|
5
6
|
/**
|
|
6
7
|
* Empty accessor — silent zeros. Used when the agent opted out of traces
|
|
7
8
|
* (`useTraces: false`). Reading any accessor returns `0` / `[]` so user
|
|
@@ -11,6 +12,7 @@ export function emptyTracesAccessor() {
|
|
|
11
12
|
return {
|
|
12
13
|
totalTokens: 0,
|
|
13
14
|
totalCost: 0,
|
|
15
|
+
costSource: 'none',
|
|
14
16
|
toolCalls: [],
|
|
15
17
|
spanDuration: () => 0,
|
|
16
18
|
spans: [],
|
|
@@ -33,48 +35,166 @@ export function unavailableTracesAccessor(reason) {
|
|
|
33
35
|
return {
|
|
34
36
|
get totalTokens() { return fail(); },
|
|
35
37
|
get totalCost() { return fail(); },
|
|
38
|
+
get costSource() { return fail(); },
|
|
36
39
|
get toolCalls() { return fail(); },
|
|
37
40
|
get spans() { return fail(); },
|
|
38
41
|
spanDuration(_name) { return fail(); },
|
|
39
42
|
};
|
|
40
43
|
}
|
|
44
|
+
const PROMPT_TOKEN_KEYS = [
|
|
45
|
+
'gen_ai.usage.prompt_tokens',
|
|
46
|
+
'gen_ai.usage.input_tokens',
|
|
47
|
+
'llm.usage.prompt_tokens',
|
|
48
|
+
'input_tokens',
|
|
49
|
+
];
|
|
50
|
+
const COMPLETION_TOKEN_KEYS = [
|
|
51
|
+
'gen_ai.usage.completion_tokens',
|
|
52
|
+
'gen_ai.usage.output_tokens',
|
|
53
|
+
'llm.usage.completion_tokens',
|
|
54
|
+
'output_tokens',
|
|
55
|
+
];
|
|
56
|
+
const CACHE_READ_TOKEN_KEYS = [
|
|
57
|
+
'gen_ai.usage.cache_read_input_tokens',
|
|
58
|
+
'gen_ai.usage.cache_read_tokens',
|
|
59
|
+
'cache_read_tokens',
|
|
60
|
+
];
|
|
61
|
+
const CACHE_CREATION_TOKEN_KEYS = [
|
|
62
|
+
'gen_ai.usage.cache_creation_input_tokens',
|
|
63
|
+
'gen_ai.usage.cache_creation_tokens',
|
|
64
|
+
'cache_creation_tokens',
|
|
65
|
+
];
|
|
66
|
+
const MODEL_ID_KEYS = ['gen_ai.request.model', 'gen_ai.response.model', 'model'];
|
|
67
|
+
const ALL_TOKEN_KEYS = [
|
|
68
|
+
...PROMPT_TOKEN_KEYS,
|
|
69
|
+
...COMPLETION_TOKEN_KEYS,
|
|
70
|
+
...CACHE_READ_TOKEN_KEYS,
|
|
71
|
+
...CACHE_CREATION_TOKEN_KEYS,
|
|
72
|
+
];
|
|
41
73
|
/** Build a TracesAccessor from a flat list of spans. */
|
|
42
74
|
export function buildTracesAccessor(spans) {
|
|
43
75
|
let totalTokens = 0;
|
|
44
|
-
let
|
|
76
|
+
let reportedCost = 0;
|
|
77
|
+
let computedCost = 0;
|
|
78
|
+
let hadReportedCost = false;
|
|
79
|
+
let hadComputedCost = false;
|
|
80
|
+
let tokenBearingSpanCount = 0;
|
|
81
|
+
// codex_review (PR #440): a span can have recognized tokens yet resolve
|
|
82
|
+
// NEITHER a reported cost NOR a computed one (unpriced/unresolved model) —
|
|
83
|
+
// this counter drives `costUnknowable` below so totalCost refuses to
|
|
84
|
+
// silently under-report just because *some other* span in the same trace
|
|
85
|
+
// happened to be priceable.
|
|
86
|
+
let unresolvedCostSpanCount = 0;
|
|
45
87
|
const toolCalls = [];
|
|
46
88
|
for (const span of spans) {
|
|
47
89
|
const attrs = span.attributes ?? {};
|
|
48
|
-
const promptTokens = pickNumber(attrs,
|
|
49
|
-
|
|
50
|
-
|
|
51
|
-
|
|
52
|
-
|
|
53
|
-
|
|
54
|
-
|
|
55
|
-
|
|
56
|
-
|
|
57
|
-
|
|
90
|
+
const promptTokens = pickNumber(attrs, PROMPT_TOKEN_KEYS);
|
|
91
|
+
const completionTokens = pickNumber(attrs, COMPLETION_TOKEN_KEYS);
|
|
92
|
+
const cacheReadTokens = pickNumber(attrs, CACHE_READ_TOKEN_KEYS);
|
|
93
|
+
const cacheCreationTokens = pickNumber(attrs, CACHE_CREATION_TOKEN_KEYS);
|
|
94
|
+
const hasTokenAttr = promptTokens !== undefined ||
|
|
95
|
+
completionTokens !== undefined ||
|
|
96
|
+
cacheReadTokens !== undefined ||
|
|
97
|
+
cacheCreationTokens !== undefined;
|
|
98
|
+
if (hasTokenAttr)
|
|
99
|
+
tokenBearingSpanCount += 1;
|
|
58
100
|
if (promptTokens)
|
|
59
101
|
totalTokens += promptTokens;
|
|
60
102
|
if (completionTokens)
|
|
61
103
|
totalTokens += completionTokens;
|
|
62
|
-
|
|
104
|
+
if (cacheReadTokens)
|
|
105
|
+
totalTokens += cacheReadTokens;
|
|
106
|
+
if (cacheCreationTokens)
|
|
107
|
+
totalTokens += cacheCreationTokens;
|
|
108
|
+
const reported = pickNumber(attrs, [
|
|
63
109
|
'gen_ai.usage.cost_usd',
|
|
64
110
|
'gen_ai.usage.cost',
|
|
65
111
|
'llm.usage.cost_usd',
|
|
66
112
|
]);
|
|
67
|
-
if (
|
|
68
|
-
|
|
113
|
+
if (reported !== undefined) {
|
|
114
|
+
hadReportedCost = true;
|
|
115
|
+
reportedCost += reported;
|
|
116
|
+
}
|
|
117
|
+
else if (hasTokenAttr) {
|
|
118
|
+
// No reported cost attribute on this span — fall back to a computed
|
|
119
|
+
// estimate from token counts, if we know pricing for the model.
|
|
120
|
+
const modelId = pickString(attrs, MODEL_ID_KEYS);
|
|
121
|
+
const rates = lookupModelRates(modelId);
|
|
122
|
+
if (rates) {
|
|
123
|
+
hadComputedCost = true;
|
|
124
|
+
computedCost += computeCost({
|
|
125
|
+
input: promptTokens ?? 0,
|
|
126
|
+
output: completionTokens ?? 0,
|
|
127
|
+
cacheRead: cacheReadTokens ?? 0,
|
|
128
|
+
cacheCreation: cacheCreationTokens ?? 0,
|
|
129
|
+
}, rates);
|
|
130
|
+
}
|
|
131
|
+
else {
|
|
132
|
+
// This span's tokens contribute to totalTokens but NOT to totalCost —
|
|
133
|
+
// without this counter that would be a silent, plausible-looking
|
|
134
|
+
// undercount rather than the loud failure #230 asks for.
|
|
135
|
+
unresolvedCostSpanCount += 1;
|
|
136
|
+
}
|
|
137
|
+
}
|
|
69
138
|
// Tool spans
|
|
70
139
|
const toolName = pickString(attrs, ['gen_ai.tool.name', 'llm.tool.name']);
|
|
71
140
|
if (toolName) {
|
|
72
141
|
toolCalls.push({ name: toolName, durationMs: spanDurationMs(span) });
|
|
73
142
|
}
|
|
74
143
|
}
|
|
144
|
+
const totalCostValue = reportedCost + computedCost;
|
|
145
|
+
const costSource = hadReportedCost && hadComputedCost
|
|
146
|
+
? 'mixed'
|
|
147
|
+
: hadReportedCost
|
|
148
|
+
? 'reported'
|
|
149
|
+
: hadComputedCost
|
|
150
|
+
? 'computed'
|
|
151
|
+
: 'none';
|
|
152
|
+
// Issue class: spans exist, but none of them carry any token-usage
|
|
153
|
+
// attribute we recognize. Reading totalTokens/totalCost would silently
|
|
154
|
+
// return 0, letting matchers like `expect(traces.totalCost).lessThan(2)`
|
|
155
|
+
// pass vacuously (mirrors the #230 fail-loud philosophy). A trace that
|
|
156
|
+
// genuinely has recognized-but-zero-valued attributes is fine and must
|
|
157
|
+
// NOT throw — only the "we found nothing we understand" case throws.
