observability-toolkit 2.0.0 → 2.1.0

package/dist/lib/agent-judge/agent-judge-consensus.test.js

@@ -0,0 +1,170 @@
+import { describe, it } from 'vitest';
+import assert from 'node:assert/strict';
+import { z } from 'zod';
+import { calculateVariance, calculateMedian, collectiveConsensus, MAX_CONCURRENT_EVALUATORS, DEFAULT_CONVERGENCE_THRESHOLD, } from './agent-as-judge.js';
+import { InputValidationError } from '../core/input-validator.js';
+import { P_VALUE_STRICT, PROBABILITY_FIFTY_PERCENT, PROBABILITY_EIGHTY_PERCENT, COUNT_EIGHT, COUNT_FIVE, COUNT_FOUR, COUNT_NINE, COUNT_SEVEN, COUNT_SIX, COUNT_THREE, COUNT_TWO, } from '../../test-helpers/test-constants.js';
+import { TEST_SCORE_PASSING, TEST_SCORE_STRONG, TEST_SCORE_VERY_LOW, } from '../quality/quality-test-constants.js';
+const TEST_ZERO = 0;
+const TEST_ONE = 1;
+const VARIANCE_NUMERATOR = 32;
+const MEDIAN_EVEN_EXPECTED = 2.5;
+const STEP_SCORE_NEAR_STRONG = 0.82;
+const STEP_SCORE_NEAR_GOOD = 0.79;
+const TEST_CONVERGENCE_THRESHOLD_TIGHT = 0.01;
+const TEST_CONVERGENCE_THRESHOLD_STRICT = 0.0001;
+const MAX_ROUNDS_STRESS_TEST = 100;
+const VARIANCE_SAMPLE_VALUES = [
+    COUNT_TWO,
+    COUNT_FOUR,
+    COUNT_FOUR,
+    COUNT_FOUR,
+    COUNT_FIVE,
+    COUNT_FIVE,
+    COUNT_SEVEN,
+    COUNT_NINE,
+];
+const VARIANCE_TWO_POINT_VALUES = [COUNT_TWO, COUNT_SIX];
+describe('agent-judge-consensus', () => {
+    describe('calculateVariance', () => {
+        it('should return 0 for empty array', () => {
+            assert.equal(calculateVariance([]), TEST_ZERO);
+        });
+        it('should return 0 for single value', () => {
+            assert.equal(calculateVariance([COUNT_FIVE]), TEST_ZERO);
+        });
+        it('should return 0 for identical values', () => {
+            assert.equal(calculateVariance([
+                COUNT_THREE,
+                COUNT_THREE,
+                COUNT_THREE,
+                COUNT_THREE,
+            ]), TEST_ZERO);
+        });
+        it('should calculate sample variance with Bessel correction (M7)', () => {
+            // Values: [2, 4, 4, 4, 5, 5, 7, 9], n=8, mean = 5
+            // Sum of squared diffs = (2-5)^2 + (4-5)^2*3 + (5-5)^2*2 + (7-5)^2 + (9-5)^2
+            //                      = 9 + 3 + 0 + 4 + 16 = 32
+            // Sample variance (Bessel's correction) = 32 / (n-1) = 32/7 ≈ 4.571
+            const result = calculateVariance(VARIANCE_SAMPLE_VALUES);
+            assert.ok(Math.abs(result - VARIANCE_NUMERATOR / COUNT_SEVEN) < P_VALUE_STRICT, `Expected ~4.571, got ${result}`);
