@qcobro/common 1.34.4 → 1.35.0
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/dist/schemas/agentEvaluations.d.ts +14 -9
- package/dist/schemas/agentEvaluations.d.ts.map +1 -1
- package/dist/schemas/agentEvaluations.js +14 -9
- package/dist/schemas/agentEvaluations.js.map +1 -1
- package/dist/types/agentEvaluations.d.ts +2 -1
- package/dist/types/agentEvaluations.d.ts.map +1 -1
- package/dist/types/evalJudge.d.ts +31 -0
- package/dist/types/evalJudge.d.ts.map +1 -0
- package/dist/types/evalJudge.js +2 -0
- package/dist/types/evalJudge.js.map +1 -0
- package/dist/types/index.d.ts +1 -0
- package/dist/types/index.d.ts.map +1 -1
- package/dist/types/index.js +1 -0
- package/dist/types/index.js.map +1 -1
- package/package.json +1 -1
|
@@ -36,15 +36,20 @@ export declare const evalExpectedToolSchema: z.ZodObject<{
|
|
|
36
36
|
parameters: z.ZodOptional<z.ZodRecord<z.ZodString, z.ZodUnknown>>;
|
|
37
37
|
}, z.core.$strip>;
|
|
38
38
|
/**
|
|
39
|
-
* A single turn's optional expectation. `
|
|
40
|
-
*
|
|
41
|
-
*
|
|
42
|
-
*
|
|
43
|
-
* `
|
|
44
|
-
*
|
|
45
|
-
*
|
|
46
|
-
*
|
|
47
|
-
*
|
|
39
|
+
* A single turn's optional expectation. `tools` is graded for `VOICE_AI` only (relayed from
|
|
40
|
+
* Fonoster); `action`/`resultado` are graded for `EMAIL`/`WHATSAPP` (the behavioral analog of
|
|
41
|
+
* an expected tool call — see design.md). `text` is graded for all three: VOICE_AI relays it
|
|
42
|
+
* to Fonoster's own `evaluateIntelligence` judge; EMAIL/WHATSAPP grade it in-process —
|
|
43
|
+
* `EXACT` is a literal match, `SIMILAR` runs QCobro's own judge (`TextSimilarityJudge`),
|
|
44
|
+
* which checks intent match AND that the reply introduces no fact/entity absent from both
|
|
45
|
+
* the expected reply and the scenario's account context (catches hallucination, unlike an
|
|
46
|
+
* intent-only judge). A turn with no `expected` still runs and streams a result, it just has
|
|
47
|
+
* nothing to grade — **except for VOICE_AI**: Fonoster's `evaluateIntelligence` requires
|
|
48
|
+
* `expected.text` on every conversation turn (confirmed against the live service, correcting
|
|
49
|
+
* an earlier assumption in design.md), so `resolveEvalTarget` rejects a VOICE_AI scenario
|
|
50
|
+
* with any turn missing it before ever calling Fonoster. The VOICE_AI-specific requirement is
|
|
51
|
+
* enforced at resolution time, not here, since this shape is shared across all three channel
|
|
52
|
+
* types.
|
|
48
53
|
*/
|
|
49
54
|
export declare const evalExpectedSchema: z.ZodObject<{
|
|
50
55
|
text: z.ZodOptional<z.ZodObject<{
|
|
@@ -1 +1 @@
|
|
|
1
|
-
{"version":3,"file":"agentEvaluations.d.ts","sourceRoot":"","sources":["../../src/schemas/agentEvaluations.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,CAAC,EAAE,MAAM,KAAK,CAAC;AAIxB;;;;;GAKG;AACH,eAAO,MAAM,iBAAiB;;;;;;;;;;;;;;;iBAe5B,CAAC;AACH,MAAM,MAAM,gBAAgB,GAAG,CAAC,CAAC,KAAK,CAAC,OAAO,iBAAiB,CAAC,CAAC;AAEjE,2FAA2F;AAC3F,eAAO,MAAM,sBAAsB;;;;;;iBAGjC,CAAC;AAEH,kGAAkG;AAClG,eAAO,MAAM,sBAAsB;;;iBAGjC,CAAC;AAEH
|
|
1
|
+
