@cogitator-ai/evals 0.6.0 → 0.7.0

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 CHANGED
@@ -17,7 +17,7 @@ Website docs: [Evals](https://cogitator.app/docs/evals), [Datasets](https://cogi
17
17
  ## Features
18
18
 
19
19
  - **EvalSuite** — Run datasets against agents or plain functions with configurable concurrency, timeouts, and retries
20
- - **4 Deterministic Metrics** — exactMatch, contains, regex, jsonSchema (Zod)
20
+ - **5 Deterministic Metrics** — exactMatch, contains, similarity, regex, jsonSchema (Zod)
21
21
  - **5 LLM-as-Judge Metrics** — faithfulness, relevance, coherence, helpfulness, custom llmMetric
22
22
  - **3 Statistical Metrics** — latency, cost, tokenUsage with full percentile breakdowns
23
23
  - **Custom Metrics** — `metric()` factory for anything domain-specific
@@ -152,25 +152,37 @@ All transformations return new `Dataset` instances — the original is never mut
152
152
 
153
153
  ### Deterministic
154
154
 
155
- Binary (0 or 1) metrics that compare output against expected values.
155
+ Metrics that compare output against expected values using string operations or schema validation.
156
156
 
157
- | Metric | Description | Requires `expected` |
158
- | ------------ | ------------------------------------------ | ------------------- |
159
- | `exactMatch` | Exact string match (case optional) | Yes |
160
- | `contains` | Output contains expected substring | Yes |
161
- | `regex` | Output matches a regex pattern | No |
162
- | `jsonSchema` | Output is valid JSON matching a Zod schema | No |
157
+ | Metric | Description | Requires `expected` |
158
+ | ------------ | ---------------------------------------------- | ------------------- |
159
+ | `exactMatch` | Exact string match (case optional) | Yes |
160
+ | `contains` | Output contains expected substring | Yes |
161
+ | `similarity` | Levenshtein distance similarity score (0 to 1) | Yes |
162
+ | `regex` | Output matches a regex pattern | No |
163
+ | `jsonSchema` | Output is valid JSON matching a Zod schema | No |
163
164
 
164
165
  ```typescript
165
- import { exactMatch, contains, regex, jsonSchema } from '@cogitator-ai/evals';
166
+ import {
167
+ exactMatch,
168
+ contains,
169
+ similarity,
170
+ regex,
171
+ jsonSchema,
172
+ threshold,
173
+ } from '@cogitator-ai/evals';
166
174
  import { z } from 'zod';
167
175
 
168
176
  const metrics = [
169
177
  exactMatch({ caseSensitive: true }),
170
178
  contains(),
179
+ similarity(),
171
180
  regex(/\d{4}-\d{2}-\d{2}/),
172
181
  jsonSchema(z.object({ answer: z.string(), confidence: z.number() })),
173
182
  ];
183
+
184
+ // Assert similarity is at least 0.8
185
+ const assertions = [threshold('similarity', 0.8)];
174
186
  ```
175
187
 
176
188
  Every built-in per-case metric takes a `name` option. A suite refuses two metrics with the same name, because their scores would merge into one aggregate, so name the second of a kind:
@@ -278,6 +290,7 @@ import { threshold } from '@cogitator-ai/evals';
278
290
 
279
291
  const assertions = [
280
292
  threshold('exactMatch', 0.9),
293
+ threshold('similarity', 0.8),
281
294
  threshold('latency', 5000),
282
295
  threshold('relevance', 0.7),
283
296
  ];
@@ -463,21 +476,22 @@ result.report('ci');
463
476
 
464
477
  ### Metrics
465
478
 
466
- | Export | Type | Description |
467
- | -------------- | ------------- | ----------------------------------- |
468
- | `exactMatch` | Deterministic | Exact string match |
469
- | `contains` | Deterministic | Substring match |
470
- | `regex` | Deterministic | Regex pattern match |
471
- | `jsonSchema` | Deterministic | Zod schema validation |
472