+        });
+        it('should calculate variance for two values', () => {
+            // Values: [2, 6], mean = 4
+            // Sum of squared diffs = (2-4)^2 + (6-4)^2 = 4 + 4 = 8
+            // Sample variance = 8 / (2-1) = 8
+            const result = calculateVariance(VARIANCE_TWO_POINT_VALUES);
+            assert.equal(result, COUNT_EIGHT);
+        });
+    });
+    describe('calculateMedian', () => {
+        it('should return 0 for empty array', () => {
+            assert.equal(calculateMedian([]), TEST_ZERO);
+        });
+        it('should return value for single element', () => {
+            assert.equal(calculateMedian([COUNT_FIVE]), COUNT_FIVE);
+        });
+        it('should calculate median for odd count', () => {
+            assert.equal(calculateMedian([TEST_ONE, COUNT_THREE, COUNT_FIVE]), COUNT_THREE);
+        });
+        it('should calculate median for even count', () => {
+            assert.equal(calculateMedian([TEST_ONE, COUNT_TWO, COUNT_THREE, COUNT_FOUR]), MEDIAN_EVEN_EXPECTED);
+        });
+        it('should handle unsorted input', () => {
+            assert.equal(calculateMedian([COUNT_FIVE, TEST_ONE, COUNT_THREE]), COUNT_THREE);
+        });
+    });
+    // ============================================================================
+    // Consensus Tests
+    // ============================================================================
+    describe('collectiveConsensus', () => {
+        const validEvaluand = { input: 'test', output: 'result' };
+        const config = {
+            rounds: COUNT_THREE,
+            convergenceThreshold: DEFAULT_CONVERGENCE_THRESHOLD,
+        };
+        it('should throw for invalid evaluand', async () => {
+            await assert.rejects(collectiveConsensus({ input: '', output: 'test' }, [], config), z.ZodError);
+        });
+        // Judge-count constraints remain as InputValidationError (function-type
+        // validation cannot be expressed in Zod). Config field constraints (rounds,
+        // convergenceThreshold) are enforced by consensusConfigSchema → z.ZodError.
+        it('should throw for too many judges (H2)', async () => {
+            const judges = Array(MAX_CONCURRENT_EVALUATORS + 1).fill({
+                id: 'judge',
+                evaluate: async () => PROBABILITY_FIFTY_PERCENT,
+            });
+            await assert.rejects(collectiveConsensus(validEvaluand, judges, config), InputValidationError);