{"version":3,"file":"agentEvaluations.d.ts","sourceRoot":"","sources":["../../src/schemas/agentEvaluations.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,CAAC,EAAE,MAAM,KAAK,CAAC;AAIxB;;;;;GAKG;AACH,eAAO,MAAM,iBAAiB;;;;;;;;;;;;;;;iBAe5B,CAAC;AACH,MAAM,MAAM,gBAAgB,GAAG,CAAC,CAAC,KAAK,CAAC,OAAO,iBAAiB,CAAC,CAAC;AAEjE,2FAA2F;AAC3F,eAAO,MAAM,sBAAsB;;;;;;iBAGjC,CAAC;AAEH,kGAAkG;AAClG,eAAO,MAAM,sBAAsB;;;iBAGjC,CAAC;AAEH;;;;;;;;;;;;;;;GAeG;AACH,eAAO,MAAM,kBAAkB;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;kBAUpB,CAAC;AAEZ,eAAO,MAAM,cAAc;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;iBAGzB,CAAC;AAEH,eAAO,MAAM,kBAAkB;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;iBAO7B,CAAC;AACH,MAAM,MAAM,YAAY,GAAG,CAAC,CAAC,KAAK,CAAC,OAAO,kBAAkB,CAAC,CAAC;AAC9D,MAAM,MAAM,QAAQ,GAAG,CAAC,CAAC,KAAK,CAAC,OAAO,cAAc,CAAC,CAAC;AA0BtD;;;;;;GAMG;AACH,eAAO,MAAM,kBAAkB;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;2BAI7B,CAAC;AACH,MAAM,MAAM,iBAAiB,GAAG,CAAC,CAAC,KAAK,CAAC,OAAO,kBAAkB,CAAC,CAAC;AAEnE;8CAC8C;AAC9C,eAAO,MAAM,4BAA4B;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;iBAGvC,CAAC;AAEH,iGAAiG;AACjG,eAAO,MAAM,wBAAwB;;iBAEnC,CAAC;AAEH,eAAO,MAAM,mBAAmB;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;mBAG9B,CAAC;AACH,MAAM,MAAM,qBAAqB,GAAG,CAAC,CAAC,KAAK,CAAC,OAAO,4BAA4B,CAAC,CAAC;AACjF,MAAM,MAAM,iBAAiB,GAAG,CAAC,CAAC,KAAK,CAAC,OAAO,wBAAwB,CAAC,CAAC;AACzE,MAAM,MAAM,aAAa,GAAG,CAAC,CAAC,KAAK,CAAC,OAAO,mBAAmB,CAAC,CAAC;AAEhE;qFACqF;AACrF,eAAO,MAAM,sBAAsB;;;;;;;;;;;;;;;;;;2BAGjC,CAAC;AAEH,eAAO,MAAM,2BAA2B;;;;;;;;;;;;;;;;;;iBAGtC,CAAC;AAEH,eAAO,MAAM,uBAAuB;;;;;;;;;;;;;;;;;;iBAGlC,CAAC;AAEH,eAAO,MAAM,kBAAkB;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;mBAAkE,CAAC;AAClG,MAAM,MAAM,YAAY,GAAG,CAAC,CAAC,KAAK,CAAC,OAAO,kBAAkB,CAAC,CAAC"}
|
|
@@ -34,15 +34,20 @@ export const evalExpectedToolSchema = z.object({
|
|
|
34
34
|
parameters: z.record(z.string(), z.unknown()).optional()
|
|
35
35
|
});
|
|
36
36
|
/**
|
|
37
|
-
* A single turn's optional expectation. `
|
|
38
|
-
*
|
|
39
|
-
*
|
|
40
|
-
*
|
|
41
|
-
* `
|
|
42
|
-
*
|
|
43
|
-
*
|
|
44
|
-
*
|
|
45
|
-
*
|
|
37
|
+
* A single turn's optional expectation. `tools` is graded for `VOICE_AI` only (relayed from
|
|
38
|
+
* Fonoster); `action`/`resultado` are graded for `EMAIL`/`WHATSAPP` (the behavioral analog of
|
|
39
|
+
* an expected tool call — see design.md). `text` is graded for all three: VOICE_AI relays it
|
|
40
|
+
* to Fonoster's own `evaluateIntelligence` judge; EMAIL/WHATSAPP grade it in-process —
|
|
41
|
+
* `EXACT` is a literal match, `SIMILAR` runs QCobro's own judge (`TextSimilarityJudge`),
|
|
42
|
+
* which checks intent match AND that the reply introduces no fact/entity absent from both
|
|
43
|
+
* the expected reply and the scenario's account context (catches hallucination, unlike an
|
|
44
|
+
* intent-only judge). A turn with no `expected` still runs and streams a result, it just has
|
|
45
|
+
* nothing to grade — **except for VOICE_AI**: Fonoster's `evaluateIntelligence` requires
|
|
46
|
+
* `expected.text` on every conversation turn (confirmed against the live service, correcting
|
|
47
|
+
* an earlier assumption in design.md), so `resolveEvalTarget` rejects a VOICE_AI scenario
|
|
48
|
+
* with any turn missing it before ever calling Fonoster. The VOICE_AI-specific requirement is
|
|
49
|
+
* enforced at resolution time, not here, since this shape is shared across all three channel
|
|
50
|
+
* types.