- | `faithfulness` | LLM Judge | Factual accuracy |
473
- | `relevance` | LLM Judge | Topical relevance |
474
- | `coherence` | LLM Judge | Logical structure |
475
- | `helpfulness` | LLM Judge | Practical usefulness |
476
- | `llmMetric` | LLM Judge | Custom judge prompt |
477
- | `latency` | Statistical | Response time percentiles |
478
- | `cost` | Statistical | Token cost aggregation |
479
- | `tokenUsage` | Statistical | Input/output token counts |
480
- | `metric` | Custom | Factory for domain-specific metrics |
479
+ | Export | Type | Description |
480
+ | -------------- | ------------- | ------------------------------------- |
481
+ | `exactMatch` | Deterministic | Exact string match |
482
+ | `contains` | Deterministic | Substring match |
483
+ | `similarity` | Deterministic | Levenshtein distance similarity score |
484
+ | `regex` | Deterministic | Regex pattern match |
485
+ | `jsonSchema` | Deterministic | Zod schema validation |
486
+ | `faithfulness` | LLM Judge | Factual accuracy |
487
+ | `relevance` | LLM Judge | Topical relevance |
488
+ | `coherence` | LLM Judge | Logical structure |
489
+ | `helpfulness` | LLM Judge | Practical usefulness |
490
+ | `llmMetric` | LLM Judge | Custom judge prompt |
491
+ | `latency` | Statistical | Response time percentiles |
492
+ | `cost` | Statistical | Token cost aggregation |
493
+ | `tokenUsage` | Statistical | Input/output token counts |
494
+ | `metric` | Custom | Factory for domain-specific metrics |
481
495
 
482
496
  ### Assertions
483
497
 
package/dist/index.d.ts CHANGED
@@ -7,7 +7,7 @@ export { EvalBuilder } from './eval-builder.js';
7
7
  export { Dataset } from './datasets/index.js';
8
8
  export { loadJsonl } from './datasets/index.js';
9
9
  export { loadCsv } from './datasets/index.js';
10
- export { exactMatch, contains, regex, jsonSchema } from './metrics/deterministic.js';
10
+ export { exactMatch, contains, similarity, regex, jsonSchema } from './metrics/deterministic.js';
11
11
  export { faithfulness, relevance, coherence, helpfulness, llmMetric, bindJudgeContext, } from './metrics/llm-judge.js';
12
12
  export type { LLMMetricFn, JudgeContext } from './metrics/llm-judge.js';
13
13
  export { latency, cost, tokenUsage } from './metrics/statistical.js';
@@ -1 +1 @@
1
- {"version":3,"file":"index.d.ts","sourceRoot":"","sources":["../src/index.ts"],"names":[],"mappings":"AAIA,eAAO,MAAM,OAAO,EAAE,MAA4B,CAAC;AAEnD,OAAO,EAAE,SAAS,EAAE,WAAW,EAAE,MAAM,cAAc,CAAC;AACtD,YAAY,EACV,UAAU,EACV,iBAAiB,EACjB,YAAY,EACZ,cAAc,EACd,gBAAgB,EAChB,eAAe,EACf,cAAc,GACf,MAAM,cAAc,CAAC;AACtB,OAAO,EAAE,cAAc,EAAE,MAAM,mBAAmB,CAAC;AACnD,YAAY,EACV,qBAAqB,EACrB,wBAAwB,EACxB,gBAAgB,EAChB,gBAAgB,GACjB,MAAM,mBAAmB,CAAC;AAC3B,OAAO,EAAE,WAAW,EAAE,MAAM,gBAAgB,CAAC;AAE7C,OAAO,EAAE,OAAO,EAAE,MAAM,YAAY,CAAC;AACrC,OAAO,EAAE,SAAS,EAAE,MAAM,YAAY,CAAC;AACvC,OAAO,EAAE,OAAO,EAAE,MAAM,YAAY,CAAC;AAErC,OAAO,EAAE,UAAU,EAAE,QAAQ,EAAE,KAAK,EAAE,UAAU,EAAE,MAAM,yBAAyB,CAAC;AAClF,OAAO,EACL,YAAY,EACZ,SAAS,EACT,SAAS,EACT,WAAW,EACX,SAAS,EACT,gBAAgB,GACjB,MAAM,qBAAqB,CAAC;AAC7B,YAAY,EAAE,WAAW,EAAE,YAAY,EAAE,MAAM,qBAAqB,CAAC;AACrE,OAAO,EAAE,OAAO,EAAE,IAAI,EAAE,UAAU,EAAE,MAAM,uBAAuB,CAAC;AAClE,OAAO,EAAE,MAAM,EAAE,MAAM,kBAAkB,CAAC;AAC1C,YAAY,EACV,QAAQ,EACR,WAAW,EACX,WAAW,EACX,aAAa,EACb,cAAc,EACd,mBAAmB,EACnB,gBAAgB,GACjB,MAAM,iBAAiB,CAAC;AACzB,YAAY,EAAE,kBAAkB,EAAE,MAAM,kBAAkB,CAAC;AAE3D,OAAO,EAAE,SAAS,EAAE,MAAM,cAAc,CAAC;AACzC,OAAO,EAAE,YAAY,EAAE,MAAM,cAAc,CAAC;AAC5C,OAAO,EAAE,SAAS,EAAE,MAAM,cAAc,CAAC;AACzC,YAAY,EAAE,WAAW,EAAE,eAAe,EAAE,gBAAgB,EAAE,SAAS,EAAE,MAAM,cAAc,CAAC;AAE9F,OAAO,EAAE,MAAM,EAAE,MAAM,aAAa,CAAC;AACrC,YAAY,EAAE,YAAY,EAAE,eAAe,EAAE,MAAM,aAAa,CAAC;AAEjE,OAAO,EAAE,WAAW,EAAE,MAAM,gBAAgB,CAAC;AAC7C,YAAY,EAAE,WAAW,EAAE,MAAM,gBAAgB,CAAC;AAClD,OAAO,EAAE,YAAY,EAAE,MAAM,iBAAiB,CAAC;AAC/C,YAAY,EAAE,aAAa,EAAE,MAAM,iBAAiB,CAAC;AACrD,OAAO,EAAE,IAAI,EAAE,MAAM,EAAE,MAAM,EAAE,UAAU,EAAE,SAAS,EAAE,MAAM,qBAAqB,CAAC;AAElF,YAAY,EACV,QAAQ,EACR,eAAe,EACf,WAAW,EACX,gBAAgB,EAChB,oBAAoB,GACrB,MAAM,UAAU,CAAC;AAElB,OAAO,EAAE,iBAAiB,EAAE,SAAS,EAAE,MAAM,SAAS,CAAC;AACvD,YAAY,EAAE,QAAQ,EAAE,MAAM,SAAS,CAAC"}