+        });
+        it('should throw for empty judges array', async () => {
+            await assert.rejects(collectiveConsensus(validEvaluand, [], config), InputValidationError);
+        });
+        it('should throw for rounds < 1 (M9)', async () => {
+            const judges = [{ id: 'j1', evaluate: async () => PROBABILITY_FIFTY_PERCENT }];
+            await assert.rejects(collectiveConsensus(validEvaluand, judges, { ...config, rounds: TEST_ZERO }), z.ZodError);
+        });
+        it('should reach consensus with agreeing judges', async () => {
+            const judges = [
+                { id: 'j1', evaluate: async () => PROBABILITY_EIGHTY_PERCENT },
+                { id: 'j2', evaluate: async () => STEP_SCORE_NEAR_STRONG },
+                { id: 'j3', evaluate: async () => STEP_SCORE_NEAR_GOOD },
+            ];
+            const result = await collectiveConsensus(validEvaluand, judges, config);
+            assert.ok(result.converged);
+            assert.ok(result.finalScore > TEST_SCORE_PASSING);
+        });
+        it('should handle disagreeing judges', async () => {
+            const judges = [
+                { id: 'j1', evaluate: async () => TEST_SCORE_VERY_LOW },
+                { id: 'j2', evaluate: async () => PROBABILITY_EIGHTY_PERCENT },
+            ];
+            const result = await collectiveConsensus(validEvaluand, judges, {
+                rounds: COUNT_TWO,
+                convergenceThreshold: TEST_CONVERGENCE_THRESHOLD_TIGHT, // Very tight
+            });
+            assert.equal(result.converged, false);
+        });
+        it('should handle judge failures gracefully (H3)', async () => {
+            const judges = [
+                { id: 'j1', evaluate: async () => PROBABILITY_EIGHTY_PERCENT },
+                {
+                    id: 'j2',
+                    evaluate: async () => {
+                        throw new Error('Judge failed');
+                    },
+                },
+                { id: 'j3', evaluate: async () => TEST_SCORE_STRONG },
+            ];
+            const result = await collectiveConsensus(validEvaluand, judges, config);
+            // Should succeed with 2 working judges
+            assert.ok(result.finalScore > TEST_ZERO);
+        });
+        it('should throw when all judges fail (H3)', async () => {
+            const judges = [
+                {
+                    id: 'j1',
+                    evaluate: async () => {
+                        throw new Error('Failed 1');