|
|
158
|
+
const noRecognizedTokenAttrs = spans.length > 0 && tokenBearingSpanCount === 0;
|
|
159
|
+
// Issue class: tokens were found, but AT LEAST ONE token-bearing span
|
|
160
|
+
// resolved neither a reported cost nor a computed one (unpriced model,
|
|
161
|
+
// no cost attribute). Pre-codex_review this only threw when EVERY span
|
|
162
|
+
// was unresolved — a trace with 9 priced spans and 1 unpriced span
|
|
163
|
+
// silently under-reported totalCost by exactly the unpriced span's real
|
|
164
|
+
// spend while still returning a plausible-looking number (PR #440 review
|
|
165
|
+
// finding, blocker). Throw only when reading totalCost (tokens themselves
|
|
166
|
+
// are legitimately known in this case) — any unresolved span makes the
|
|
167
|
+
// total untrustworthy, not just a total absence of resolved spans.
|
|
168
|
+
const costUnknowable = totalTokens > 0 && unresolvedCostSpanCount > 0;
|
|
75
169
|
return {
|
|
76
|
-
totalTokens
|
|
77
|
-
|
|
170
|
+
get totalTokens() {
|
|
171
|
+
if (noRecognizedTokenAttrs) {
|
|
172
|
+
throw new Error(`traces fixture: ${spans.length} span(s) present but none carried a recognized ` +
|
|
173
|
+
`token-usage attribute (checked ${ALL_TOKEN_KEYS.join(', ')}). Reading totalTokens ` +
|
|
174
|
+
`would silently return 0 — either this agent's spans use an unsupported attribute ` +
|
|
175
|
+
`shape (extend buildTracesAccessor() in lib/matchers/traces.ts), or this test case ` +
|
|
176
|
+
`has no LLM spans and should use traces.spans / traces.toolCalls instead of totalTokens.`);
|
|
177
|
+
}
|
|
178
|
+
return totalTokens;
|
|
179
|
+
},
|
|
180
|
+
get totalCost() {
|
|
181
|
+
if (noRecognizedTokenAttrs && !hadReportedCost) {
|
|
182
|
+
throw new Error(`traces fixture: ${spans.length} span(s) present but none carried a recognized ` +
|
|
183
|
+
`token-usage attribute or a cost attribute, so totalCost cannot be determined. ` +
|
|
184
|
+
`Reading totalCost would silently return 0.`);
|
|
185
|
+
}
|
|
186
|
+
if (costUnknowable) {
|
|
187
|
+
throw new Error(`traces fixture: ${unresolvedCostSpanCount} of ${tokenBearingSpanCount} span(s) with token usage ` +
|
|
188
|
+
`carried neither a cost attribute nor a model id matching the fallback pricing table (see ` +
|
|
189
|
+
`lib/matchers/tracesPricing.ts). Reading totalCost would silently UNDER-report real spend by ` +
|
|
190
|
+
`omitting those spans rather than reflect it -- add the model(s) to the pricing table or stamp a ` +
|
|
191
|
+
`cost attribute on the span(s), or use traces.spans to inspect per-span usage directly.`);
|
|
192
|
+
}
|
|
193
|
+
return totalCostValue;
|
|
194
|
+
},
|
|
195
|
+
get costSource() {
|
|
196
|
+
return costSource;
|
|
197
|
+
},
|
|
78
198
|
toolCalls,
|
|
79
199
|
spanDuration(name) {
|
|
80
200
|
const hit = spans.find(s => s.name === name);
|
|
@@ -1 +1 @@
|
|
|
1
|
-
{"version":3,"file":"traces.js","sourceRoot":"","sources":["../../../matchers/traces.ts"],"names":[],"mappings":"AAAA;;;GAGG;
|
|
1
|
+
{"version":3,"file":"traces.js","sourceRoot":"","sources":["../../../matchers/traces.ts"],"names":[],"mappings":"AAAA;;;GAGG;AAuBH,OAAO,EAAE,gBAAgB,EAAE,WAAW,EAAE,MAAM,oBAAoB,CAAC;AA+BnE;;;;GAIG;AACH,MAAM,UAAU,mBAAmB;IACjC,OAAO;QACL,WAAW,EAAE,CAAC;QACd,SAAS,EAAE,CAAC;QACZ,UAAU,EAAE,MAAM;QAClB,SAAS,EAAE,EAAE;QACb,YAAY,EAAE,GAAG,EAAE,CAAC,CAAC;QACrB,KAAK,EAAE,EAAE;KACV,CAAC;AACJ,CAAC;AAED;;;;;;;;;GASG;AACH,MAAM,UAAU,yBAAyB,CAAC,MAAc;IACtD,MAAM,IAAI,GAAG,GAAU,EAAE;QACvB,MAAM,IAAI,KAAK,CAAC,+BAA+B,MAAM,EAAE,CAAC,CAAC;IAC3D,CAAC,CAAC;IACF,OAAO;QACL,IAAI,WAAW,KAAa,OAAO,IAAI,EAAE,CAAC,CAAC,CAAC;QAC5C,IAAI,SAAS,KAAa,OAAO,IAAI,EAAE,CAAC,CAAC,CAAC;QAC1C,IAAI,UAAU,KAAiB,OAAO,IAAI,EAAE,CAAC,CAAC,CAAC;QAC/C,IAAI,SAAS,KAA0D,OAAO,IAAI,EAAE,CAAC,CAAC,CAAC;QACvF,IAAI,KAAK,KAA0B,OAAO,IAAI,EAAE,CAAC,CAAC,CAAC;QACnD,YAAY,CAAC,KAAa,IAAY,OAAO,IAAI,EAAE,CAAC,CAAC,CAAC;KACrC,CAAC;AACtB,CAAC;AAED,MAAM,iBAAiB,GAAG;IACxB,4BAA4B;IAC5B,2BAA2B;IAC3B,yBAAyB;IACzB,cAAc;CACf,CAAC;AACF,MAAM,qBAAqB,GAAG;IAC5B,gCAAgC;IAChC,4BAA4B;IAC5B,6BAA6B;IAC7B,eAAe;CAChB,CAAC;AACF,MAAM,qBAAqB,GAAG;IAC5B,sCAAsC;IACtC,gCAAgC;IAChC,mBAAmB;CACpB,CAAC;AACF,MAAM,yBAAyB,GAAG;IAChC,0CAA0C;IAC1C,oCAAoC;IACpC,uBAAuB;CACxB,CAAC;AACF,MAAM,aAAa,GAAG,CAAC,sBAAsB,EAAE,uBAAuB,EAAE,OAAO,CAAC,CAAC;AACjF,MAAM,cAAc,GAAG;IACrB,GAAG,iBAAiB;IACpB,GAAG,qBAAqB;IACxB,GAAG,qBAAqB;IACxB,GAAG,yBAAyB;CAC7B,CAAC;AAEF,wDAAwD;AACxD,MAAM,UAAU,mBAAmB,CAAC,KAAa;IAC/C,IAAI,WAAW,GAAG,CAAC,CAAC;IACpB,IAAI,YAAY,GAAG,CAAC,CAAC;IACrB,IAAI,YAAY,GAAG,CAAC,CAAC;IACrB,IAAI,eAAe,GAAG,KAAK,CAAC;IAC5B,IAAI,eAAe,GAAG,KAAK,CAAC;IAC5B,IAAI,qBAAqB,GAAG,CAAC,CAAC;IAC9B,wEAAwE;IACxE,2EAA2E;IAC3E,qEAAqE;IACrE,yEAAyE;IACzE,4BAA4B;IAC5B,IAAI,uBAAuB,GAAG,CAAC,CAAC;IAChC,MAAM,SAAS,GAA2C,EAAE,CAAC;IAE7D,KAAK,MAAM,IAAI,IAAI,KAAK,EAAE,CAAC;QACzB,MAAM,KAAK,GAAG,IAAI,CAAC,UAAU,IAAI,EAAE,CAAC;QACpC,MAAM,YAAY,GAAG,UAAU,CAAC,KAAK,EAAE,iBAAiB,CAAC,CAAC;QAC1D,MAAM,gBAAgB,GAAG,UAAU,CAAC,KAAK,EAAE,qBAAqB,CAAC,CAAC;QAClE,MAAM,eAAe,GAAG,UAAU,CAAC,KAAK,EAAE,qBAAqB,CAAC,CAAC;QACjE,MAAM,mBAAmB,GAAG,UAAU,CAAC,KAAK,EAAE,yBAAyB,CAAC,CAAC;QAEzE,MAAM,YAAY,GAChB,YAAY,KAAK,SAAS;YAC1B,gBAAgB,KAAK,SAAS;YAC9B,eAAe,KAAK,SAAS;YAC7B,mBAAmB,KAAK,SAAS,CAAC;QAEpC,IAAI,YAAY;YAAE,qBAAqB,IAAI,CAAC,CAAC;QAC7C,IAAI,YAAY;YAAE,WAAW,IAAI,YAAY,CAAC;QAC9C,IAAI,gBAAgB;YAAE,WAAW,IAAI,gBAAgB,CAAC;QACtD,IAAI,eAAe;YAAE,WAAW,IAAI,eAAe,CAAC;QACpD,IAAI,mBAAmB;YAAE,WAAW,IAAI,mBAAmB,CAAC;QAE5D,MAAM,QAAQ,GAAG,UAAU,CAAC,KAAK,EAAE;YACjC,uBAAuB;YACvB,mBAAmB;YACnB,oBAAoB;SACrB,CAAC,CAAC;QACH,IAAI,QAAQ,KAAK,SAAS,EAAE,CAAC;YAC3B,eAAe,GAAG,IAAI,CAAC;YACvB,YAAY,IAAI,QAAQ,CAAC;QAC3B,CAAC;aAAM,IAAI,YAAY,EAAE,CAAC;YACxB,oEAAoE;YACpE,gEAAgE;YAChE,MAAM,OAAO,GAAG,UAAU,CAAC,KAAK,EAAE,aAAa,CAAC,CAAC;YACjD,MAAM,KAAK,GAAG,gBAAgB,CAAC,OAAO,CAAC,CAAC;YACxC,IAAI,KAAK,EAAE,CAAC;gBACV,eAAe,GAAG,IAAI,CAAC;gBACvB,YAAY,IAAI,WAAW,CACzB;oBACE,KAAK,EAAE,YAAY,IAAI,CAAC;oBACxB,MAAM,EAAE,gBAAgB,IAAI,CAAC;oBAC7B,SAAS,EAAE,eAAe,IAAI,CAAC;oBAC/B,aAAa,EAAE,mBAAmB,IAAI,CAAC;iBACxC,EACD,KAAK,CACN,CAAC;YACJ,CAAC;iBAAM,CAAC;gBACN,sEAAsE;gBACtE,iEAAiE;gBACjE,yDAAyD;gBACzD,uBAAuB,IAAI,CAAC,CAAC;YAC/B,CAAC;QACH,CAAC;QAED,aAAa;QACb,MAAM,QAAQ,GAAG,UAAU,CAAC,KAAK,EAAE,CAAC,kBAAkB,EAAE,eAAe,CAAC,CAAC,CAAC;QAC1E,IAAI,QAAQ,EAAE,CAAC;YACb,SAAS,CAAC,IAAI,CAAC,EAAE,IAAI,EAAE,QAAQ,EAAE,UAAU,EAAE,cAAc,CAAC,IAAI,CAAC,EAAE,CAAC,CAAC;QACvE,CAAC;IACH,CAAC;IAED,MAAM,cAAc,GAAG,YAAY,GAAG,YAAY,CAAC;IACnD,MAAM,UAAU,GAAe,eAAe,IAAI,eAAe;QAC/D,CAAC,CAAC,OAAO;QACT,CAAC,CAAC,eAAe;YACf,CAAC,CAAC,UAAU;YACZ,CAAC,CAAC,eAAe;gBACf,CAAC,CAAC,UAAU;gBACZ,CAAC,CAAC,MAAM,CAAC;IAEf,mEAAmE;IACnE,uEAAuE;IACvE,yEAAyE;IACzE,uEAAuE;IACvE,uEAAuE;IACvE,qEAAqE;IACrE,MAAM,sBAAsB,GAAG,KAAK,CAAC,MAAM,GAAG,CAAC,IAAI,qBAAqB,KAAK,CAAC,CAAC;IAE/E,sEAAsE;IACtE,uEAAuE;IACvE,uEAAuE;IACvE,mEAAmE;IACnE,wEAAwE;IACxE,yEAAyE;IACzE,0EAA0E;IAC1E,uEAAuE;IACvE,mEAAmE;IACnE,MAAM,cAAc,GAAG,WAAW,GAAG,CAAC,IAAI,uBAAuB,GAAG,CAAC,CAAC;IAEtE,OAAO;QACL,IAAI,WAAW;YACb,IAAI,sBAAsB,EAAE,CAAC;gBAC3B,MAAM,IAAI,KAAK,CACb,mBAAmB,KAAK,CAAC,MAAM,iDAAiD;oBAChF,kCAAkC,cAAc,CAAC,IAAI,CAAC,IAAI,CAAC,yBAAyB;oBACpF,mFAAmF;oBACnF,oFAAoF;oBACpF,yFAAyF,CAC1F,CAAC;YACJ,CAAC;YACD,OAAO,WAAW,CAAC;QACrB,CAAC;QACD,IAAI,SAAS;YACX,IAAI,sBAAsB,IAAI,CAAC,eAAe,EAAE,CAAC;gBAC/C,MAAM,IAAI,KAAK,CACb,mBAAmB,KAAK,CAAC,MAAM,iDAAiD;oBAChF,gFAAgF;oBAChF,4CAA4C,CAC7C,CAAC;YACJ,CAAC;YACD,IAAI,cAAc,EAAE,CAAC;gBACnB,MAAM,IAAI,KAAK,CACb,mBAAmB,uBAAuB,OAAO,qBAAqB,4BAA4B;oBAClG,2FAA2F;oBAC3F,8FAA8F;oBAC9F,kGAAkG;oBAClG,wFAAwF,CACzF,CAAC;YACJ,CAAC;YACD,OAAO,cAAc,CAAC;QACxB,CAAC;QACD,IAAI,UAAU;YACZ,OAAO,UAAU,CAAC;QACpB,CAAC;QACD,SAAS;QACT,YAAY,CAAC,IAAY;YACvB,MAAM,GAAG,GAAG,KAAK,CAAC,IAAI,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC,IAAI,KAAK,IAAI,CAAC,CAAC;YAC7C,OAAO,GAAG,CAAC,CAAC,CAAC,cAAc,CAAC,GAAG,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC;QACvC,CAAC;QACD,KAAK;KACN,CAAC;AACJ,CAAC;AAED,SAAS,cAAc,CAAC,IAAU;IAChC,MAAM,KAAK,GAAI,IAAY,CAAC,iBAAiB,IAAK,IAAY,CAAC,SAAS,CAAC;IACzE,MAAM,GAAG,GAAI,IAAY,CAAC,eAAe,IAAK,IAAY,CAAC,OAAO,CAAC;IACnE,IAAI,OAAO,KAAK,KAAK,QAAQ,IAAI,OAAO,GAAG,KAAK,QAAQ,EAAE,CAAC;QACzD,4DAA4D;QAC5D,MAAM,MAAM,GAAG,IAAI,CAAC,GAAG,CAAC,KAAK,CAAC,GAAG,IAAI,CAAC,CAAC,CAAC,GAAG,CAAC,CAAC,CAAC,CAAC,CAAC;QAChD,OAAO,IAAI,CAAC,GAAG,CAAC,CAAC,EAAE,IAAI,CAAC,KAAK,CAAC,CAAC,GAAG,GAAG,KAAK,CAAC,GAAG,MAAM,CAAC,CAAC,CAAC;IACzD,CAAC;IACD,IAAI,OAAQ,IAAY,CAAC,UAAU,KAAK,QAAQ;QAAE,OAAQ,IAAY,CAAC,UAAU,CAAC;IAClF,OAAO,CAAC,CAAC;AACX,CAAC;AAED,SAAS,UAAU,CAAC,KAA8B,EAAE,IAAc;IAChE,KAAK,MAAM,CAAC,IAAI,IAAI,EAAE,CAAC;QACrB,MAAM,CAAC,GAAG,KAAK,CAAC,CAAC,CAAC,CAAC;QACnB,IAAI,OAAO,CAAC,KAAK,QAAQ,IAAI,MAAM,CAAC,QAAQ,CAAC,CAAC,CAAC;YAAE,OAAO,CAAC,CAAC;QAC1D,IAAI,OAAO,CAAC,KAAK,QAAQ,IAAI,CAAC,CAAC,IAAI,EAAE,KAAK,EAAE,IAAI,MAAM,CAAC,QAAQ,CAAC,MAAM,CAAC,CAAC,CAAC,CAAC;YAAE,OAAO,MAAM,CAAC,CAAC,CAAC,CAAC;IAC/F,CAAC;IACD,OAAO,SAAS,CAAC;AACnB,CAAC;AAED,SAAS,UAAU,CAAC,KAA8B,EAAE,IAAc;IAChE,KAAK,MAAM,CAAC,IAAI,IAAI,EAAE,CAAC;QACrB,MAAM,CAAC,GAAG,KAAK,CAAC,CAAC,CAAC,CAAC;QACnB,IAAI,OAAO,CAAC,KAAK,QAAQ,IAAI,CAAC,CAAC,MAAM,GAAG,CAAC;YAAE,OAAO,CAAC,CAAC;IACtD,CAAC;IACD,OAAO,SAAS,CAAC;AACnB,CAAC"}
|
|
@@ -0,0 +1,38 @@
|
|
|
1
|
+
/**
|
|
2
|
+
* Fallback pricing table used by `buildTracesAccessor()` (see
|
|
3
|
+
* `lib/matchers/traces.ts`) to *compute* `totalCost` when spans carry
|
|
4
|
+
* token-usage attributes but no cost attribute (this is the common case —
|
|
5
|
+
* most OTel exporters, including Claude Code's, emit raw token counts and
|
|
6
|
+
* leave cost calculation to the consumer).