1
+ {"version":3,"file":"index.d.ts","sourceRoot":"","sources":["../src/index.ts"],"names":[],"mappings":"AAIA,eAAO,MAAM,OAAO,EAAE,MAA4B,CAAC;AAEnD,OAAO,EAAE,SAAS,EAAE,WAAW,EAAE,MAAM,cAAc,CAAC;AACtD,YAAY,EACV,UAAU,EACV,iBAAiB,EACjB,YAAY,EACZ,cAAc,EACd,gBAAgB,EAChB,eAAe,EACf,cAAc,GACf,MAAM,cAAc,CAAC;AACtB,OAAO,EAAE,cAAc,EAAE,MAAM,mBAAmB,CAAC;AACnD,YAAY,EACV,qBAAqB,EACrB,wBAAwB,EACxB,gBAAgB,EAChB,gBAAgB,GACjB,MAAM,mBAAmB,CAAC;AAC3B,OAAO,EAAE,WAAW,EAAE,MAAM,gBAAgB,CAAC;AAE7C,OAAO,EAAE,OAAO,EAAE,MAAM,YAAY,CAAC;AACrC,OAAO,EAAE,SAAS,EAAE,MAAM,YAAY,CAAC;AACvC,OAAO,EAAE,OAAO,EAAE,MAAM,YAAY,CAAC;AAErC,OAAO,EAAE,UAAU,EAAE,QAAQ,EAAE,UAAU,EAAE,KAAK,EAAE,UAAU,EAAE,MAAM,yBAAyB,CAAC;AAC9F,OAAO,EACL,YAAY,EACZ,SAAS,EACT,SAAS,EACT,WAAW,EACX,SAAS,EACT,gBAAgB,GACjB,MAAM,qBAAqB,CAAC;AAC7B,YAAY,EAAE,WAAW,EAAE,YAAY,EAAE,MAAM,qBAAqB,CAAC;AACrE,OAAO,EAAE,OAAO,EAAE,IAAI,EAAE,UAAU,EAAE,MAAM,uBAAuB,CAAC;AAClE,OAAO,EAAE,MAAM,EAAE,MAAM,kBAAkB,CAAC;AAC1C,YAAY,EACV,QAAQ,EACR,WAAW,EACX,WAAW,EACX,aAAa,EACb,cAAc,EACd,mBAAmB,EACnB,gBAAgB,GACjB,MAAM,iBAAiB,CAAC;AACzB,YAAY,EAAE,kBAAkB,EAAE,MAAM,kBAAkB,CAAC;AAE3D,OAAO,EAAE,SAAS,EAAE,MAAM,cAAc,CAAC;AACzC,OAAO,EAAE,YAAY,EAAE,MAAM,cAAc,CAAC;AAC5C,OAAO,EAAE,SAAS,EAAE,MAAM,cAAc,CAAC;AACzC,YAAY,EAAE,WAAW,EAAE,eAAe,EAAE,gBAAgB,EAAE,SAAS,EAAE,MAAM,cAAc,CAAC;AAE9F,OAAO,EAAE,MAAM,EAAE,MAAM,aAAa,CAAC;AACrC,YAAY,EAAE,YAAY,EAAE,eAAe,EAAE,MAAM,aAAa,CAAC;AAEjE,OAAO,EAAE,WAAW,EAAE,MAAM,gBAAgB,CAAC;AAC7C,YAAY,EAAE,WAAW,EAAE,MAAM,gBAAgB,CAAC;AAClD,OAAO,EAAE,YAAY,EAAE,MAAM,iBAAiB,CAAC;AAC/C,YAAY,EAAE,aAAa,EAAE,MAAM,iBAAiB,CAAC;AACrD,OAAO,EAAE,IAAI,EAAE,MAAM,EAAE,MAAM,EAAE,UAAU,EAAE,SAAS,EAAE,MAAM,qBAAqB,CAAC;AAElF,YAAY,EACV,QAAQ,EACR,eAAe,EACf,WAAW,EACX,gBAAgB,EAChB,oBAAoB,GACrB,MAAM,UAAU,CAAC;AAElB,OAAO,EAAE,iBAAiB,EAAE,SAAS,EAAE,MAAM,SAAS,CAAC;AACvD,YAAY,EAAE,QAAQ,EAAE,MAAM,SAAS,CAAC"}
package/dist/index.js CHANGED
@@ -7,7 +7,7 @@ export { EvalBuilder } from './eval-builder.js';
7
7
  export { Dataset } from './datasets/index.js';
8
8
  export { loadJsonl } from './datasets/index.js';
9
9
  export { loadCsv } from './datasets/index.js';
10
- export { exactMatch, contains, regex, jsonSchema } from './metrics/deterministic.js';
10
+ export { exactMatch, contains, similarity, regex, jsonSchema } from './metrics/deterministic.js';
11
11
  export { faithfulness, relevance, coherence, helpfulness, llmMetric, bindJudgeContext, } from './metrics/llm-judge.js';
12
12
  export { latency, cost, tokenUsage } from './metrics/statistical.js';
13
13
  export { metric } from './metrics/custom.js';
package/dist/index.js.map CHANGED
@@ -1 +1 @@
1
- {"version":3,"file":"index.js","sourceRoot":"","sources":["../src/index.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,aAAa,EAAE,MAAM,aAAa,CAAC;AAE5C,MAAM,WAAW,GAAG,aAAa,CAAC,MAAM,CAAC,IAAI,CAAC,GAAG,CAAC,CAAC,iBAAiB,CAAwB,CAAC;AAE7F,MAAM,CAAC,MAAM,OAAO,GAAW,WAAW,CAAC,OAAO,CAAC;AAEnD,OAAO,EAAE,SAAS,EAAE,WAAW,EAAE,MAAM,cAAc,CAAC;AAUtD,OAAO,EAAE,cAAc,EAAE,MAAM,mBAAmB,CAAC;AAOnD,OAAO,EAAE,WAAW,EAAE,MAAM,gBAAgB,CAAC;AAE7C,OAAO,EAAE,OAAO,EAAE,MAAM,YAAY,CAAC;AACrC,OAAO,EAAE,SAAS,EAAE,MAAM,YAAY,CAAC;AACvC,OAAO,EAAE,OAAO,EAAE,MAAM,YAAY,CAAC;AAErC,OAAO,EAAE,UAAU,EAAE,QAAQ,EAAE,KAAK,EAAE,UAAU,EAAE,MAAM,yBAAyB,CAAC;AAClF,OAAO,EACL,YAAY,EACZ,SAAS,EACT,SAAS,EACT,WAAW,EACX,SAAS,EACT,gBAAgB,GACjB,MAAM,qBAAqB,CAAC;AAE7B,OAAO,EAAE,OAAO,EAAE,IAAI,EAAE,UAAU,EAAE,MAAM,uBAAuB,CAAC;AAClE,OAAO,EAAE,MAAM,EAAE,MAAM,kBAAkB,CAAC;AAY1C,OAAO,EAAE,SAAS,EAAE,MAAM,cAAc,CAAC;AACzC,OAAO,EAAE,YAAY,EAAE,MAAM,cAAc,CAAC;AAC5C,OAAO,EAAE,SAAS,EAAE,MAAM,cAAc,CAAC;AAGzC,OAAO,EAAE,MAAM,EAAE,MAAM,aAAa,CAAC;AAGrC,OAAO,EAAE,WAAW,EAAE,MAAM,gBAAgB,CAAC;AAE7C,OAAO,EAAE,YAAY,EAAE,MAAM,iBAAiB,CAAC;AAE/C,OAAO,EAAE,IAAI,EAAE,MAAM,EAAE,MAAM,EAAE,UAAU,EAAE,SAAS,EAAE,MAAM,qBAAqB,CAAC;AAUlF,OAAO,EAAE,iBAAiB,EAAE,SAAS,EAAE,MAAM,SAAS,CAAC"}