|
|
46
51
|
*/
|
|
47
52
|
export const evalExpectedSchema = z
|
|
48
53
|
.object({
|
|
@@ -1 +1 @@
|
|
|
1
|
-
{"version":3,"file":"agentEvaluations.js","sourceRoot":"","sources":["../../src/schemas/agentEvaluations.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,CAAC,EAAE,MAAM,KAAK,CAAC;AACxB,OAAO,EAAE,yBAAyB,EAAE,MAAM,qBAAqB,CAAC;AAChE,OAAO,EAAE,eAAe,EAAE,MAAM,iBAAiB,CAAC;AAElD;;;;;GAKG;AACH,MAAM,CAAC,MAAM,iBAAiB,GAAG,CAAC,CAAC,MAAM,CAAC;IACxC,QAAQ,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,GAAG,CAAC,CAAC,CAAC;IAC3B,eAAe,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,OAAO,CAAC,CAAC,CAAC;IACtC,kBAAkB,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,OAAO,CAAC,CAAC,CAAC;IACzC,WAAW,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,GAAG,EAAE,CAAC,OAAO,CAAC,CAAC,CAAC;IACxC,WAAW,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,OAAO,CAAC,CAAC,CAAC;IAClC,cAAc,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,QAAQ,EAAE;IACrC,WAAW,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,GAAG,EAAE,CAAC,OAAO,CAAC,CAAC,CAAC;IACxC,kBAAkB,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,GAAG,EAAE,CAAC,OAAO,CAAC,CAAC,CAAC;IAC/C,eAAe,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,QAAQ,EAAE,CAAC,QAAQ,EAAE;IACjD,iBAAiB,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,QAAQ,EAAE;IACxC,iBAAiB,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,QAAQ,EAAE;IACxC,eAAe,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,QAAQ,EAAE;IACtC,cAAc,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,QAAQ,EAAE;IACrC,QAAQ,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,OAAO,CAAC,KAAK,CAAC;CACpC,CAAC,CAAC;AAGH,2FAA2F;AAC3F,MAAM,CAAC,MAAM,sBAAsB,GAAG,CAAC,CAAC,MAAM,CAAC;IAC7C,IAAI,EAAE,CAAC,CAAC,IAAI,CAAC,CAAC,OAAO,EAAE,SAAS,CAAC,CAAC;IAClC,QAAQ,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,GAAG,CAAC,CAAC,CAAC;CAC5B,CAAC,CAAC;AAEH,kGAAkG;AAClG,MAAM,CAAC,MAAM,sBAAsB,GAAG,CAAC,CAAC,MAAM,CAAC;IAC7C,IAAI,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,GAAG,CAAC,CAAC,CAAC;IACvB,UAAU,EAAE,CAAC,CAAC,MAAM,CAAC,CAAC,CAAC,MAAM,EAAE,EAAE,CAAC,CAAC,OAAO,EAAE,CAAC,CAAC,QAAQ,EAAE;CACzD,CAAC,CAAC;AAEH
|
|
1
|
+