|
|
7
|
+
*
|
|
8
|
+
* Rates are USD per **million tokens** (MTok), converted to per-token below.
|
|
9
|
+
* These are point-in-time *estimates* maintained by hand — update them when
|
|
10
|
+
* pricing changes. They are intentionally separate from
|
|
11
|
+
* `server/services/codingAgents/pricing.ts` (used by the coding-agent
|
|
12
|
+
* analytics/leaderboard feature): that table silently falls back to Sonnet
|
|
13
|
+
* pricing for unknown models, which is the wrong behavior here — an eval
|
|
14
|
+
* reading `traces.totalCost` for an unpriced model must know the number is
|
|
15
|
+
* incomplete (see `buildTracesAccessor`'s fail-loud handling), not get a
|
|
16
|
+
* silently-substituted estimate.
|
|
17
|
+
*/
|
|
18
|
+
export interface ModelRates {
|
|
19
|
+
/** USD per input (prompt) token. */
|
|
20
|
+
input: number;
|
|
21
|
+
/** USD per output (completion) token. */
|
|
22
|
+
output: number;
|
|
23
|
+
/** USD per cache-read token (cheaper than a fresh input token). */
|
|
24
|
+
cacheRead: number;
|
|
25
|
+
/** USD per cache-creation/write token (pricier than a fresh input token). */
|
|
26
|
+
cacheCreation: number;
|
|
27
|
+
}
|
|
28
|
+
/** Look up rates for a model id. Returns `undefined` when no rule matches. */
|
|
29
|
+
export declare function lookupModelRates(modelId: string | undefined): ModelRates | undefined;
|
|
30
|
+
export interface UsageTokens {
|
|
31
|
+
input: number;
|
|
32
|
+
output: number;
|
|
33
|
+
cacheRead: number;
|
|
34
|
+
cacheCreation: number;
|
|
35
|
+
}
|
|
36
|
+
/** Compute USD cost for a set of token counts against a model's rates. */
|
|
37
|
+
export declare function computeCost(tokens: UsageTokens, rates: ModelRates): number;
|
|
38
|
+
//# sourceMappingURL=tracesPricing.d.ts.map
|
|
@@ -0,0 +1 @@
|
|
|
1
|
+
{"version":3,"file":"tracesPricing.d.ts","sourceRoot":"","sources":["../../../matchers/tracesPricing.ts"],"names":[],"mappings":"AAKA;;;;;;;;;;;;;;;;GAgBG;AAEH,MAAM,WAAW,UAAU;IACzB,oCAAoC;IACpC,KAAK,EAAE,MAAM,CAAC;IACd,yCAAyC;IACzC,MAAM,EAAE,MAAM,CAAC;IACf,mEAAmE;IACnE,SAAS,EAAE,MAAM,CAAC;IAClB,6EAA6E;IAC7E,aAAa,EAAE,MAAM,CAAC;CACvB;AA8CD,8EAA8E;AAC9E,wBAAgB,gBAAgB,CAAC,OAAO,EAAE,MAAM,GAAG,SAAS,GAAG,UAAU,GAAG,SAAS,CAMpF;AAED,MAAM,WAAW,WAAW;IAC1B,KAAK,EAAE,MAAM,CAAC;IACd,MAAM,EAAE,MAAM,CAAC;IACf,SAAS,EAAE,MAAM,CAAC;IAClB,aAAa,EAAE,MAAM,CAAC;CACvB;AAED,0EAA0E;AAC1E,wBAAgB,WAAW,CAAC,MAAM,EAAE,WAAW,EAAE,KAAK,EAAE,UAAU,GAAG,MAAM,CAO1E"}
|
|
@@ -0,0 +1,64 @@
|
|
|
1
|
+
/*
|
|
2
|
+
* Copyright OpenSearch Contributors
|
|
3
|
+
* SPDX-License-Identifier: Apache-2.0
|
|
4
|
+
*/
|
|
5
|
+
const PER_MTOK = 1_000_000;
|
|
6
|
+
/**
|
|
7
|
+
* Estimated Bedrock/Anthropic Claude rates, USD per MTok:
|
|
8
|
+
* Sonnet class: $3 in / $15 out / $0.30 cache-read / $3.75 cache-write
|
|
9
|
+
* Opus class: $15 in / $75 out / $1.50 cache-read / $18.75 cache-write
|
|
10
|
+
* Haiku class: $0.80 in / $4 out / $0.08 cache-read / $1.00 cache-write
|
|
11
|
+
* Keyed by substrings matched case-insensitively against the span's model
|
|
12
|
+
* id (see `resolveModelId`), so both raw Anthropic ids
|
|
13
|
+
* (`claude-sonnet-4-6-...`) and Bedrock ids
|
|
14
|
+
* (`us.anthropic.claude-sonnet-4-6-...`) match.
|
|
15
|
+
*/
|
|
16
|
+
const RATE_TABLE = [
|
|
17
|
+
{
|
|
18
|
+
match: /opus/i,
|
|
19
|
+
rates: {
|
|
20
|
+
input: 15 / PER_MTOK,
|
|
21
|
+
output: 75 / PER_MTOK,
|
|
22
|
+
cacheRead: 1.5 / PER_MTOK,
|
|
23
|
+
cacheCreation: 18.75 / PER_MTOK,
|
|
24
|
+
},
|
|
25
|
+
},
|
|
26
|
+
{
|
|
27
|
+
match: /haiku/i,
|
|
28
|
+
rates: {
|
|
29
|
+
input: 0.8 / PER_MTOK,
|
|
30
|
+
output: 4 / PER_MTOK,
|
|
31
|
+
cacheRead: 0.08 / PER_MTOK,
|
|
32
|
+
cacheCreation: 1.0 / PER_MTOK,
|
|
33
|
+
},
|
|
34
|
+
},
|
|
35
|
+
{
|
|
36
|
+
// Sonnet is the default Claude class; matched last so opus/haiku take
|
|
37
|
+
// priority, but also serves as the generic "claude" match.