+                    },
+                },
+                {
+                    id: 'j2',
+                    evaluate: async () => {
+                        throw new Error('Failed 2');
+                    },
+                },
+            ];
+            await assert.rejects(collectiveConsensus(validEvaluand, judges, config), /All judge evaluations failed/);
+        });
+        it('should reject rounds exceeding MAX_CONSENSUS_ROUNDS via schema', async () => {
+            const judges = [
+                { id: 'j1', evaluate: async () => PROBABILITY_FIFTY_PERCENT },
+            ];
+            await assert.rejects(collectiveConsensus(validEvaluand, judges, {
+                rounds: MAX_ROUNDS_STRESS_TEST,
+                convergenceThreshold: TEST_CONVERGENCE_THRESHOLD_STRICT,
+            }), z.ZodError);
+        });
+    });
+});
+//# sourceMappingURL=agent-judge-consensus.test.js.map

package/dist/lib/agent-judge/agent-judge-consensus.test.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"agent-judge-consensus.test.js","sourceRoot":"","sources":["../../../src/lib/agent-judge/agent-judge-consensus.test.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,QAAQ,EAAE,EAAE,EAAE,MAAM,QAAQ,CAAC;AACtC,OAAO,MAAM,MAAM,oBAAoB,CAAC;AACxC,OAAO,EAAE,CAAC,EAAE,MAAM,KAAK,CAAC;AAExB,OAAO,EACL,iBAAiB,EACjB,eAAe,EACf,mBAAmB,EACnB,yBAAyB,EACzB,6BAA6B,GAG9B,MAAM,qBAAqB,CAAC;AAE7B,OAAO,EAAE,oBAAoB,EAAE,MAAM,4BAA4B,CAAC;AAClE,OAAO,EACL,cAAc,EACd,yBAAyB,EACzB,0BAA0B,EAC1B,WAAW,EACX,UAAU,EACV,UAAU,EACV,UAAU,EACV,WAAW,EACX,SAAS,EACT,WAAW,EACX,SAAS,GACV,MAAM,sCAAsC,CAAC;AAC9C,OAAO,EACL,kBAAkB,EAClB,iBAAiB,EACjB,mBAAmB,GACpB,MAAM,sCAAsC,CAAC;AAE9C,MAAM,SAAS,GAAG,CAAC,CAAC;AACpB,MAAM,QAAQ,GAAG,CAAC,CAAC;AACnB,MAAM,kBAAkB,GAAG,EAAE,CAAC;AAC9B,MAAM,oBAAoB,GAAG,GAAG,CAAC;AACjC,MAAM,sBAAsB,GAAG,IAAI,CAAC;AACpC,MAAM,oBAAoB,GAAG,IAAI,CAAC;AAClC,MAAM,gCAAgC,GAAG,IAAI,CAAC;AAC9C,MAAM,iCAAiC,GAAG,MAAM,CAAC;AACjD,MAAM,sBAAsB,GAAG,GAAG,CAAC;AACnC,MAAM,sBAAsB,GAAG;IAC7B,SAAS;IACT,UAAU;IACV,UAAU;IACV,UAAU;IACV,UAAU;IACV,UAAU;IACV,WAAW;IACX,UAAU;CACX,CAAC;AACF,MAAM,yBAAyB,GAAG,CAAC,SAAS,EAAE,SAAS,CAAC,CAAC;AAEzD,QAAQ,CAAC,uBAAuB,EAAE,GAAG,EAAE;IACrC,QAAQ,CAAC,mBAAmB,EAAE,GAAG,EAAE;QACjC,EAAE,CAAC,iCAAiC,EAAE,GAAG,EAAE;YACzC,MAAM,CAAC,KAAK,CAAC,iBAAiB,CAAC,EAAE,CAAC,EAAE,SAAS,CAAC,CAAC;QACjD,CAAC,CAAC,CAAC;QAEH,EAAE,CAAC,kCAAkC,EAAE,GAAG,EAAE;YAC1C,MAAM,CAAC,KAAK,CAAC,iBAAiB,CAAC,CAAC,UAAU,CAAC,CAAC,EAAE,SAAS,CAAC,CAAC;QAC3D,CAAC,CAAC,CAAC;QAEH,EAAE,CAAC,sCAAsC,EAAE,GAAG,EAAE;YAC9C,MAAM,CAAC,KAAK,CACV,iBAAiB,CAAC;gBAChB,WAAW;gBACX,WAAW;gBACX,WAAW;gBACX,WAAW;aACZ,CAAC,EACF,SAAS,CACV,CAAC;QACJ,CAAC,CAAC,CAAC;QAEH,EAAE,CAAC,8DAA8D,EAAE,GAAG,EAAE;YACtE,kDAAkD;YAClD,6EAA6E;YAC7E,iDAAiD;YACjD,oEAAoE;YACpE,MAAM,MAAM,GAAG,iBAAiB,CAAC,sBAAsB,CAAC,CAAC;YACzD,MAAM,CAAC,EAAE,CACP,IAAI,CAAC,GAAG,CAAC,MAAM,GAAG,kBAAkB,GAAG,WAAW,CAAC,GAAG,cAAc,EACpE,wBAAwB,MAAM,EAAE,CACjC,CAAC;QACJ,CAAC,CAAC,CAAC;QAEH,EAAE,CAAC,0CAA0C,EAAE,GAAG,EAAE;YAClD,2BAA2B;YAC3B,uDAAuD;YACvD,kCAAkC;YAClC,MAAM,MAAM,GAAG,iBAAiB,CAAC,yBAAyB,CAAC,CAAC;YAC5D,MAAM,CAAC,KAAK,CAAC,MAAM,EAAE,WAAW,CAAC,CAAC;QACpC,CAAC,CAAC,CAAC;IACL,CAAC,CAAC,CAAC;IAEH,QAAQ,CAAC,iBAAiB,EAAE,GAAG,EAAE;QAC/B,EAAE,CAAC,iCAAiC,EAAE,GAAG,EAAE;YACzC,MAAM,CAAC,KAAK,CAAC,eAAe,CAAC,EAAE,CAAC,EAAE,SAAS,CAAC,CAAC;QAC/C,CAAC,CAAC,CAAC;QAEH,EAAE,CAAC,wCAAwC,EAAE,GAAG,EAAE;YAChD,MAAM,CAAC,KAAK,CAAC,eAAe,CAAC,CAAC,UAAU,CAAC,CAAC,EAAE,UAAU,CAAC,CAAC;QAC1D,CAAC,CAAC,CAAC;QAEH,EAAE,CAAC,uCAAuC,EAAE,GAAG,EAAE;YAC/C,MAAM,CAAC,KAAK,CAAC,eAAe,CAAC,CAAC,QAAQ,EAAE,WAAW,EAAE,UAAU,CAAC,CAAC,EAAE,WAAW,CAAC,CAAC;QAClF,CAAC,CAAC,CAAC;QAEH,EAAE,CAAC,wCAAwC,EAAE,GAAG,EAAE;YAChD,MAAM,CAAC,KAAK,CACV,eAAe,CAAC,CAAC,QAAQ,EAAE,SAAS,EAAE,WAAW,EAAE,UAAU,CAAC,CAAC,EAC/D,oBAAoB,CACrB,CAAC;QACJ,CAAC,CAAC,CAAC;QAEH,EAAE,CAAC,8BAA8B,EAAE,GAAG,EAAE;YACtC,MAAM,CAAC,KAAK,CAAC,eAAe,CAAC,CAAC,UAAU,EAAE,QAAQ,EAAE,WAAW,CAAC,CAAC,EAAE,WAAW,CAAC,CAAC;QAClF,CAAC,CAAC,CAAC;IACL,CAAC,CAAC,CAAC;IAEH,+EAA+E;IAC/E,kBAAkB;IAClB,+EAA+E;IAE/E,QAAQ,CAAC,qBAAqB,EAAE,GAAG,EAAE;QACnC,MAAM,aAAa,GAAa,EAAE,KAAK,EAAE,MAAM,EAAE,MAAM,EAAE,QAAQ,EAAE,CAAC;QACpE,MAAM,MAAM,GAAoB;YAC9B,MAAM,EAAE,WAAW;YACnB,oBAAoB,EAAE,6BAA6B;SACpD,CAAC;QAEF,EAAE,CAAC,mCAAmC,EAAE,KAAK,IAAI,EAAE;YACjD,MAAM,MAAM,CAAC,OAAO,CAClB,mBAAmB,CAAC,EAAE,KAAK,EAAE,EAAE,EAAE,MAAM,EAAE,MAAM,EAAE,EAAE,EAAE,EAAE,MAAM,CAAC,EAC9D,CAAC,CAAC,QAAQ,CACX,CAAC;QACJ,CAAC,CAAC,CAAC;QAEH,wEAAwE;QACxE,4EAA4E;QAC5E,4EAA4E;QAC5E,EAAE,CAAC,uCAAuC,EAAE,KAAK,IAAI,EAAE;YACrD,MAAM,MAAM,GAAG,KAAK,CAAC,yBAAyB,GAAG,CAAC,CAAC,CAAC,IAAI,CAAC;gBACvD,EAAE,EAAE,OAAO;gBACX,QAAQ,EAAE,KAAK,IAAI,EAAE,CAAC,yBAAyB;aAChD,CAAC,CAAC;YACH,MAAM,MAAM,CAAC,OAAO,CAClB,mBAAmB,CAAC,aAAa,EAAE,MAAM,EAAE,MAAM,CAAC,EAClD,oBAAoB,CACrB,CAAC;QACJ,CAAC,CAAC,CAAC;QAEH,EAAE,CAAC,qCAAqC,EAAE,KAAK,IAAI,EAAE;YACnD,MAAM,MAAM,CAAC,OAAO,CAClB,mBAAmB,CAAC,aAAa,EAAE,EAAE,EAAE,MAAM,CAAC,EAC9C,oBAAoB,CACrB,CAAC;QACJ,CAAC,CAAC,CAAC;QAEH,EAAE,CAAC,kCAAkC,EAAE,KAAK,IAAI,EAAE;YAChD,MAAM,MAAM,GAAG,CAAC,EAAE,EAAE,EAAE,IAAI,EAAE,QAAQ,EAAE,KAAK,IAAI,EAAE,CAAC,yBAAyB,EAAE,CAAC,CAAC;YAC/E,MAAM,MAAM,CAAC,OAAO,CAClB,mBAAmB,CAAC,aAAa,EAAE,MAAM,EAAE,EAAE,GAAG,MAAM,EAAE,MAAM,EAAE,SAAS,EAAE,CAAC,EAC5E,CAAC,CAAC,QAAQ,CACX,CAAC;QACJ,CAAC,CAAC,CAAC;QAEH,EAAE,CAAC,6CAA6C,EAAE,KAAK,IAAI,EAAE;YAC3D,MAAM,MAAM,GAAG;gBACb,EAAE,EAAE,EAAE,IAAI,EAAE,QAAQ,EAAE,KAAK,IAAI,EAAE,CAAC,0BAA0B,EAAE;gBAC9D,EAAE,EAAE,EAAE,IAAI,EAAE,QAAQ,EAAE,KAAK,IAAI,EAAE,CAAC,sBAAsB,EAAE;gBAC1D,EAAE,EAAE,EAAE,IAAI,EAAE,QAAQ,EAAE,KAAK,IAAI,EAAE,CAAC,oBAAoB,EAAE;aACzD,CAAC;YACF,MAAM,MAAM,GAAG,MAAM,mBAAmB,CAAC,aAAa,EAAE,MAAM,EAAE,MAAM,CAAC,CAAC;YACxE,MAAM,CAAC,EAAE,CAAC,MAAM,CAAC,SAAS,CAAC,CAAC;YAC5B,MAAM,CAAC,EAAE,CAAC,MAAM,CAAC,UAAU,GAAG,kBAAkB,CAAC,CAAC;QACpD,CAAC,CAAC,CAAC;QAEH,EAAE,CAAC,kCAAkC,EAAE,KAAK,IAAI,EAAE;YAChD,MAAM,MAAM,GAAG;gBACb,EAAE,EAAE,EAAE,IAAI,EAAE,QAAQ,EAAE,KAAK,IAAI,EAAE,CAAC,mBAAmB,EAAE;gBACvD,EAAE,EAAE,EAAE,IAAI,EAAE,QAAQ,EAAE,KAAK,IAAI,EAAE,CAAC,0BAA0B,EAAE;aAC/D,CAAC;YACF,MAAM,MAAM,GAAG,MAAM,mBAAmB,CAAC,aAAa,EAAE,MAAM,EAAE;gBAC9D,MAAM,EAAE,SAAS;gBACjB,oBAAoB,EAAE,gCAAgC,EAAE,aAAa;aACtE,CAAC,CAAC;YACH,MAAM,CAAC,KAAK,CAAC,MAAM,CAAC,SAAS,EAAE,KAAK,CAAC,CAAC;QACxC,CAAC,CAAC,CAAC;QAEH,EAAE,CAAC,8CAA8C,EAAE,KAAK,IAAI,EAAE;YAC5D,MAAM,MAAM,GAAG;gBACb,EAAE,EAAE,EAAE,IAAI,EAAE,QAAQ,EAAE,KAAK,IAAI,EAAE,CAAC,0BAA0B,EAAE;gBAC9D;oBACE,EAAE,EAAE,IAAI;oBACR,QAAQ,EAAE,KAAK,IAAI,EAAE;wBACnB,MAAM,IAAI,KAAK,CAAC,cAAc,CAAC,CAAC;oBAClC,CAAC;iBACF;gBACD,EAAE,EAAE,EAAE,IAAI,EAAE,QAAQ,EAAE,KAAK,IAAI,EAAE,CAAC,iBAAiB,EAAE;aACtD,CAAC;YACF,MAAM,MAAM,GAAG,MAAM,mBAAmB,CAAC,aAAa,EAAE,MAAM,EAAE,MAAM,CAAC,CAAC;YACxE,uCAAuC;YACvC,MAAM,CAAC,EAAE,CAAC,MAAM,CAAC,UAAU,GAAG,SAAS,CAAC,CAAC;QAC3C,CAAC,CAAC,CAAC;QAEH,EAAE,CAAC,wCAAwC,EAAE,KAAK,IAAI,EAAE;YACtD,MAAM,MAAM,GAAG;gBACb;oBACE,EAAE,EAAE,IAAI;oBACR,QAAQ,EAAE,KAAK,IAAI,EAAE;wBACnB,MAAM,IAAI,KAAK,CAAC,UAAU,CAAC,CAAC;oBAC9B,CAAC;iBACF;gBACD;oBACE,EAAE,EAAE,IAAI;oBACR,QAAQ,EAAE,KAAK,IAAI,EAAE;wBACnB,MAAM,IAAI,KAAK,CAAC,UAAU,CAAC,CAAC;oBAC9B,CAAC;iBACF;aACF,CAAC;YACF,MAAM,MAAM,CAAC,OAAO,CAClB,mBAAmB,CAAC,aAAa,EAAE,MAAM,EAAE,MAAM,CAAC,EAClD,8BAA8B,CAC/B,CAAC;QACJ,CAAC,CAAC,CAAC;QAEH,EAAE,CAAC,gEAAgE,EAAE,KAAK,IAAI,EAAE;YAC9E,MAAM,MAAM,GAAG;gBACb,EAAE,EAAE,EAAE,IAAI,EAAE,QAAQ,EAAE,KAAK,IAAI,EAAE,CAAC,yBAAyB,EAAE;aAC9D,CAAC;YACF,MAAM,MAAM,CAAC,OAAO,CAClB,mBAAmB,CAAC,aAAa,EAAE,MAAM,EAAE;gBACzC,MAAM,EAAE,sBAAsB;gBAC9B,oBAAoB,EAAE,iCAAiC;aACxD,CAAC,EACF,CAAC,CAAC,QAAQ,CACX,CAAC;QACJ,CAAC,CAAC,CAAC;IACL,CAAC,CAAC,CAAC;AACL,CAAC,CAAC,CAAC"}

package/dist/lib/agent-judge/agent-judge-verification.d.ts

@@ -0,0 +1,89 @@
+/**
+ * Agent Judge Verification & Scoring Utilities