{"version":3,"file":"agentEvaluations.js","sourceRoot":"","sources":["../../src/schemas/agentEvaluations.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,CAAC,EAAE,MAAM,KAAK,CAAC;AACxB,OAAO,EAAE,yBAAyB,EAAE,MAAM,qBAAqB,CAAC;AAChE,OAAO,EAAE,eAAe,EAAE,MAAM,iBAAiB,CAAC;AAElD;;;;;GAKG;AACH,MAAM,CAAC,MAAM,iBAAiB,GAAG,CAAC,CAAC,MAAM,CAAC;IACxC,QAAQ,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,GAAG,CAAC,CAAC,CAAC;IAC3B,eAAe,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,OAAO,CAAC,CAAC,CAAC;IACtC,kBAAkB,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,OAAO,CAAC,CAAC,CAAC;IACzC,WAAW,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,GAAG,EAAE,CAAC,OAAO,CAAC,CAAC,CAAC;IACxC,WAAW,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,OAAO,CAAC,CAAC,CAAC;IAClC,cAAc,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,QAAQ,EAAE;IACrC,WAAW,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,GAAG,EAAE,CAAC,OAAO,CAAC,CAAC,CAAC;IACxC,kBAAkB,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,GAAG,EAAE,CAAC,OAAO,CAAC,CAAC,CAAC;IAC/C,eAAe,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,QAAQ,EAAE,CAAC,QAAQ,EAAE;IACjD,iBAAiB,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,QAAQ,EAAE;IACxC,iBAAiB,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,QAAQ,EAAE;IACxC,eAAe,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,QAAQ,EAAE;IACtC,cAAc,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,QAAQ,EAAE;IACrC,QAAQ,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,OAAO,CAAC,KAAK,CAAC;CACpC,CAAC,CAAC;AAGH,2FAA2F;AAC3F,MAAM,CAAC,MAAM,sBAAsB,GAAG,CAAC,CAAC,MAAM,CAAC;IAC7C,IAAI,EAAE,CAAC,CAAC,IAAI,CAAC,CAAC,OAAO,EAAE,SAAS,CAAC,CAAC;IAClC,QAAQ,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,GAAG,CAAC,CAAC,CAAC;CAC5B,CAAC,CAAC;AAEH,kGAAkG;AAClG,MAAM,CAAC,MAAM,sBAAsB,GAAG,CAAC,CAAC,MAAM,CAAC;IAC7C,IAAI,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,GAAG,CAAC,CAAC,CAAC;IACvB,UAAU,EAAE,CAAC,CAAC,MAAM,CAAC,CAAC,CAAC,MAAM,EAAE,EAAE,CAAC,CAAC,OAAO,EAAE,CAAC,CAAC,QAAQ,EAAE;CACzD,CAAC,CAAC;AAEH;;;;;;;;;;;;;;;GAeG;AACH,MAAM,CAAC,MAAM,kBAAkB,GAAG,CAAC;KAChC,MAAM,CAAC;IACN,IAAI,EAAE,sBAAsB,CAAC,QAAQ,EAAE;IACvC,KAAK,EAAE,CAAC,CAAC,KAAK,CAAC,sBAAsB,CAAC,CAAC,QAAQ,EAAE;IACjD,MAAM,EAAE,CAAC,CAAC,IAAI,CAAC,CAAC,OAAO,EAAE,QAAQ,EAAE,SAAS,EAAE,UAAU,CAAC,CAAC,CAAC,QAAQ,EAAE;IACrE,SAAS,EAAE,eAAe,CAAC,QAAQ,EAAE;CACtC,CAAC;IACF,wFAAwF;IACxF,4FAA4F;IAC5F,sEAAsE;KACrE,MAAM,EAAE,CAAC;AAEZ,MAAM,CAAC,MAAM,cAAc,GAAG,CAAC,CAAC,MAAM,CAAC;IACrC,KAAK,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,GAAG,CAAC,CAAC,CAAC;IACxB,QAAQ,EAAE,kBAAkB,CAAC,QAAQ,EAAE;CACxC,CAAC,CAAC;AAEH,MAAM,CAAC,MAAM,kBAAkB,GAAG,CAAC,CAAC,MAAM,CAAC;IACzC,GAAG,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,GAAG,CAAC,CAAC,CAAC;IACtB;oFACgF;IAChF,WAAW,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,QAAQ,EAAE;IAClC,OAAO,EAAE,iBAAiB;IAC1B,KAAK,EAAE,CAAC,CAAC,KAAK,CAAC,cAAc,CAAC,CAAC,GAAG,CAAC,CAAC,CAAC;CACtC,CAAC,CAAC;AAIH,0EAA0E;AAC1E,6FAA6F;AAC7F,8FAA8F;AAC9F,sFAAsF;AACtF,kCAAkC;AAClC,MAAM,CAAC,aAAa,EAAE,sBAAsB,EAAE,SAAS,EAAE,WAAW,EAAE,cAAc,CAAC,GACnF,yBAAyB,CAAC,OAAO,CAAC;AAEpC,SAAS,gBAAgB,CAAC,MAA8C,EAAE,QAAgB;IACxF,IAAI,MAAM,CAAC,KAAK,CAAC,IAAI,CAAC,KAAK,KAAK,QAAQ,EAAE,CAAC;QACzC,MAAM,IAAI,KAAK,CACb,6DAA6D,QAAQ,cAAc;YACjF,IAAI,MAAM,CAAC,KAAK,CAAC,IAAI,CAAC,KAAK,8DAA8D,CAC5F,CAAC;IACJ,CAAC;AACH,CAAC;AACD,gBAAgB,CAAC,aAAa,EAAE,UAAU,CAAC,CAAC;AAC5C,gBAAgB,CAAC,sBAAsB,EAAE,mBAAmB,CAAC,CAAC;AAC9D,gBAAgB,CAAC,SAAS,EAAE,KAAK,CAAC,CAAC;AACnC,gBAAgB,CAAC,WAAW,EAAE,OAAO,CAAC,CAAC;AACvC,gBAAgB,CAAC,cAAc,EAAE,UAAU,CAAC,CAAC;AAE7C,MAAM,aAAa,GAAG,EAAE,SAAS,EAAE,CAAC,CAAC,KAAK,CAAC,kBAAkB,CAAC,CAAC,GAAG,CAAC,CAAC,CAAC,EAAE,CAAC;AAExE;;;;;;GAMG;AACH,MAAM,CAAC,MAAM,kBAAkB,GAAG,CAAC,CAAC,kBAAkB,CAAC,MAAM,EAAE;IAC7D,aAAa,CAAC,MAAM,CAAC,aAAa,CAAC;IACnC,WAAW,CAAC,MAAM,CAAC,aAAa,CAAC;IACjC,cAAc,CAAC,MAAM,CAAC,aAAa,CAAC;CACrC,CAAC,CAAC;AAGH;8CAC8C;AAC9C,MAAM,CAAC,MAAM,4BAA4B,GAAG,CAAC,CAAC,MAAM,CAAC;IACnD,eAAe,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,GAAG,CAAC,CAAC,CAAC;IAClC,SAAS,EAAE,CAAC,CAAC,KAAK,CAAC,kBAAkB,CAAC,CAAC,GAAG,CAAC,CAAC,CAAC;CAC9C,CAAC,CAAC;AAEH,iGAAiG;AACjG,MAAM,CAAC,MAAM,wBAAwB,GAAG,CAAC,CAAC,MAAM,CAAC;IAC/C,IAAI,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,GAAG,CAAC,CAAC,CAAC;CACxB,CAAC,CAAC;AAEH,MAAM,CAAC,MAAM,mBAAmB,GAAG,CAAC,CAAC,KAAK,CAAC;IACzC,4BAA4B;IAC5B,wBAAwB;CACzB,CAAC,CAAC;AAKH;qFACqF;AACrF,MAAM,CAAC,MAAM,sBAAsB,GAAG,CAAC,CAAC,kBAAkB,CAAC,MAAM,EAAE;IACjE,SAAS;IACT,sBAAsB;CACvB,CAAC,CAAC;AAEH,MAAM,CAAC,MAAM,2BAA2B,GAAG,CAAC,CAAC,MAAM,CAAC;IAClD,eAAe,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,GAAG,CAAC,CAAC,CAAC;IAClC,OAAO,EAAE,iBAAiB;CAC3B,CAAC,CAAC;AAEH,MAAM,CAAC,MAAM,uBAAuB,GAAG,CAAC,CAAC,MAAM,CAAC;IAC9C,IAAI,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,GAAG,CAAC,CAAC,CAAC;IACvB,OAAO,EAAE,iBAAiB;CAC3B,CAAC,CAAC;AAEH,MAAM,CAAC,MAAM,kBAAkB,GAAG,CAAC,CAAC,KAAK,CAAC,CAAC,2BAA2B,EAAE,uBAAuB,CAAC,CAAC,CAAC"}
|
|
@@ -10,7 +10,8 @@ export interface EvalStepResult {
|
|
|
10
10
|
input: string;
|
|
11
11
|