|
|
38
|
+
match: /sonnet|claude/i,
|
|
39
|
+
rates: {
|
|
40
|
+
input: 3 / PER_MTOK,
|
|
41
|
+
output: 15 / PER_MTOK,
|
|
42
|
+
cacheRead: 0.3 / PER_MTOK,
|
|
43
|
+
cacheCreation: 3.75 / PER_MTOK,
|
|
44
|
+
},
|
|
45
|
+
},
|
|
46
|
+
];
|
|
47
|
+
/** Look up rates for a model id. Returns `undefined` when no rule matches. */
|
|
48
|
+
export function lookupModelRates(modelId) {
|
|
49
|
+
if (!modelId)
|
|
50
|
+
return undefined;
|
|
51
|
+
for (const { match, rates } of RATE_TABLE) {
|
|
52
|
+
if (match.test(modelId))
|
|
53
|
+
return rates;
|
|
54
|
+
}
|
|
55
|
+
return undefined;
|
|
56
|
+
}
|
|
57
|
+
/** Compute USD cost for a set of token counts against a model's rates. */
|
|
58
|
+
export function computeCost(tokens, rates) {
|
|
59
|
+
return (tokens.input * rates.input +
|
|
60
|
+
tokens.output * rates.output +
|
|
61
|
+
tokens.cacheRead * rates.cacheRead +
|
|
62
|
+
tokens.cacheCreation * rates.cacheCreation);
|
|
63
|
+
}
|
|
64
|
+
//# sourceMappingURL=tracesPricing.js.map
|
|
@@ -0,0 +1 @@
|
|
|
1
|
+
{"version":3,"file":"tracesPricing.js","sourceRoot":"","sources":["../../../matchers/tracesPricing.ts"],"names":[],"mappings":"AAAA;;;GAGG;AA+BH,MAAM,QAAQ,GAAG,SAAS,CAAC;AAE3B;;;;;;;;;GASG;AACH,MAAM,UAAU,GAAgD;IAC9D;QACE,KAAK,EAAE,OAAO;QACd,KAAK,EAAE;YACL,KAAK,EAAE,EAAE,GAAG,QAAQ;YACpB,MAAM,EAAE,EAAE,GAAG,QAAQ;YACrB,SAAS,EAAE,GAAG,GAAG,QAAQ;YACzB,aAAa,EAAE,KAAK,GAAG,QAAQ;SAChC;KACF;IACD;QACE,KAAK,EAAE,QAAQ;QACf,KAAK,EAAE;YACL,KAAK,EAAE,GAAG,GAAG,QAAQ;YACrB,MAAM,EAAE,CAAC,GAAG,QAAQ;YACpB,SAAS,EAAE,IAAI,GAAG,QAAQ;YAC1B,aAAa,EAAE,GAAG,GAAG,QAAQ;SAC9B;KACF;IACD;QACE,sEAAsE;QACtE,2DAA2D;QAC3D,KAAK,EAAE,gBAAgB;QACvB,KAAK,EAAE;YACL,KAAK,EAAE,CAAC,GAAG,QAAQ;YACnB,MAAM,EAAE,EAAE,GAAG,QAAQ;YACrB,SAAS,EAAE,GAAG,GAAG,QAAQ;YACzB,aAAa,EAAE,IAAI,GAAG,QAAQ;SAC/B;KACF;CACF,CAAC;AAEF,8EAA8E;AAC9E,MAAM,UAAU,gBAAgB,CAAC,OAA2B;IAC1D,IAAI,CAAC,OAAO;QAAE,OAAO,SAAS,CAAC;IAC/B,KAAK,MAAM,EAAE,KAAK,EAAE,KAAK,EAAE,IAAI,UAAU,EAAE,CAAC;QAC1C,IAAI,KAAK,CAAC,IAAI,CAAC,OAAO,CAAC;YAAE,OAAO,KAAK,CAAC;IACxC,CAAC;IACD,OAAO,SAAS,CAAC;AACnB,CAAC;AASD,0EAA0E;AAC1E,MAAM,UAAU,WAAW,CAAC,MAAmB,EAAE,KAAiB;IAChE,OAAO,CACL,MAAM,CAAC,KAAK,GAAG,KAAK,CAAC,KAAK;QAC1B,MAAM,CAAC,MAAM,GAAG,KAAK,CAAC,MAAM;QAC5B,MAAM,CAAC,SAAS,GAAG,KAAK,CAAC,SAAS;QAClC,MAAM,CAAC,aAAa,GAAG,KAAK,CAAC,aAAa,CAC3C,CAAC;AACJ,CAAC"}
|
|
@@ -88,5 +88,29 @@ export declare function getReportIdsFromRun(run: BenchmarkRun): string[];
|
|
|
88
88
|
* @param reports - Array of reports for this run
|
|
89
89
|
* @returns Denormalized stats object (passed, failed, pending, total)
|
|
90
90
|
*/
|
|
91
|
+
/**
|
|
92
|
+
* Canonical "display" stats for a run detail/list view: prefer recomputing
|
|
93
|
+
* from the persisted per-test-case verdicts (`run.results`) via
|
|
94
|
+
* {@link bucketRunResults} — the single source of truth — and only fall
|
|
95
|
+
* back to the denormalized `run.stats` when `results` is empty (e.g. a run
|
|
96
|
+
* that hasn't started, or legacy data with no results map at all).
|
|
97
|
+
*
|
|
98
|
+
* This is THE single shared helper for pass/fail/errored/pending/total
|
|
99
|
+
* across every surface that renders run stats (runs list, benchmark runs
|
|
100
|
+
* list, run detail page, comparison page) — do not reimplement this logic
|
|
101
|
+
* locally in a component or add a differently-named twin. Historically this
|
|
102
|
+
* logic was duplicated per-page (each with its own subtly different bugs,
|
|
103
|
+
* e.g. `errored` not tracked, or `run.stats` trusted directly even when
|
|
104
|
+
* stale/inflated for trace-judged runs — a 66/84-real-passed run displaying
|
|
105
|
+
* as 84/84). Callers that only need `{passed, failed, errored, total}` can
|
|
106
|
+
* simply ignore the extra `pending` field.
|
|
107
|
+
*/
|
|
108
|
+
export declare function computeRunStats(run: {
|
|
109
|
+
results?: Record<string, {
|
|
110
|
+
status?: string;
|
|
111
|
+
passFailStatus?: string;
|
|
112
|
+
}>;
|
|
113
|
+
stats?: Partial<RunStatsType> | null;
|
|
114
|
+
}): Pick<RunStatsType, 'passed' | 'failed' | 'errored' | 'pending' | 'total'>;
|
|
91
115
|
export declare function computeRunStatsFromReports(run: BenchmarkRun, reports: EvaluationReport[]): RunStatsType;
|
|
92
116
|
//# sourceMappingURL=runStats.d.ts.map
|
|
@@ -1 +1 @@
|
|
|
1
|
-
{"version":3,"file":"runStats.d.ts","sourceRoot":"","sources":["../../runStats.ts"],"names":[],"mappings":"AAKA;;;;;;;;;;;;;GAaG;AAEH,OAAO,KAAK,EAAE,YAAY,EAAE,gBAAgB,EAAE,QAAQ,IAAI,YAAY,EAAE,MAAM,kBAAkB,CAAC;AAEjG;;GAEG;AACH,MAAM,WAAW,QAAQ;IACvB,qEAAqE;IACrE,MAAM,EAAE,MAAM,CAAC;IACf,yFAAyF;IACzF,MAAM,EAAE,MAAM,CAAC;IACf,gFAAgF;IAChF,OAAO,EAAE,MAAM,CAAC;IAChB;;;;;OAKG;IACH,OAAO,EAAE,MAAM,CAAC;IAChB,4CAA4C;IAC5C,KAAK,EAAE,MAAM,CAAC;IACd;;;;OAIG;IACH,QAAQ,EAAE,MAAM,CAAC;CAClB;AAED;;;;;;;;;;;;;;;;GAgBG;AACH,wBAAgB,gBAAgB,CAC9B,OAAO,EAAE,MAAM,CAAC,MAAM,EAAE;IAAE,MAAM,CAAC,EAAE,MAAM,CAAC;IAAC,cAAc,CAAC,EAAE,MAAM,CAAA;CAAE,CAAC,GAAG,SAAS,GAChF,IAAI,CAAC,QAAQ,EAAE,QAAQ,GAAG,QAAQ,GAAG,SAAS,GAAG,SAAS,GAAG,OAAO,CAAC,CAevE;AAED;;;;;;;;;;;GAWG;AACH,wBAAgB,iBAAiB,CAC/B,GAAG,EAAE,YAAY,EACjB,OAAO,EAAE,MAAM,CAAC,MAAM,EAAE,gBAAgB,GAAG,IAAI,CAAC,GAC/C,QAAQ,CAuEV;AAED;;;;;GAKG;AACH,wBAAgB,mBAAmB,CAAC,GAAG,EAAE,YAAY,GAAG,MAAM,EAAE,CAU/D;AAED;;;;;;;GAOG;AACH,wBAAgB,0BAA0B,CACxC,GAAG,EAAE,YAAY,EACjB,OAAO,EAAE,gBAAgB,EAAE,GAC1B,YAAY,CAed"}
|
|
1
|
+
{"version":3,"file":"runStats.d.ts","sourceRoot":"","sources":["../../runStats.ts"],"names":[],"mappings":"AAKA;;;;;;;;;;;;;GAaG;AAEH,OAAO,KAAK,EAAE,YAAY,EAAE,gBAAgB,EAAE,QAAQ,IAAI,YAAY,EAAE,MAAM,kBAAkB,CAAC;AAEjG;;GAEG;AACH,MAAM,WAAW,QAAQ;IACvB,qEAAqE;IACrE,MAAM,EAAE,MAAM,CAAC;IACf,yFAAyF;IACzF,MAAM,EAAE,MAAM,CAAC;IACf,gFAAgF;IAChF,OAAO,EAAE,MAAM,CAAC;IAChB;;;;;OAKG;IACH,OAAO,EAAE,MAAM,CAAC;IAChB,4CAA4C;IAC5C,KAAK,EAAE,MAAM,CAAC;IACd;;;;OAIG;IACH,QAAQ,EAAE,MAAM,CAAC;CAClB;AAED;;;;;;;;;;;;;;;;GAgBG;AACH,wBAAgB,gBAAgB,CAC9B,OAAO,EAAE,MAAM,CAAC,MAAM,EAAE;IAAE,MAAM,CAAC,EAAE,MAAM,CAAC;IAAC,cAAc,CAAC,EAAE,MAAM,CAAA;CAAE,CAAC,GAAG,SAAS,GAChF,IAAI,CAAC,QAAQ,EAAE,QAAQ,GAAG,QAAQ,GAAG,SAAS,GAAG,SAAS,GAAG,OAAO,CAAC,CAevE;AAED;;;;;;;;;;;GAWG;AACH,wBAAgB,iBAAiB,CAC/B,GAAG,EAAE,YAAY,EACjB,OAAO,EAAE,MAAM,CAAC,MAAM,EAAE,gBAAgB,GAAG,IAAI,CAAC,GAC/C,QAAQ,CAuEV;AAED;;;;;GAKG;AACH,wBAAgB,mBAAmB,CAAC,GAAG,EAAE,YAAY,GAAG,MAAM,EAAE,CAU/D;AAED;;;;;;;GAOG;AACH;;;;;;;;;;;;;;;;GAgBG;AACH,wBAAgB,eAAe,CAC7B,GAAG,EAAE;IACH,OAAO,CAAC,EAAE,MAAM,CAAC,MAAM,EAAE;QAAE,MAAM,CAAC,EAAE,MAAM,CAAC;QAAC,cAAc,CAAC,EAAE,MAAM,CAAA;KAAE,CAAC,CAAC;IACvE,KAAK,CAAC,EAAE,OAAO,CAAC,YAAY,CAAC,GAAG,IAAI,CAAC;CACtC,GACA,IAAI,CAAC,YAAY,EAAE,QAAQ,GAAG,QAAQ,GAAG,SAAS,GAAG,SAAS,GAAG,OAAO,CAAC,CAc3E;AAED,wBAAgB,0BAA0B,CACxC,GAAG,EAAE,YAAY,EACjB,OAAO,EAAE,gBAAgB,EAAE,GAC1B,YAAY,CAed"}
|
package/lib/dist/lib/runStats.js
CHANGED
|
@@ -143,6 +143,38 @@ export function getReportIdsFromRun(run) {
|
|
|
143
143
|
* @param reports - Array of reports for this run
|
|
144
144
|
* @returns Denormalized stats object (passed, failed, pending, total)
|
|
145
145
|
*/
|
|
146
|
+
/**
|
|
147
|
+
* Canonical "display" stats for a run detail/list view: prefer recomputing
|
|
148
|
+
* from the persisted per-test-case verdicts (`run.results`) via
|
|
149
|
+
* {@link bucketRunResults} — the single source of truth — and only fall
|
|
150
|
+
* back to the denormalized `run.stats` when `results` is empty (e.g. a run
|
|
151
|
+
* that hasn't started, or legacy data with no results map at all).