1
+ {"version":3,"file":"index.js","sourceRoot":"","sources":["../src/index.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,aAAa,EAAE,MAAM,aAAa,CAAC;AAE5C,MAAM,WAAW,GAAG,aAAa,CAAC,MAAM,CAAC,IAAI,CAAC,GAAG,CAAC,CAAC,iBAAiB,CAAwB,CAAC;AAE7F,MAAM,CAAC,MAAM,OAAO,GAAW,WAAW,CAAC,OAAO,CAAC;AAEnD,OAAO,EAAE,SAAS,EAAE,WAAW,EAAE,MAAM,cAAc,CAAC;AAUtD,OAAO,EAAE,cAAc,EAAE,MAAM,mBAAmB,CAAC;AAOnD,OAAO,EAAE,WAAW,EAAE,MAAM,gBAAgB,CAAC;AAE7C,OAAO,EAAE,OAAO,EAAE,MAAM,YAAY,CAAC;AACrC,OAAO,EAAE,SAAS,EAAE,MAAM,YAAY,CAAC;AACvC,OAAO,EAAE,OAAO,EAAE,MAAM,YAAY,CAAC;AAErC,OAAO,EAAE,UAAU,EAAE,QAAQ,EAAE,UAAU,EAAE,KAAK,EAAE,UAAU,EAAE,MAAM,yBAAyB,CAAC;AAC9F,OAAO,EACL,YAAY,EACZ,SAAS,EACT,SAAS,EACT,WAAW,EACX,SAAS,EACT,gBAAgB,GACjB,MAAM,qBAAqB,CAAC;AAE7B,OAAO,EAAE,OAAO,EAAE,IAAI,EAAE,UAAU,EAAE,MAAM,uBAAuB,CAAC;AAClE,OAAO,EAAE,MAAM,EAAE,MAAM,kBAAkB,CAAC;AAY1C,OAAO,EAAE,SAAS,EAAE,MAAM,cAAc,CAAC;AACzC,OAAO,EAAE,YAAY,EAAE,MAAM,cAAc,CAAC;AAC5C,OAAO,EAAE,SAAS,EAAE,MAAM,cAAc,CAAC;AAGzC,OAAO,EAAE,MAAM,EAAE,MAAM,aAAa,CAAC;AAGrC,OAAO,EAAE,WAAW,EAAE,MAAM,gBAAgB,CAAC;AAE7C,OAAO,EAAE,YAAY,EAAE,MAAM,iBAAiB,CAAC;AAE/C,OAAO,EAAE,IAAI,EAAE,MAAM,EAAE,MAAM,EAAE,UAAU,EAAE,SAAS,EAAE,MAAM,qBAAqB,CAAC;AAUlF,OAAO,EAAE,iBAAiB,EAAE,SAAS,EAAE,MAAM,SAAS,CAAC"}
@@ -4,6 +4,7 @@ interface MatchOptions extends MetricOptions {
4
4
  caseSensitive?: boolean;
5
5
  }
6
6
  export declare function exactMatch(opts?: MatchOptions): MetricFn;
7
+ export declare function similarity(opts?: MatchOptions): MetricFn;
7
8
  export declare function contains(opts?: MatchOptions): MetricFn;
8
9
  /**
9
10
  * Scores 1 when the output matches `pattern`. The `g` and `y` flags are dropped: they make
@@ -1 +1 @@
1
- {"version":3,"file":"deterministic.d.ts","sourceRoot":"","sources":["../../src/metrics/deterministic.ts"],"names":[],"mappings":"AAAA,OAAO,KAAK,EAAE,OAAO,EAAE,MAAM,KAAK,CAAC;AACnC,OAAO,KAAK,EAAE,QAAQ,EAAE,aAAa,EAAE,MAAM,SAAS,CAAC;AAEvD,UAAU,YAAa,SAAQ,aAAa;IAC1C,aAAa,CAAC,EAAE,OAAO,CAAC;CACzB;AAOD,wBAAgB,UAAU,CAAC,IAAI,CAAC,EAAE,YAAY,GAAG,QAAQ,CAoBxD;AAED,wBAAgB,QAAQ,CAAC,IAAI,CAAC,EAAE,YAAY,GAAG,QAAQ,CAoBtD;AAED;;;;GAIG;AACH,wBAAgB,KAAK,CAAC,OAAO,EAAE,MAAM,GAAG,MAAM,EAAE,IAAI,CAAC,EAAE,aAAa,GAAG,QAAQ,CAa9E;AAED,wBAAgB,UAAU,CAAC,MAAM,EAAE,OAAO,EAAE,IAAI,CAAC,EAAE,aAAa,GAAG,QAAQ,CA0B1E"}
1
+ {"version":3,"file":"deterministic.d.ts","sourceRoot":"","sources":["../../src/metrics/deterministic.ts"],"names":[],"mappings":"AAAA,OAAO,KAAK,EAAE,OAAO,EAAE,MAAM,KAAK,CAAC;AACnC,OAAO,KAAK,EAAE,QAAQ,EAAE,aAAa,EAAE,MAAM,SAAS,CAAC;AAEvD,UAAU,YAAa,SAAQ,aAAa;IAC1C,aAAa,CAAC,EAAE,OAAO,CAAC;CACzB;AAOD,wBAAgB,UAAU,CAAC,IAAI,CAAC,EAAE,YAAY,GAAG,QAAQ,CAoBxD;AAkCD,wBAAgB,UAAU,CAAC,IAAI,CAAC,EAAE,YAAY,GAAG,QAAQ,CAkCxD;AAED,wBAAgB,QAAQ,CAAC,IAAI,CAAC,EAAE,YAAY,GAAG,QAAQ,CAoBtD;AAED;;;;GAIG;AACH,wBAAgB,KAAK,CAAC,OAAO,EAAE,MAAM,GAAG,MAAM,EAAE,IAAI,CAAC,EAAE,aAAa,GAAG,QAAQ,CAa9E;AAED,wBAAgB,UAAU,CAAC,MAAM,EAAE,OAAO,EAAE,IAAI,CAAC,EAAE,aAAa,GAAG,QAAQ,CA0B1E"}
@@ -20,6 +20,61 @@ export function exactMatch(opts) {
20
20
  };
21
21
  }));
22
22
  }
23
+ function levenshteinDistance(s1, s2) {
24
+ const m = s1.length;
25
+ const n = s2.length;
26
+ if (m === 0)
27
+ return n;
28
+ if (n === 0)