passed?: boolean;
|
|
12
12
|
errorMessage?: string;
|
|
13
|
-
/** VOICE_AI (relayed from Fonoster's
|
|
13
|
+
/** The agent's reply/response text for this turn — VOICE_AI (relayed from Fonoster's
|
|
14
|
+
* `stepResult`) or EMAIL/WHATSAPP (the autopilot's `replyBody`, when it replied). */
|
|
14
15
|
aiResponse?: string;
|
|
15
16
|
expectedResponse?: string;
|
|
16
17
|
evaluationType?: "EXACT" | "SIMILAR";
|
|
@@ -1 +1 @@
|
|
|
1
|
-
{"version":3,"file":"agentEvaluations.d.ts","sourceRoot":"","sources":["../../src/types/agentEvaluations.ts"],"names":[],"mappings":"AAAA,OAAO,KAAK,EAAE,SAAS,EAAE,MAAM,8BAA8B,CAAC;AAC9D,OAAO,KAAK,EAAE,oBAAoB,EAAE,MAAM,YAAY,CAAC;AAEvD;;;;GAIG;AACH,MAAM,WAAW,cAAc;IAC7B,SAAS,EAAE,MAAM,CAAC;IAClB,KAAK,EAAE,MAAM,CAAC;IACd,MAAM,CAAC,EAAE,OAAO,CAAC;IACjB,YAAY,CAAC,EAAE,MAAM,CAAC;IACtB
|
|
1
|
+
{"version":3,"file":"agentEvaluations.d.ts","sourceRoot":"","sources":["../../src/types/agentEvaluations.ts"],"names":[],"mappings":"AAAA,OAAO,KAAK,EAAE,SAAS,EAAE,MAAM,8BAA8B,CAAC;AAC9D,OAAO,KAAK,EAAE,oBAAoB,EAAE,MAAM,YAAY,CAAC;AAEvD;;;;GAIG;AACH,MAAM,WAAW,cAAc;IAC7B,SAAS,EAAE,MAAM,CAAC;IAClB,KAAK,EAAE,MAAM,CAAC;IACd,MAAM,CAAC,EAAE,OAAO,CAAC;IACjB,YAAY,CAAC,EAAE,MAAM,CAAC;IACtB;yFACqF;IACrF,UAAU,CAAC,EAAE,MAAM,CAAC;IACpB,gBAAgB,CAAC,EAAE,MAAM,CAAC;IAC1B,cAAc,CAAC,EAAE,OAAO,GAAG,SAAS,CAAC;IACrC,eAAe,CAAC,EAAE;QAChB,YAAY,EAAE,MAAM,CAAC;QACrB,UAAU,CAAC,EAAE,MAAM,CAAC;QACpB,MAAM,EAAE,OAAO,CAAC;KACjB,EAAE,CAAC;IACJ,8DAA8D;IAC9D,MAAM,CAAC,EAAE,oBAAoB,CAAC;IAC9B,SAAS,CAAC,EAAE,MAAM,GAAG,IAAI,CAAC;CAC3B;AAED,MAAM,WAAW,mBAAmB;IAClC,GAAG,EAAE,MAAM,CAAC;IACZ,aAAa,EAAE,OAAO,CAAC;CACxB;AAED,8FAA8F;AAC9F,MAAM,MAAM,SAAS,GACjB;IAAE,IAAI,EAAE,MAAM,CAAC;IAAC,WAAW,EAAE,MAAM,CAAC;IAAC,MAAM,EAAE,cAAc,CAAA;CAAE,GAC7D;IAAE,IAAI,EAAE,iBAAiB,CAAC;IAAC,WAAW,EAAE,MAAM,CAAC;IAAC,aAAa,EAAE,OAAO,CAAA;CAAE,GACxE;IAAE,IAAI,EAAE,SAAS,CAAC;IAAC,OAAO,EAAE,MAAM,GAAG,MAAM,CAAC;IAAC,SAAS,EAAE,mBAAmB,EAAE,CAAA;CAAE,GAC/E;IAAE,IAAI,EAAE,OAAO,CAAC;IAAC,OAAO,EAAE,MAAM,CAAA;CAAE,CAAC;AAEvC;4EAC4E;AAC5E,MAAM,WAAW,oBAAoB;IACnC,EAAE,EAAE,MAAM,CAAC;IACX,IAAI,EAAE,SAAS,CAAC;IAChB,aAAa,EAAE;QAAE,YAAY,EAAE,MAAM,CAAC;QAAC,YAAY,EAAE,MAAM,GAAG,IAAI,CAAC;QAAC,QAAQ,EAAE,MAAM,CAAA;KAAE,GAAG,IAAI,CAAC;IAC9F,sBAAsB,EAAE;QAAE,MAAM,EAAE,MAAM,CAAC;QAAC,QAAQ,EAAE,MAAM,CAAA;KAAE,GAAG,IAAI,CAAC;IACpE,SAAS,EAAE;QAAE,WAAW,EAAE,MAAM,CAAA;KAAE,GAAG,IAAI,CAAC;IAC1C,WAAW,EAAE;QAAE,YAAY,EAAE,MAAM,CAAC;QAAC,WAAW,EAAE,MAAM,CAAC;QAAC,UAAU,EAAE,MAAM,GAAG,IAAI,CAAA;KAAE,GAAG,IAAI,CAAC;IAC7F,cAAc,EAAE;QAAE,YAAY,EAAE,MAAM,CAAC;QAAC,WAAW,EAAE,MAAM,CAAC;QAAC,UAAU,EAAE,MAAM,GAAG,IAAI,CAAA;KAAE,GAAG,IAAI,CAAC;CACjG;AAED,wFAAwF;AACxF,MAAM,WAAW,uBAAuB;IACtC,aAAa,EAAE;QACb,gBAAgB,CAAC,IAAI,EAAE;YACrB,KAAK,EAAE;gBAAE,EAAE,EAAE,MAAM,CAAC;gBAAC,YAAY,EAAE,MAAM,CAAA;aAAE,CAAC;YAC5C,OAAO,EAAE;gBACP,aAAa,EAAE,IAAI,CAAC;gBACpB,sBAAsB,EAAE,IAAI,CAAC;gBAC7B,SAAS,EAAE,IAAI,CAAC;gBAChB,WAAW,EAAE,IAAI,CAAC;gBAClB,cAAc,EAAE,IAAI,CAAC;aACtB,CAAC;SACH,GAAG,OAAO,CAAC,oBAAoB,CAAC,CAAC;KACnC,CAAC;CACH"}
|
|
@@ -0,0 +1,31 @@
|
|
|
1
|
+
/** What the judge compares: an evaluator-authored expected reply against the reply an
|
|
2
|
+
* agent under test actually produced, grounded by the scenario's rendered account context
|
|
3
|
+
* so a reply correctly citing real account data isn't mistaken for an invented fact. */
|
|
4
|
+
export interface TextSimilarityRequest {
|
|
5
|
+
expected: string;
|
|
6
|
+
actual: string;
|
|
7
|
+
/** Grounding facts the actual reply may legitimately state even when absent from `expected`
|
|
8
|
+
* verbatim: the rendered account context (see `buildOutreachContext`), plus `referenceDate`
|
|
9
|
+
* (today, for resolving a relative date the customer stated, e.g. "the 15th") and
|
|
10
|
+
* `customerMessage` (this turn's inbound message, so a detail the customer themselves
|
|
11
|
+
* supplied isn't mistaken for an invented one). */
|
|
12
|
+
context?: Record<string, unknown>;
|
|
13
|
+
}
|
|
14
|
+
export interface TextSimilarityResult {
|
|
15
|
+
passed: boolean;
|
|
16
|
+
/** One-sentence explanation, present on both pass and fail, for CLI/report output. */
|
|
17
|
+
reason?: string;
|
|
18
|
+
}
|
|
19
|
+
/**
|
|
20
|
+
* Port for judging whether an actual reply matches an expected one in intent, without
|
|
21
|
+
* introducing facts/entities absent from both the expected reply and the grounding
|
|
22
|
+
* context — an intent-only match (à la Fonoster's VOICE_AI evaluator) would pass a reply
|
|
23
|
+
* that hallucinates a bank account, since the hallucinated detail is exactly the kind of
|
|
24
|
+
* "entity" an intent-only judge is told to ignore. The production adapter wraps an LLM
|
|
25
|
+
* (provider from `qcobro.json` `ai`, shared with the insight generator); tests inject a
|
|
26
|
+
* stub. Reached through the tRPC context like other service ports.