+ *
+ * Input validation, tool call verification, step scoring,
+ * and trajectory analysis.
+ *
+ * @see agent-as-judge.ts for core types
+ */
+import { type StepScore, type ToolVerification } from '../../backends/index.js';
+import { type AgentAction, type Evaluand } from './agent-as-judge.js';
+/**
+ * Validate a StepScore object using the canonical Zod schema.
+ * @throws {InputValidationError} If validation fails
+ */
+export declare function validateStepScore(stepScore: StepScore): void;
+/**
+ * Verify a single tool call against expected behavior.
+ *
+ * @param action - The agent action containing the tool call
+ * @param expectedTool - Expected tool name (optional)
+ * @param expectedArgs - Expected arguments (optional)
+ * @param actualResult - Actual result to compare against expected (optional)
+ * @param expectedResult - Expected result for comparison (optional)
+ * @returns ToolVerification result with weighted score:
+ *   - Tool selection: 40% weight
+ *   - Arguments: 30% weight (if expectedArgs provided)
+ *   - Result: 30% weight (if expectedResult provided)
+ * @throws {InputValidationError} If action is null, undefined, or not an object
+ */
+export declare function verifyToolCall(action: AgentAction, expectedTool?: string, expectedArgs?: Record<string, unknown>, actualResult?: unknown, expectedResult?: unknown): ToolVerification;
+/**
+ * Verify all tool calls in an agent trajectory.
+ *
+ * @param actions - List of agent actions
+ * @param expectedTools - Map of expected tool calls by index or toolCallId
+ * @returns Array of ToolVerification results
+ */
+export declare function verifyToolCalls(actions: AgentAction[], expectedTools?: Map<number | string, {
+    tool: string;
+    args?: Record<string, unknown>;
+    result?: unknown;
+}>): ToolVerification[];
+/**
+ * Score a single step in the agent trajectory.
+ *
+ * @param action - The agent action to score
+ * @param stepIndex - Index of this step (non-negative integer)
+ * @param evaluation - Score and explanation from LLM judge
+ * @returns StepScore result with score clamped to [0, 1]
+ * @throws {InputValidationError} If evaluation.score is not a finite number
+ * @throws {InputValidationError} If stepIndex is invalid (validated via validateStepScore)
+ */
+export declare function scoreStep(action: AgentAction, stepIndex: number, evaluation: {
+    score: number;
+    explanation?: string;
+}): StepScore;
+/**
+ * Calculate aggregate step scores from individual scores.