29
+ return m;
30
+ let prev = new Array(n + 1);
31
+ let curr = new Array(n + 1);
32
+ for (let j = 0; j <= n; j++) {
33
+ prev[j] = j;
34
+ }
35
+ for (let i = 1; i <= m; i++) {
36
+ curr[0] = i;
37
+ const char1 = s1[i - 1];
38
+ for (let j = 1; j <= n; j++) {
39
+ const cost = char1 === s2[j - 1] ? 0 : 1;
40
+ curr[j] = Math.min(prev[j] + 1, // deletion
41
+ curr[j - 1] + 1, // insertion
42
+ prev[j - 1] + cost // substitution
43
+ );
44
+ }
45
+ const temp = prev;
46
+ prev = curr;
47
+ curr = temp;
48
+ }
49
+ return prev[n];
50
+ }
51
+ export function similarity(opts) {
52
+ const caseSensitive = opts?.caseSensitive ?? false;
53
+ const name = opts?.name ?? 'similarity';
54
+ return createMetricFn(name, (async (result) => {
55
+ const expected = result.case.expected;
56
+ if (expected === undefined) {
57
+ return { name, score: 0, details: 'no expected value provided' };
58
+ }
59
+ const output = result.output.trim();
60
+ const target = expected.trim();
61
+ const s1 = caseSensitive ? output : output.toLowerCase();
62
+ const s2 = caseSensitive ? target : target.toLowerCase();
63
+ const outChars = [...s1];
64
+ const expChars = [...s2];
65
+ const maxLen = Math.max(outChars.length, expChars.length);
66
+ if (maxLen === 0) {
67
+ return { name, score: 1 };
68
+ }
69
+ const distance = levenshteinDistance(outChars, expChars);
70
+ const score = 1 - distance / maxLen;
71
+ return {
72
+ name,
73
+ score,
74
+ details: score < 1 ? `expected "${target}", got "${output}" (distance: ${distance})` : undefined,
75
+ };
76
+ }));
77
+ }
23
78
  export function contains(opts) {
24
79
  const caseSensitive = opts?.caseSensitive ?? false;
25
80
  const name = opts?.name ?? 'contains';
@@ -1 +1 @@
1
- {"version":3,"file":"deterministic.js","sourceRoot":"","sources":["../../src/metrics/deterministic.ts"],"names":[],"mappings":"AAOA,SAAS,cAAc,CAAC,IAAY,EAAE,EAAY;IAChD,EAAE,CAAC,UAAU,GAAG,IAAI,CAAC;IACrB,OAAO,EAAE,CAAC;AACZ,CAAC;AAED,MAAM,UAAU,UAAU,CAAC,IAAmB;IAC5C,MAAM,aAAa,GAAG,IAAI,EAAE,aAAa,IAAI,KAAK,CAAC;IACnD,MAAM,IAAI,GAAG,IAAI,EAAE,IAAI,IAAI,YAAY,CAAC;IAExC,OAAO,cAAc,CAAC,IAAI,EAAE,CAAC,KAAK,EAAE,MAAM,EAAE,EAAE;QAC5C,MAAM,QAAQ,GAAG,MAAM,CAAC,IAAI,CAAC,QAAQ,CAAC;QACtC,IAAI,QAAQ,KAAK,SAAS,EAAE,CAAC;YAC3B,OAAO,EAAE,IAAI,EAAE,KAAK,EAAE,CAAC,EAAE,OAAO,EAAE,4BAA4B,EAAE,CAAC;QACnE,CAAC;QAED,MAAM,MAAM,GAAG,MAAM,CAAC,MAAM,CAAC,IAAI,EAAE,CAAC;QACpC,MAAM,MAAM,GAAG,QAAQ,CAAC,IAAI,EAAE,CAAC;QAC/B,MAAM,KAAK,GAAG,aAAa,CAAC,CAAC,CAAC,MAAM,KAAK,MAAM,CAAC,CAAC,CAAC,MAAM,CAAC,WAAW,EAAE,KAAK,MAAM,CAAC,WAAW,EAAE,CAAC;QAEhG,OAAO;YACL,IAAI;YACJ,KAAK,EAAE,KAAK,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC;YACpB,OAAO,EAAE,KAAK,CAAC,CAAC,CAAC,SAAS,CAAC,CAAC,CAAC,aAAa,MAAM,WAAW,MAAM,GAAG;SACrE,CAAC;IACJ,CAAC,CAAa,CAAC,CAAC;AAClB,CAAC;AAED,MAAM,UAAU,QAAQ,CAAC,IAAmB;IAC1C,MAAM,aAAa,GAAG,IAAI,EAAE,aAAa,IAAI,KAAK,CAAC;IACnD,MAAM,IAAI,GAAG,IAAI,EAAE,IAAI,IAAI,UAAU,CAAC;IAEtC,OAAO,cAAc,CAAC,IAAI,EAAE,CAAC,KAAK,EAAE,MAAM,EAAE,EAAE;QAC5C,MAAM,QAAQ,GAAG,MAAM,CAAC,IAAI,CAAC,QAAQ,CAAC;QACtC,IAAI,QAAQ,KAAK,SAAS,EAAE,CAAC;YAC3B,OAAO,EAAE,IAAI,EAAE,KAAK,EAAE,CAAC,EAAE,OAAO,EAAE,4BAA4B,EAAE,CAAC;QACnE,CAAC;QAED,MAAM,MAAM,GAAG,aAAa,CAAC,CAAC,CAAC,MAAM,CAAC,MAAM,CAAC,CAAC,CAAC,MAAM,CAAC,MAAM,CAAC,WAAW,EAAE,CAAC;QAC3E,MAAM,MAAM,GAAG,aAAa,CAAC,CAAC,CAAC,QAAQ,CAAC,CAAC,CAAC,QAAQ,CAAC,WAAW,EAAE,CAAC;QACjE,MAAM,KAAK,GAAG,MAAM,CAAC,QAAQ,CAAC,MAAM,CAAC,CAAC;QAEtC,OAAO;YACL,IAAI;YACJ,KAAK,EAAE,KAAK,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC;YACpB,OAAO,EAAE,KAAK,CAAC,CAAC,CAAC,SAAS,CAAC,CAAC,CAAC,4BAA4B,QAAQ,GAAG;SACrE,CAAC;IACJ,CAAC,CAAa,CAAC,CAAC;AAClB,CAAC;AAED;;;;GAIG;AACH,MAAM,UAAU,KAAK,CAAC,OAAwB,EAAE,IAAoB;IAClE,MAAM,MAAM,GAAG,OAAO,OAAO,KAAK,QAAQ,CAAC,CAAC,CAAC,IAAI,MAAM,CAAC,OAAO,CAAC,CAAC,CAAC,CAAC,OAAO,CAAC;IAC3E,MAAM,EAAE,GAAG,IAAI,MAAM,CAAC,MAAM,CAAC,MAAM,EAAE,MAAM,CAAC,KAAK,CAAC,OAAO,CAAC,OAAO,EAAE,EAAE,CAAC,CAAC,CAAC;IACxE,MAAM,IAAI,GAAG,IAAI,EAAE,IAAI,IAAI,OAAO,CAAC;IAEnC,OAAO,cAAc,CAAC,IAAI,EAAE,CAAC,KAAK,EAAE,MAAM,EAAE,EAAE;QAC5C,MAAM,KAAK,GAAG,EAAE,CAAC,IAAI,CAAC,MAAM,CAAC,MAAM,CAAC,CAAC;QACrC,OAAO;YACL,IAAI;YACJ,KAAK,EAAE,KAAK,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC;YACpB,OAAO,EAAE,KAAK,CAAC,CAAC,CAAC,SAAS,CAAC,CAAC,CAAC,iCAAiC,MAAM,EAAE;SACvE,CAAC;IACJ,CAAC,CAAa,CAAC,CAAC;AAClB,CAAC;AAED,MAAM,UAAU,UAAU,CAAC,MAAe,EAAE,IAAoB;IAC9D,MAAM,IAAI,GAAG,IAAI,EAAE,IAAI,IAAI,YAAY,CAAC;IAExC,OAAO,cAAc,CAAC,IAAI,EAAE,CAAC,KAAK,EAAE,MAAM,EAAE,EAAE;QAC5C,IAAI,MAAe,CAAC;QACpB,IAAI,CAAC;YACH,MAAM,GAAG,IAAI,CAAC,KAAK,CAAC,MAAM,CAAC,MAAM,CAAC,CAAC;QACrC,CAAC;QAAC,OAAO,CAAC,EAAE,CAAC;YACX,OAAO;gBACL,IAAI;gBACJ,KAAK,EAAE,CAAC;gBACR,OAAO,EAAE,iBAAkB,CAAW,CAAC,OAAO,EAAE;aACjD,CAAC;QACJ,CAAC;QAED,MAAM,UAAU,GAAG,MAAM,CAAC,SAAS,CAAC,MAAM,CAAC,CAAC;QAC5C,IAAI,UAAU,CAAC,OAAO,EAAE,CAAC;YACvB,OAAO,EAAE,IAAI,EAAE,KAAK,EAAE,CAAC,EAAE,CAAC;QAC5B,CAAC;QAED,OAAO;YACL,IAAI;YACJ,KAAK,EAAE,CAAC;YACR,OAAO,EAAE,6BAA6B,UAAU,CAAC,KAAK,CAAC,OAAO,EAAE;SACjE,CAAC;IACJ,CAAC,CAAa,CAAC,CAAC;AAClB,CAAC"}
1
+ {"version":3,"file":"deterministic.js","sourceRoot":"","sources":["../../src/metrics/deterministic.ts"],"names":[],"mappings":"AAOA,SAAS,cAAc,CAAC,IAAY,EAAE,EAAY;IAChD,EAAE,CAAC,UAAU,GAAG,IAAI,CAAC;IACrB,OAAO,EAAE,CAAC;AACZ,CAAC;AAED,MAAM,UAAU,UAAU,CAAC,IAAmB;IAC5C,MAAM,aAAa,GAAG,IAAI,EAAE,aAAa,IAAI,KAAK,CAAC;IACnD,MAAM,IAAI,GAAG,IAAI,EAAE,IAAI,IAAI,YAAY,CAAC;IAExC,OAAO,cAAc,CAAC,IAAI,EAAE,CAAC,KAAK,EAAE,MAAM,EAAE,EAAE;QAC5C,MAAM,QAAQ,GAAG,MAAM,CAAC,IAAI,CAAC,QAAQ,CAAC;QACtC,IAAI,QAAQ,KAAK,SAAS,EAAE,CAAC;YAC3B,OAAO,EAAE,IAAI,EAAE,KAAK,EAAE,CAAC,EAAE,OAAO,EAAE,4BAA4B,EAAE,CAAC;QACnE,CAAC;QAED,MAAM,MAAM,GAAG,MAAM,CAAC,MAAM,CAAC,IAAI,EAAE,CAAC;QACpC,MAAM,MAAM,GAAG,QAAQ,CAAC,IAAI,EAAE,CAAC;QAC/B,MAAM,KAAK,GAAG,aAAa,CAAC,CAAC,CAAC,MAAM,KAAK,MAAM,CAAC,CAAC,CAAC,MAAM,CAAC,WAAW,EAAE,KAAK,MAAM,CAAC,WAAW,EAAE,CAAC;QAEhG,OAAO;YACL,IAAI;YACJ,KAAK,EAAE,KAAK,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC;YACpB,OAAO,EAAE,KAAK,CAAC,CAAC,CAAC,SAAS,CAAC,CAAC,CAAC,aAAa,MAAM,WAAW,MAAM,GAAG;SACrE,CAAC;IACJ,CAAC,CAAa,CAAC,CAAC;AAClB,CAAC;AAED,SAAS,mBAAmB,CAAC,EAAqB,EAAE,EAAqB;IACvE,MAAM,CAAC,GAAG,EAAE,CAAC,MAAM,CAAC;IACpB,MAAM,CAAC,GAAG,EAAE,CAAC,MAAM,CAAC;IACpB,IAAI,CAAC,KAAK,CAAC;QAAE,OAAO,CAAC,CAAC;IACtB,IAAI,CAAC,KAAK,CAAC;QAAE,OAAO,CAAC,CAAC;IAEtB,IAAI,IAAI,GAAG,IAAI,KAAK,CAAS,CAAC,GAAG,CAAC,CAAC,CAAC;IACpC,IAAI,IAAI,GAAG,IAAI,KAAK,CAAS,CAAC,GAAG,CAAC,CAAC,CAAC;IAEpC,KAAK,IAAI,CAAC,GAAG,CAAC,EAAE,CAAC,IAAI,CAAC,EAAE,CAAC,EAAE,EAAE,CAAC;QAC5B,IAAI,CAAC,CAAC,CAAC,GAAG,CAAC,CAAC;IACd,CAAC;IAED,KAAK,IAAI,CAAC,GAAG,CAAC,EAAE,CAAC,IAAI,CAAC,EAAE,CAAC,EAAE,EAAE,CAAC;QAC5B,IAAI,CAAC,CAAC,CAAC,GAAG,CAAC,CAAC;QACZ,MAAM,KAAK,GAAG,EAAE,CAAC,CAAC,GAAG,CAAC,CAAC,CAAC;QACxB,KAAK,IAAI,CAAC,GAAG,CAAC,EAAE,CAAC,IAAI,CAAC,EAAE,CAAC,EAAE,EAAE,CAAC;YAC5B,MAAM,IAAI,GAAG,KAAK,KAAK,EAAE,CAAC,CAAC,GAAG,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC;YACzC,IAAI,CAAC,CAAC,CAAC,GAAG,IAAI,CAAC,GAAG,CAChB,IAAI,CAAC,CAAC,CAAC,GAAG,CAAC,EAAE,WAAW;YACxB,IAAI,CAAC,CAAC,GAAG,CAAC,CAAC,GAAG,CAAC,EAAE,YAAY;YAC7B,IAAI,CAAC,CAAC,GAAG,CAAC,CAAC,GAAG,IAAI,CAAC,eAAe;aACnC,CAAC;QACJ,CAAC;QACD,MAAM,IAAI,GAAG,IAAI,CAAC;QAClB,IAAI,GAAG,IAAI,CAAC;QACZ,IAAI,GAAG,IAAI,CAAC;IACd,CAAC;IAED,OAAO,IAAI,CAAC,CAAC,CAAC,CAAC;AACjB,CAAC;AAED,MAAM,UAAU,UAAU,CAAC,IAAmB;IAC5C,MAAM,aAAa,GAAG,IAAI,EAAE,aAAa,IAAI,KAAK,CAAC;IACnD,MAAM,IAAI,GAAG,IAAI,EAAE,IAAI,IAAI,YAAY,CAAC;IAExC,OAAO,cAAc,CAAC,IAAI,EAAE,CAAC,KAAK,EAAE,MAAM,EAAE,EAAE;QAC5C,MAAM,QAAQ,GAAG,MAAM,CAAC,IAAI,CAAC,QAAQ,CAAC;QACtC,IAAI,QAAQ,KAAK,SAAS,EAAE,CAAC;YAC3B,OAAO,EAAE,IAAI,EAAE,KAAK,EAAE,CAAC,EAAE,OAAO,EAAE,4BAA4B,EAAE,CAAC;QACnE,CAAC;QAED,MAAM,MAAM,GAAG,MAAM,CAAC,MAAM,CAAC,IAAI,EAAE,CAAC;QACpC,MAAM,MAAM,GAAG,QAAQ,CAAC,IAAI,EAAE,CAAC;QAE/B,MAAM,EAAE,GAAG,aAAa,CAAC,CAAC,CAAC,MAAM,CAAC,CAAC,CAAC,MAAM,CAAC,WAAW,EAAE,CAAC;QACzD,MAAM,EAAE,GAAG,aAAa,CAAC,CAAC,CAAC,MAAM,CAAC,CAAC,CAAC,MAAM,CAAC,WAAW,EAAE,CAAC;QAEzD,MAAM,QAAQ,GAAG,CAAC,GAAG,EAAE,CAAC,CAAC;QACzB,MAAM,QAAQ,GAAG,CAAC,GAAG,EAAE,CAAC,CAAC;QAEzB,MAAM,MAAM,GAAG,IAAI,CAAC,GAAG,CAAC,QAAQ,CAAC,MAAM,EAAE,QAAQ,CAAC,MAAM,CAAC,CAAC;QAC1D,IAAI,MAAM,KAAK,CAAC,EAAE,CAAC;YACjB,OAAO,EAAE,IAAI,EAAE,KAAK,EAAE,CAAC,EAAE,CAAC;QAC5B,CAAC;QAED,MAAM,QAAQ,GAAG,mBAAmB,CAAC,QAAQ,EAAE,QAAQ,CAAC,CAAC;QACzD,MAAM,KAAK,GAAG,CAAC,GAAG,QAAQ,GAAG,MAAM,CAAC;QAEpC,OAAO;YACL,IAAI;YACJ,KAAK;YACL,OAAO,EACL,KAAK,GAAG,CAAC,CAAC,CAAC,CAAC,aAAa,MAAM,WAAW,MAAM,gBAAgB,QAAQ,GAAG,CAAC,CAAC,CAAC,SAAS;SAC1F,CAAC;IACJ,CAAC,CAAa,CAAC,CAAC;AAClB,CAAC;AAED,MAAM,UAAU,QAAQ,CAAC,IAAmB;IAC1C,MAAM,aAAa,GAAG,IAAI,EAAE,aAAa,IAAI,KAAK,CAAC;IACnD,MAAM,IAAI,GAAG,IAAI,EAAE,IAAI,IAAI,UAAU,CAAC;IAEtC,OAAO,cAAc,CAAC,IAAI,EAAE,CAAC,KAAK,EAAE,MAAM,EAAE,EAAE;QAC5C,MAAM,QAAQ,GAAG,MAAM,CAAC,IAAI,CAAC,QAAQ,CAAC;QACtC,IAAI,QAAQ,KAAK,SAAS,EAAE,CAAC;YAC3B,OAAO,EAAE,IAAI,EAAE,KAAK,EAAE,CAAC,EAAE,OAAO,EAAE,4BAA4B,EAAE,CAAC;QACnE,CAAC;QAED,MAAM,MAAM,GAAG,aAAa,CAAC,CAAC,CAAC,MAAM,CAAC,MAAM,CAAC,CAAC,CAAC,MAAM,CAAC,MAAM,CAAC,WAAW,EAAE,CAAC;QAC3E,MAAM,MAAM,GAAG,aAAa,CAAC,CAAC,CAAC,QAAQ,CAAC,CAAC,CAAC,QAAQ,CAAC,WAAW,EAAE,CAAC;QACjE,MAAM,KAAK,GAAG,MAAM,CAAC,QAAQ,CAAC,MAAM,CAAC,CAAC;QAEtC,OAAO;YACL,IAAI;YACJ,KAAK,EAAE,KAAK,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC;YACpB,OAAO,EAAE,KAAK,CAAC,CAAC,CAAC,SAAS,CAAC,CAAC,CAAC,4BAA4B,QAAQ,GAAG;SACrE,CAAC;IACJ,CAAC,CAAa,CAAC,CAAC;AAClB,CAAC;AAED;;;;GAIG;AACH,MAAM,UAAU,KAAK,CAAC,OAAwB,EAAE,IAAoB;IAClE,MAAM,MAAM,GAAG,OAAO,OAAO,KAAK,QAAQ,CAAC,CAAC,CAAC,IAAI,MAAM,CAAC,OAAO,CAAC,CAAC,CAAC,CAAC,OAAO,CAAC;IAC3E,MAAM,EAAE,GAAG,IAAI,MAAM,CAAC,MAAM,CAAC,MAAM,EAAE,MAAM,CAAC,KAAK,CAAC,OAAO,CAAC,OAAO,EAAE,EAAE,CAAC,CAAC,CAAC;IACxE,MAAM,IAAI,GAAG,IAAI,EAAE,IAAI,IAAI,OAAO,CAAC;IAEnC,OAAO,cAAc,CAAC,IAAI,EAAE,CAAC,KAAK,EAAE,MAAM,EAAE,EAAE;QAC5C,MAAM,KAAK,GAAG,EAAE,CAAC,IAAI,CAAC,MAAM,CAAC,MAAM,CAAC,CAAC;QACrC,OAAO;YACL,IAAI;YACJ,KAAK,EAAE,KAAK,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC;YACpB,OAAO,EAAE,KAAK,CAAC,CAAC,CAAC,SAAS,CAAC,CAAC,CAAC,iCAAiC,MAAM,EAAE;SACvE,CAAC;IACJ,CAAC,CAAa,CAAC,CAAC;AAClB,CAAC;AAED,MAAM,UAAU,UAAU,CAAC,MAAe,EAAE,IAAoB;IAC9D,MAAM,IAAI,GAAG,IAAI,EAAE,IAAI,IAAI,YAAY,CAAC;IAExC,OAAO,cAAc,CAAC,IAAI,EAAE,CAAC,KAAK,EAAE,MAAM,EAAE,EAAE;QAC5C,IAAI,MAAe,CAAC;QACpB,IAAI,CAAC;YACH,MAAM,GAAG,IAAI,CAAC,KAAK,CAAC,MAAM,CAAC,MAAM,CAAC,CAAC;QACrC,CAAC;QAAC,OAAO,CAAC,EAAE,CAAC;YACX,OAAO;gBACL,IAAI;gBACJ,KAAK,EAAE,CAAC;gBACR,OAAO,EAAE,iBAAkB,CAAW,CAAC,OAAO,EAAE;aACjD,CAAC;QACJ,CAAC;QAED,MAAM,UAAU,GAAG,MAAM,CAAC,SAAS,CAAC,MAAM,CAAC,CAAC;QAC5C,IAAI,UAAU,CAAC,OAAO,EAAE,CAAC;YACvB,OAAO,EAAE,IAAI,EAAE,KAAK,EAAE,CAAC,EAAE,CAAC;QAC5B,CAAC;QAED,OAAO;YACL,IAAI;YACJ,KAAK,EAAE,CAAC;YACR,OAAO,EAAE,6BAA6B,UAAU,CAAC,KAAK,CAAC,OAAO,EAAE;SACjE,CAAC;IACJ,CAAC,CAAa,CAAC,CAAC;AAClB,CAAC"}
@@ -1,5 +1,5 @@
1
1
  export type { MetricScore, MetricUsage, MetricOptions, EvalCaseResult, MetricFn, StatisticalMetricFn, StatisticalScore, } from './types.js';
2
- export { exactMatch, contains, regex, jsonSchema } from './deterministic.js';
2
+ export { exactMatch, contains, similarity, regex, jsonSchema } from './deterministic.js';
3
3
  export { metric } from './custom.js';
4
4
  export type { CustomMetricConfig } from './custom.js';