|
|
152
|
+
*
|
|
153
|
+
* This is THE single shared helper for pass/fail/errored/pending/total
|
|
154
|
+
* across every surface that renders run stats (runs list, benchmark runs
|
|
155
|
+
* list, run detail page, comparison page) — do not reimplement this logic
|
|
156
|
+
* locally in a component or add a differently-named twin. Historically this
|
|
157
|
+
* logic was duplicated per-page (each with its own subtly different bugs,
|
|
158
|
+
* e.g. `errored` not tracked, or `run.stats` trusted directly even when
|
|
159
|
+
* stale/inflated for trace-judged runs — a 66/84-real-passed run displaying
|
|
160
|
+
* as 84/84). Callers that only need `{passed, failed, errored, total}` can
|
|
161
|
+
* simply ignore the extra `pending` field.
|
|
162
|
+
*/
|
|
163
|
+
export function computeRunStats(run) {
|
|
164
|
+
if (run.results && Object.keys(run.results).length > 0) {
|
|
165
|
+
return bucketRunResults(run.results);
|
|
166
|
+
}
|
|
167
|
+
if (run.stats && (run.stats.total ?? 0) > 0) {
|
|
168
|
+
return {
|
|
169
|
+
passed: run.stats.passed ?? 0,
|
|
170
|
+
failed: run.stats.failed ?? 0,
|
|
171
|
+
errored: run.stats.errored ?? 0,
|
|
172
|
+
pending: run.stats.pending ?? 0,
|
|
173
|
+
total: run.stats.total ?? 0,
|
|
174
|
+
};
|
|
175
|
+
}
|
|
176
|
+
return { passed: 0, failed: 0, errored: 0, pending: 0, total: 0 };
|
|
177
|
+
}
|
|
146
178
|
export function computeRunStatsFromReports(run, reports) {
|
|
147
179
|
const reportsMap = {};
|
|
148
180
|
reports.forEach(report => {
|
|
@@ -1 +1 @@
|
|
|
1
|
-
{"version":3,"file":"runStats.js","sourceRoot":"","sources":["../../runStats.ts"],"names":[],"mappings":"AAAA;;;GAGG;AA8CH;;;;;;;;;;;;;;;;GAgBG;AACH,MAAM,UAAU,gBAAgB,CAC9B,OAAiF;IAEjF,IAAI,MAAM,GAAG,CAAC,EAAE,MAAM,GAAG,CAAC,EAAE,OAAO,GAAG,CAAC,EAAE,OAAO,GAAG,CAAC,EAAE,KAAK,GAAG,CAAC,CAAC;IAChE,KAAK,MAAM,CAAC,IAAI,MAAM,CAAC,MAAM,CAAC,OAAO,IAAI,EAAE,CAAC,EAAE,CAAC;QAC7C,KAAK,EAAE,CAAC;QACR,IAAI,CAAC,CAAC,MAAM,KAAK,SAAS,IAAI,CAAC,CAAC,MAAM,KAAK,SAAS,EAAE,CAAC;YAAC,OAAO,EAAE,CAAC;YAAC,SAAS;QAAC,CAAC;QAC9E,IAAI,CAAC,CAAC,MAAM,KAAK,QAAQ,IAAI,CAAC,CAAC,MAAM,KAAK,WAAW,EAAE,CAAC;YAAC,MAAM,EAAE,CAAC;YAAC,SAAS;QAAC,CAAC;QAC9E,IAAI,CAAC,CAAC,MAAM,KAAK,WAAW,EAAE,CAAC;YAC7B,IAAI,CAAC,CAAC,cAAc,KAAK,QAAQ;gBAAE,MAAM,EAAE,CAAC;iBACvC,IAAI,CAAC,CAAC,cAAc,KAAK,QAAQ;gBAAE,MAAM,EAAE,CAAC;;gBAC5C,OAAO,EAAE,CAAC,CAAC,qDAAqD;YACrE,SAAS;QACX,CAAC;QACD,OAAO,EAAE,CAAC,CAAC,sBAAsB;IACnC,CAAC;IACD,OAAO,EAAE,MAAM,EAAE,MAAM,EAAE,OAAO,EAAE,OAAO,EAAE,KAAK,EAAE,CAAC;AACrD,CAAC;AAED;;;;;;;;;;;GAWG;AACH,MAAM,UAAU,iBAAiB,CAC/B,GAAiB,EACjB,OAAgD;IAEhD,IAAI,MAAM,GAAG,CAAC,CAAC;IACf,IAAI,MAAM,GAAG,CAAC,CAAC;IACf,IAAI,OAAO,GAAG,CAAC,CAAC;IAChB,IAAI,OAAO,GAAG,CAAC,CAAC;IAChB,IAAI,KAAK,GAAG,CAAC,CAAC;IAEd,MAAM,CAAC,OAAO,CAAC,GAAG,CAAC,OAAO,IAAI,EAAE,CAAC,CAAC,OAAO,CAAC,CAAC,CAAC,UAAU,EAAE,MAAM,CAAC,EAAE,EAAE;QACjE,KAAK,EAAE,CAAC;QAER,4BAA4B;QAC5B,IAAI,MAAM,CAAC,MAAM,KAAK,SAAS,IAAI,MAAM,CAAC,MAAM,KAAK,SAAS,EAAE,CAAC;YAC/D,OAAO,EAAE,CAAC;YACV,OAAO;QACT,CAAC;QAED,IAAI,MAAM,CAAC,MAAM,KAAK,QAAQ,IAAI,MAAM,CAAC,MAAM,KAAK,WAAW,EAAE,CAAC;YAChE,MAAM,EAAE,CAAC;YACT,OAAO;QACT,CAAC;QAED,2DAA2D;QAC3D,IAAI,MAAM,CAAC,MAAM,KAAK,WAAW,IAAI,MAAM,CAAC,QAAQ,EAAE,CAAC;YACrD,MAAM,MAAM,GAAG,OAAO,CAAC,MAAM,CAAC,QAAQ,CAAC,CAAC;YAExC,IAAI,CAAC,MAAM,EAAE,CAAC;gBACZ,yCAAyC;gBACzC,OAAO,EAAE,CAAC;gBACV,OAAO;YACT,CAAC;YAED,oDAAoD;YACpD,IAAI,MAAM,CAAC,aAAa,KAAK,SAAS,IAAI,MAAM,CAAC,aAAa,KAAK,aAAa,EAAE,CAAC;gBACjF,OAAO,EAAE,CAAC;gBACV,OAAO;YACT,CAAC;YAED,sEAAsE;YACtE,8DAA8D;YAC9D,+DAA+D;YAC/D,IAAI,MAAM,CAAC,aAAa,KAAK,OAAO,EAAE,CAAC;gBACrC,OAAO,EAAE,CAAC;gBACV,OAAO;YACT,CAAC;YAED,+CAA+C;YAC/C,IAAI,MAAM,CAAC,cAAc,KAAK,QAAQ,EAAE,CAAC;gBACvC,MAAM,EAAE,CAAC;YACX,CAAC;iBAAM,CAAC;gBACN,6DAA6D;gBAC7D,MAAM,EAAE,CAAC;YACX,CAAC;QACH,CAAC;aAAM,CAAC;YACN,yDAAyD;YACzD,OAAO,EAAE,CAAC;QACZ,CAAC;IACH,CAAC,CAAC,CAAC;IAEH,sEAAsE