|
|
27
|
+
*/
|
|
28
|
+
export interface TextSimilarityJudge {
|
|
29
|
+
compare(req: TextSimilarityRequest): Promise<TextSimilarityResult>;
|
|
30
|
+
}
|
|
31
|
+
//# sourceMappingURL=evalJudge.d.ts.map
|
|
@@ -0,0 +1 @@
|
|
|
1
|
+
{"version":3,"file":"evalJudge.d.ts","sourceRoot":"","sources":["../../src/types/evalJudge.ts"],"names":[],"mappings":"AAAA;;wFAEwF;AACxF,MAAM,WAAW,qBAAqB;IACpC,QAAQ,EAAE,MAAM,CAAC;IACjB,MAAM,EAAE,MAAM,CAAC;IACf;;;;uDAImD;IACnD,OAAO,CAAC,EAAE,MAAM,CAAC,MAAM,EAAE,OAAO,CAAC,CAAC;CACnC;AAED,MAAM,WAAW,oBAAoB;IACnC,MAAM,EAAE,OAAO,CAAC;IAChB,sFAAsF;IACtF,MAAM,CAAC,EAAE,MAAM,CAAC;CACjB;AAED;;;;;;;;GAQG;AACH,MAAM,WAAW,mBAAmB;IAClC,OAAO,CAAC,GAAG,EAAE,qBAAqB,GAAG,OAAO,CAAC,oBAAoB,CAAC,CAAC;CACpE"}
|
|
@@ -0,0 +1 @@
|
|
|
1
|
+
{"version":3,"file":"evalJudge.js","sourceRoot":"","sources":["../../src/types/evalJudge.ts"],"names":[],"mappings":""}
|
package/dist/types/index.d.ts
CHANGED
|
@@ -1 +1 @@
|
|
|
1
|
-
{"version":3,"file":"index.d.ts","sourceRoot":"","sources":["../../src/types/index.ts"],"names":[],"mappings":"AAAA,cAAc,iBAAiB,CAAC;AAChC,cAAc,wBAAwB,CAAC;AACvC,cAAc,mBAAmB,CAAC;AAClC,cAAc,qBAAqB,CAAC;AACpC,cAAc,gBAAgB,CAAC;AAC/B,cAAc,uBAAuB,CAAC;AACtC,cAAc,uBAAuB,CAAC;AACtC,cAAc,eAAe,CAAC;AAC9B,cAAc,YAAY,CAAC;AAC3B,cAAc,eAAe,CAAC;AAC9B,cAAc,aAAa,CAAC;AAC5B,cAAc,cAAc,CAAC;AAC7B,cAAc,cAAc,CAAC"}
|
|
1
|
+
{"version":3,"file":"index.d.ts","sourceRoot":"","sources":["../../src/types/index.ts"],"names":[],"mappings":"AAAA,cAAc,iBAAiB,CAAC;AAChC,cAAc,wBAAwB,CAAC;AACvC,cAAc,mBAAmB,CAAC;AAClC,cAAc,qBAAqB,CAAC;AACpC,cAAc,gBAAgB,CAAC;AAC/B,cAAc,uBAAuB,CAAC;AACtC,cAAc,uBAAuB,CAAC;AACtC,cAAc,eAAe,CAAC;AAC9B,cAAc,YAAY,CAAC;AAC3B,cAAc,eAAe,CAAC;AAC9B,cAAc,aAAa,CAAC;AAC5B,cAAc,cAAc,CAAC;AAC7B,cAAc,gBAAgB,CAAC;AAC/B,cAAc,cAAc,CAAC"}
|
package/dist/types/index.js
CHANGED
package/dist/types/index.js.map
CHANGED
|
@@ -1 +1 @@
|
|
|
1
|
-
{"version":3,"file":"index.js","sourceRoot":"","sources":["../../src/types/index.ts"],"names":[],"mappings":"AAAA,cAAc,iBAAiB,CAAC;AAChC,cAAc,wBAAwB,CAAC;AACvC,cAAc,mBAAmB,CAAC;AAClC,cAAc,qBAAqB,CAAC;AACpC,cAAc,gBAAgB,CAAC;AAC/B,cAAc,uBAAuB,CAAC;AACtC,cAAc,uBAAuB,CAAC;AACtC,cAAc,eAAe,CAAC;AAC9B,cAAc,YAAY,CAAC;AAC3B,cAAc,eAAe,CAAC;AAC9B,cAAc,aAAa,CAAC;AAC5B,cAAc,cAAc,CAAC;AAC7B,cAAc,cAAc,CAAC"}
|
|
1
|
+
{"version":3,"file":"index.js","sourceRoot":"","sources":["../../src/types/index.ts"],"names":[],"mappings":"AAAA,cAAc,iBAAiB,CAAC;AAChC,cAAc,wBAAwB,CAAC;AACvC,cAAc,mBAAmB,CAAC;AAClC,cAAc,qBAAqB,CAAC;AACpC,cAAc,gBAAgB,CAAC;AAC/B,cAAc,uBAAuB,CAAC;AACtC,cAAc,uBAAuB,CAAC;AACtC,cAAc,eAAe,CAAC;AAC9B,cAAc,YAAY,CAAC;AAC3B,cAAc,eAAe,CAAC;AAC9B,cAAc,aAAa,CAAC;AAC5B,cAAc,cAAc,CAAC;AAC7B,cAAc,gBAAgB,CAAC;AAC/B,cAAc,cAAc,CAAC"}
|