5
5
  export { latency, cost, tokenUsage } from './statistical.js';
@@ -1 +1 @@
1
- {"version":3,"file":"index.d.ts","sourceRoot":"","sources":["../../src/metrics/index.ts"],"names":[],"mappings":"AAAA,YAAY,EACV,WAAW,EACX,WAAW,EACX,aAAa,EACb,cAAc,EACd,QAAQ,EACR,mBAAmB,EACnB,gBAAgB,GACjB,MAAM,SAAS,CAAC;AACjB,OAAO,EAAE,UAAU,EAAE,QAAQ,EAAE,KAAK,EAAE,UAAU,EAAE,MAAM,iBAAiB,CAAC;AAC1E,OAAO,EAAE,MAAM,EAAE,MAAM,UAAU,CAAC;AAClC,YAAY,EAAE,kBAAkB,EAAE,MAAM,UAAU,CAAC;AACnD,OAAO,EAAE,OAAO,EAAE,IAAI,EAAE,UAAU,EAAE,MAAM,eAAe,CAAC;AAC1D,OAAO,EACL,YAAY,EACZ,SAAS,EACT,SAAS,EACT,WAAW,EACX,SAAS,EACT,gBAAgB,GACjB,MAAM,aAAa,CAAC;AACrB,YAAY,EAAE,YAAY,EAAE,WAAW,EAAE,MAAM,aAAa,CAAC"}
1
+ {"version":3,"file":"index.d.ts","sourceRoot":"","sources":["../../src/metrics/index.ts"],"names":[],"mappings":"AAAA,YAAY,EACV,WAAW,EACX,WAAW,EACX,aAAa,EACb,cAAc,EACd,QAAQ,EACR,mBAAmB,EACnB,gBAAgB,GACjB,MAAM,SAAS,CAAC;AACjB,OAAO,EAAE,UAAU,EAAE,QAAQ,EAAE,UAAU,EAAE,KAAK,EAAE,UAAU,EAAE,MAAM,iBAAiB,CAAC;AACtF,OAAO,EAAE,MAAM,EAAE,MAAM,UAAU,CAAC;AAClC,YAAY,EAAE,kBAAkB,EAAE,MAAM,UAAU,CAAC;AACnD,OAAO,EAAE,OAAO,EAAE,IAAI,EAAE,UAAU,EAAE,MAAM,eAAe,CAAC;AAC1D,OAAO,EACL,YAAY,EACZ,SAAS,EACT,SAAS,EACT,WAAW,EACX,SAAS,EACT,gBAAgB,GACjB,MAAM,aAAa,CAAC;AACrB,YAAY,EAAE,YAAY,EAAE,WAAW,EAAE,MAAM,aAAa,CAAC"}
@@ -1,4 +1,4 @@
1
- export { exactMatch, contains, regex, jsonSchema } from './deterministic.js';
1
+ export { exactMatch, contains, similarity, regex, jsonSchema } from './deterministic.js';
2
2
  export { metric } from './custom.js';
3
3
  export { latency, cost, tokenUsage } from './statistical.js';
4
4
  export { faithfulness, relevance, coherence, helpfulness, llmMetric, bindJudgeContext, } from './llm-judge.js';
@@ -1 +1 @@
1
- {"version":3,"file":"index.js","sourceRoot":"","sources":["../../src/metrics/index.ts"],"names":[],"mappings":"AASA,OAAO,EAAE,UAAU,EAAE,QAAQ,EAAE,KAAK,EAAE,UAAU,EAAE,MAAM,iBAAiB,CAAC;AAC1E,OAAO,EAAE,MAAM,EAAE,MAAM,UAAU,CAAC;AAElC,OAAO,EAAE,OAAO,EAAE,IAAI,EAAE,UAAU,EAAE,MAAM,eAAe,CAAC;AAC1D,OAAO,EACL,YAAY,EACZ,SAAS,EACT,SAAS,EACT,WAAW,EACX,SAAS,EACT,gBAAgB,GACjB,MAAM,aAAa,CAAC"}
1
+ {"version":3,"file":"index.js","sourceRoot":"","sources":["../../src/metrics/index.ts"],"names":[],"mappings":"AASA,OAAO,EAAE,UAAU,EAAE,QAAQ,EAAE,UAAU,EAAE,KAAK,EAAE,UAAU,EAAE,MAAM,iBAAiB,CAAC;AACtF,OAAO,EAAE,MAAM,EAAE,MAAM,UAAU,CAAC;AAElC,OAAO,EAAE,OAAO,EAAE,IAAI,EAAE,UAAU,EAAE,MAAM,eAAe,CAAC;AAC1D,OAAO,EACL,YAAY,EACZ,SAAS,EACT,SAAS,EACT,WAAW,EACX,SAAS,EACT,gBAAgB,GACjB,MAAM,aAAa,CAAC"}
package/package.json CHANGED
@@ -1,6 +1,6 @@
1
1
  {
2
2
  "name": "@cogitator-ai/evals",
3
- "version": "0.6.0",
3
+ "version": "0.7.0",
4
4
  "description": "Evaluation framework for Cogitator AI agents",
5
5
  "keywords": [
6
6
  "cogitator",
@@ -23,6 +23,7 @@
23
23
  "default": "./dist/index.js"
24
24
  }
25
25
  },
26
+ "sideEffects": false,
26
27
  "files": [
27
28
  "dist"
28
29
  ],
@@ -33,7 +34,7 @@
33
34
  "zod": "^4.6.5"
34
35
  },
35
36
  "peerDependencies": {
36
- "@cogitator-ai/core": ">=0.34.0"
37
+ "@cogitator-ai/core": ">=0.41.0"
37
38
  },
38
39
  "peerDependenciesMeta": {
39
40
  "@cogitator-ai/core": {
@@ -44,14 +45,14 @@
44
45
  "papaparse": "^5.7.0"
45
46
  },
46
47
  "devDependencies": {
47
- "@cogitator-ai/core": "0.34.0",
48
+ "@cogitator-ai/core": "0.41.0",
48
49
  "@types/papaparse": "^5.3.15",
49
50
  "typescript": "^6.0.3",
50
51
  "vitest": "^5.0.3"
51
52
  },
52
53
  "repository": {
53
54
  "type": "git",
54
- "url": "https://github.com/cogitator-ai/Cogitator-AI.git",
55
+ "url": "git+https://github.com/cogitator-ai/Cogitator-AI.git",
55
56
  "directory": "packages/evals"
56
57
  },
57
58
  "publishConfig": {