;IACtE,gEAAgE;IAChE,MAAM,SAAS,GAAG,IAAI,CAAC,GAAG,CAAC,CAAC,EAAE,KAAK,GAAG,OAAO,CAAC,CAAC;IAC/C,MAAM,QAAQ,GAAG,SAAS,GAAG,CAAC,CAAC,CAAC,CAAC,IAAI,CAAC,KAAK,CAAC,CAAC,MAAM,GAAG,SAAS,CAAC,GAAG,GAAG,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC;IAE5E,OAAO;QACL,MAAM;QACN,MAAM;QACN,OAAO;QACP,OAAO;QACP,KAAK;QACL,QAAQ;KACT,CAAC;AACJ,CAAC;AAED;;;;;GAKG;AACH,MAAM,UAAU,mBAAmB,CAAC,GAAiB;IACnD,MAAM,SAAS,GAAG,IAAI,GAAG,EAAU,CAAC;IAEpC,MAAM,CAAC,MAAM,CAAC,GAAG,CAAC,OAAO,IAAI,EAAE,CAAC,CAAC,OAAO,CAAC,CAAC,MAAM,EAAE,EAAE;QAClD,IAAI,MAAM,CAAC,QAAQ,EAAE,CAAC;YACpB,SAAS,CAAC,GAAG,CAAC,MAAM,CAAC,QAAQ,CAAC,CAAC;QACjC,CAAC;IACH,CAAC,CAAC,CAAC;IAEH,OAAO,KAAK,CAAC,IAAI,CAAC,SAAS,CAAC,CAAC;AAC/B,CAAC;AAED;;;;;;;GAOG;AACH,MAAM,UAAU,0BAA0B,CACxC,GAAiB,EACjB,OAA2B;IAE3B,MAAM,UAAU,GAA4C,EAAE,CAAC;IAC/D,OAAO,CAAC,OAAO,CAAC,MAAM,CAAC,EAAE;QACvB,UAAU,CAAC,MAAM,CAAC,EAAE,CAAC,GAAG,MAAM,CAAC;IACjC,CAAC,CAAC,CAAC;IAEH,MAAM,SAAS,GAAG,iBAAiB,CAAC,GAAG,EAAE,UAAU,CAAC,CAAC;IAErD,OAAO;QACL,MAAM,EAAE,SAAS,CAAC,MAAM;QACxB,MAAM,EAAE,SAAS,CAAC,MAAM;QACxB,OAAO,EAAE,SAAS,CAAC,OAAO;QAC1B,OAAO,EAAE,SAAS,CAAC,OAAO;QAC1B,KAAK,EAAE,SAAS,CAAC,KAAK;KACvB,CAAC;AACJ,CAAC"}
|
|
1
|
+
{"version":3,"file":"runStats.js","sourceRoot":"","sources":["../../runStats.ts"],"names":[],"mappings":"AAAA;;;GAGG;AA8CH;;;;;;;;;;;;;;;;GAgBG;AACH,MAAM,UAAU,gBAAgB,CAC9B,OAAiF;IAEjF,IAAI,MAAM,GAAG,CAAC,EAAE,MAAM,GAAG,CAAC,EAAE,OAAO,GAAG,CAAC,EAAE,OAAO,GAAG,CAAC,EAAE,KAAK,GAAG,CAAC,CAAC;IAChE,KAAK,MAAM,CAAC,IAAI,MAAM,CAAC,MAAM,CAAC,OAAO,IAAI,EAAE,CAAC,EAAE,CAAC;QAC7C,KAAK,EAAE,CAAC;QACR,IAAI,CAAC,CAAC,MAAM,KAAK,SAAS,IAAI,CAAC,CAAC,MAAM,KAAK,SAAS,EAAE,CAAC;YAAC,OAAO,EAAE,CAAC;YAAC,SAAS;QAAC,CAAC;QAC9E,IAAI,CAAC,CAAC,MAAM,KAAK,QAAQ,IAAI,CAAC,CAAC,MAAM,KAAK,WAAW,EAAE,CAAC;YAAC,MAAM,EAAE,CAAC;YAAC,SAAS;QAAC,CAAC;QAC9E,IAAI,CAAC,CAAC,MAAM,KAAK,WAAW,EAAE,CAAC;YAC7B,IAAI,CAAC,CAAC,cAAc,KAAK,QAAQ;gBAAE,MAAM,EAAE,CAAC;iBACvC,IAAI,CAAC,CAAC,cAAc,KAAK,QAAQ;gBAAE,MAAM,EAAE,CAAC;;gBAC5C,OAAO,EAAE,CAAC,CAAC,qDAAqD;YACrE,SAAS;QACX,CAAC;QACD,OAAO,EAAE,CAAC,CAAC,sBAAsB;IACnC,CAAC;IACD,OAAO,EAAE,MAAM,EAAE,MAAM,EAAE,OAAO,EAAE,OAAO,EAAE,KAAK,EAAE,CAAC;AACrD,CAAC;AAED;;;;;;;;;;;GAWG;AACH,MAAM,UAAU,iBAAiB,CAC/B,GAAiB,EACjB,OAAgD;IAEhD,IAAI,MAAM,GAAG,CAAC,CAAC;IACf,IAAI,MAAM,GAAG,CAAC,CAAC;IACf,IAAI,OAAO,GAAG,CAAC,CAAC;IAChB,IAAI,OAAO,GAAG,CAAC,CAAC;IAChB,IAAI,KAAK,GAAG,CAAC,CAAC;IAEd,MAAM,CAAC,OAAO,CAAC,GAAG,CAAC,OAAO,IAAI,EAAE,CAAC,CAAC,OAAO,CAAC,CAAC,CAAC,UAAU,EAAE,MAAM,CAAC,EAAE,EAAE;QACjE,KAAK,EAAE,CAAC;QAER,4BAA4B;QAC5B,IAAI,MAAM,CAAC,MAAM,KAAK,SAAS,IAAI,MAAM,CAAC,MAAM,KAAK,SAAS,EAAE,CAAC;YAC/D,OAAO,EAAE,CAAC;YACV,OAAO;QACT,CAAC;QAED,IAAI,MAAM,CAAC,MAAM,KAAK,QAAQ,IAAI,MAAM,CAAC,MAAM,KAAK,WAAW,EAAE,CAAC;YAChE,MAAM,EAAE,CAAC;YACT,OAAO;QACT,CAAC;QAED,2DAA2D;QAC3D,IAAI,MAAM,CAAC,MAAM,KAAK,WAAW,IAAI,MAAM,CAAC,QAAQ,EAAE,CAAC;YACrD,MAAM,MAAM,GAAG,OAAO,CAAC,MAAM,CAAC,QAAQ,CAAC,CAAC;YAExC,IAAI,CAAC,MAAM,EAAE,CAAC;gBACZ,yCAAyC;gBACzC,OAAO,EAAE,CAAC;gBACV,OAAO;YACT,CAAC;YAED,oDAAoD;YACpD,IAAI,MAAM,CAAC,aAAa,KAAK,SAAS,IAAI,MAAM,CAAC,aAAa,KAAK,aAAa,EAAE,CAAC;gBACjF,OAAO,EAAE,CAAC;gBACV,OAAO;YACT,CAAC;YAED,sEAAsE;YACtE,8DAA8D;YAC9D,+DAA+D;YAC/D,IAAI,MAAM,CAAC,aAAa,KAAK,OAAO,EAAE,CAAC;gBACrC,OAAO,EAAE,CAAC;gBACV,OAAO;YACT,CAAC;YAED,+CAA+C;YAC/C,IAAI,MAAM,CAAC,cAAc,KAAK,QAAQ,EAAE,CAAC;gBACvC,MAAM,EAAE,CAAC;YACX,CAAC;iBAAM,CAAC;gBACN,6DAA6D;gBAC7D,MAAM,EAAE,CAAC;YACX,CAAC;QACH,CAAC;aAAM,CAAC;YACN,yDAAyD;YACzD,OAAO,EAAE,CAAC;QACZ,CAAC;IACH,CAAC,CAAC,CAAC;IAEH,sEAAsE;IACtE,gEAAgE;IAChE,MAAM,SAAS,GAAG,IAAI,CAAC,GAAG,CAAC,CAAC,EAAE,KAAK,GAAG,OAAO,CAAC,CAAC;IAC/C,MAAM,QAAQ,GAAG,SAAS,GAAG,CAAC,CAAC,CAAC,CAAC,IAAI,CAAC,KAAK,CAAC,CAAC,MAAM,GAAG,SAAS,CAAC,GAAG,GAAG,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC;IAE5E,OAAO;QACL,MAAM;QACN,MAAM;QACN,OAAO;QACP,OAAO;QACP,KAAK;QACL,QAAQ;KACT,CAAC;AACJ,CAAC;AAED;;;;;GAKG;AACH,MAAM,UAAU,mBAAmB,CAAC,GAAiB;IACnD,MAAM,SAAS,GAAG,IAAI,GAAG,EAAU,CAAC;IAEpC,MAAM,CAAC,MAAM,CAAC,GAAG,CAAC,OAAO,IAAI,EAAE,CAAC,CAAC,OAAO,CAAC,CAAC,MAAM,EAAE,EAAE;QAClD,IAAI,MAAM,CAAC,QAAQ,EAAE,CAAC;YACpB,SAAS,CAAC,GAAG,CAAC,MAAM,CAAC,QAAQ,CAAC,CAAC;QACjC,CAAC;IACH,CAAC,CAAC,CAAC;IAEH,OAAO,KAAK,CAAC,IAAI,CAAC,SAAS,CAAC,CAAC;AAC/B,CAAC;AAED;;;;;;;GAOG;AACH;;;;;;;;;;;;;;;;GAgBG;AACH,MAAM,UAAU,eAAe,CAC7B,GAGC;IAED,IAAI,GAAG,CAAC,OAAO,IAAI,MAAM,CAAC,IAAI,CAAC,GAAG,CAAC,OAAO,CAAC,CAAC,MAAM,GAAG,CAAC,EAAE,CAAC;QACvD,OAAO,gBAAgB,CAAC,GAAG,CAAC,OAAO,CAAC,CAAC;IACvC,CAAC;IACD,IAAI,GAAG,CAAC,KAAK,IAAI,CAAC,GAAG,CAAC,KAAK,CAAC,KAAK,IAAI,CAAC,CAAC,GAAG,CAAC,EAAE,CAAC;QAC5C,OAAO;YACL,MAAM,EAAE,GAAG,CAAC,KAAK,CAAC,MAAM,IAAI,CAAC;YAC7B,MAAM,EAAE,GAAG,CAAC,KAAK,CAAC,MAAM,IAAI,CAAC;YAC7B,OAAO,EAAE,GAAG,CAAC,KAAK,CAAC,OAAO,IAAI,CAAC;YAC/B,OAAO,EAAE,GAAG,CAAC,KAAK,CAAC,OAAO,IAAI,CAAC;YAC/B,KAAK,EAAE,GAAG,CAAC,KAAK,CAAC,KAAK,IAAI,CAAC;SAC5B,CAAC;IACJ,CAAC;IACD,OAAO,EAAE,MAAM,EAAE,CAAC,EAAE,MAAM,EAAE,CAAC,EAAE,OAAO,EAAE,CAAC,EAAE,OAAO,EAAE,CAAC,EAAE,KAAK,EAAE,CAAC,EAAE,CAAC;AACpE,CAAC;AAED,MAAM,UAAU,0BAA0B,CACxC,GAAiB,EACjB,OAA2B;IAE3B,MAAM,UAAU,GAA4C,EAAE,CAAC;IAC/D,OAAO,CAAC,OAAO,CAAC,MAAM,CAAC,EAAE;QACvB,UAAU,CAAC,MAAM,CAAC,EAAE,CAAC,GAAG,MAAM,CAAC;IACjC,CAAC,CAAC,CAAC;IAEH,MAAM,SAAS,GAAG,iBAAiB,CAAC,GAAG,EAAE,UAAU,CAAC,CAAC;IAErD,OAAO;QACL,MAAM,EAAE,SAAS,CAAC,MAAM;QACxB,MAAM,EAAE,SAAS,CAAC,MAAM;QACxB,OAAO,EAAE,SAAS,CAAC,OAAO;QAC1B,OAAO,EAAE,SAAS,CAAC,OAAO;QAC1B,KAAK,EAAE,SAAS,CAAC,KAAK;KACvB,CAAC;AACJ,CAAC"}
|
|
@@ -1,6 +1,7 @@
|
|
|
1
1
|
import type { CodeTestCase, RegisteredHook } from './types.js';
|
|
2
2
|
export declare function isCodeFile(filename: string): boolean;
|
|
3
|
-
export
|
|
3
|
+
export { detectSourceLanguage } from '../../lib/utils.js';
|
|
4
|
+
export declare function computeTestCaseHash(tc: CodeTestCase, fileSource?: string): string;
|
|
4
5
|
export interface LoadedTestCase extends CodeTestCase {
|
|
5
6
|
hash: string;
|
|
6
7
|
}
|
|
@@ -15,6 +16,15 @@ export interface LoadResult {
|
|
|
15
16
|
* (CLI / server names it after the filename).
|
|
16
17
|
*/
|
|
17
18
|
benchmarks: Map<string, string[]>;
|
|
19
|
+
/**
|
|
20
|
+
* Raw text of the eval file, read once up front (before execution) for
|
|
21
|
+
* BOTH .js and .ts/.mjs paths. Persisted verbatim as `sourceCode` on each
|
|
22
|
+
* imported TestCase (see cli/commands/benchmark.ts) so the Test Case
|
|
23
|
+
* detail page can render the whole file as an IDE-style code view --
|
|
24
|
+
* without this we'd only have the parsed `initialPrompt`/`expectedOutcomes`
|
|
25
|
+
* fields, not the `evaluate()` body that produced them.
|
|
26
|
+
*/
|
|
27
|
+
fileSource: string;
|
|
18
28
|
/**
|
|
19
29
|
* All lifecycle hooks (`beforeEach`/`afterEach`/`beforeAll`/`afterAll`)
|
|
20
30
|
* registered while loading this file. Empty when the file declares none.
|
|
@@ -1 +1 @@
|
|
|
1
|
-
{"version":3,"file":"loader.d.ts","sourceRoot":"","sources":["../../../testCases/loader.ts"],"names":[],"mappings":"AAUA,OAAO,KAAK,EAAE,YAAY,EAAE,cAAc,EAAE,MAAM,YAAY,CAAC;AAW/D,wBAAgB,UAAU,CAAC,QAAQ,EAAE,MAAM,GAAG,OAAO,CAGpD;
|
|
1
|
+
{"version":3,"file":"loader.d.ts","sourceRoot":"","sources":["../../../testCases/loader.ts"],"names":[],"mappings":"AAUA,OAAO,KAAK,EAAE,YAAY,EAAE,cAAc,EAAE,MAAM,YAAY,CAAC;AAW/D,wBAAgB,UAAU,CAAC,QAAQ,EAAE,MAAM,GAAG,OAAO,CAGpD;AAQD,OAAO,EAAE,oBAAoB,EAAE,MAAM,gBAAgB,CAAC;AAEtD,wBAAgB,mBAAmB,CAAC,EAAE,EAAE,YAAY,EAAE,UAAU,CAAC,EAAE,MAAM,GAAG,MAAM,CAyBjF;AAED,MAAM,WAAW,cAAe,SAAQ,YAAY;IAClD,IAAI,EAAE,MAAM,CAAC;CACd;AAED,MAAM,WAAW,UAAU;IACzB,SAAS,EAAE,cAAc,EAAE,CAAC;IAC5B,QAAQ,EAAE,MAAM,CAAC;IACjB;;;;;;OAMG;IACH,UAAU,EAAE,GAAG,CAAC,MAAM,EAAE,MAAM,EAAE,CAAC,CAAC;IAClC;;;;;;;OAOG;IACH,UAAU,EAAE,MAAM,CAAC;IACnB;;;;OAIG;IACH,KAAK,EAAE,cAAc,EAAE,CAAC;CACzB;AAED,wBAAsB,uBAAuB,CAAC,QAAQ,EAAE,MAAM,GAAG,OAAO,CAAC,UAAU,CAAC,CAiJnF"}
|