+ *
+ * @param stepScores - Array of step scores
+ * @param aggregation - Aggregation method ('average', 'weighted', 'min')
+ * @param weights - Weights for weighted aggregation (must match stepScores length, finite non-negative)
+ * @param options - Additional options
+ * @param options.emptyBehavior - What to return for empty arrays: 'vacuous' (1.0, default), 'zero' (0.0), or 'NaN'
+ * @returns Aggregated score (0-1)
+ * @throws {Error} If aggregation is 'weighted' but weights array is missing or wrong length
+ * @throws {Error} If any weight is negative, NaN, or Infinity
+ */
+export declare function aggregateStepScores(stepScores: StepScore[], aggregation?: 'average' | 'weighted' | 'min', weights?: number[], options?: {
+    emptyBehavior?: 'vacuous' | 'zero' | 'NaN';
+}): number;
+/**
+ * Calculate trajectory efficiency metrics.
+ *
+ * Uses JSON.stringify for redundancy detection. For large argument objects,
+ * consider implementing hash-based comparison for better performance.
+ *
+ * @param evaluand - The evaluand with actions
+ * @param optimalLength - Expected optimal trajectory length (optional)
+ * @returns Efficiency metrics
+ */
+export declare function analyzeTrajectory(evaluand: Evaluand, optimalLength?: number): {
+    length: number;
+    toolCallCount: number;
+    uniqueTools: number;
+    efficiencyRatio: number;
+    redundantActions: number;
+};
+//# sourceMappingURL=agent-judge-verification.d.ts.map

package/dist/lib/agent-judge/agent-judge-verification.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"agent-judge-verification.d.ts","sourceRoot":"","sources":["../../../src/lib/agent-judge/agent-judge-verification.ts"],"names":[],"mappings":"AAAA;;;;;;;GAOG;AAEH,OAAO,EAEL,KAAK,SAAS,EACd,KAAK,gBAAgB,EACtB,MAAM,yBAAyB,CAAC;AAGjC,OAAO,EAAE,KAAK,WAAW,EAAE,KAAK,QAAQ,EAAE,MAAM,qBAAqB,CAAC;AAQtE;;;GAGG;AACH,wBAAgB,iBAAiB,CAAC,SAAS,EAAE,SAAS,GAAG,IAAI,CAU5D;AAMD;;;;;;;;;;;;;GAaG;AACH,wBAAgB,cAAc,CAC5B,MAAM,EAAE,WAAW,EACnB,YAAY,CAAC,EAAE,MAAM,EACrB,YAAY,CAAC,EAAE,MAAM,CAAC,MAAM,EAAE,OAAO,CAAC,EACtC,YAAY,CAAC,EAAE,OAAO,EACtB,cAAc,CAAC,EAAE,OAAO,GACvB,gBAAgB,CAqDlB;AAED;;;;;;GAMG;AACH,wBAAgB,eAAe,CAC7B,OAAO,EAAE,WAAW,EAAE,EACtB,aAAa,CAAC,EAAE,GAAG,CAAC,MAAM,GAAG,MAAM,EAAE;IAAE,IAAI,EAAE,MAAM,CAAC;IAAC,IAAI,CAAC,EAAE,MAAM,CAAC,MAAM,EAAE,OAAO,CAAC,CAAC;IAAC,MAAM,CAAC,EAAE,OAAO,CAAA;CAAE,CAAC,GACvG,gBAAgB,EAAE,CAuBpB;AAMD;;;;;;;;;GASG;AACH,wBAAgB,SAAS,CACvB,MAAM,EAAE,WAAW,EACnB,SAAS,EAAE,MAAM,EACjB,UAAU,EAAE;IAAE,KAAK,EAAE,MAAM,CAAC;IAAC,WAAW,CAAC,EAAE,MAAM,CAAA;CAAE,GAClD,SAAS,CAuBX;AAED;;;;;;;;;;;GAWG;AACH,wBAAgB,mBAAmB,CACjC,UAAU,EAAE,SAAS,EAAE,EACvB,WAAW,GAAE,SAAS,GAAG,UAAU,GAAG,KAAiB,EACvD,OAAO,CAAC,EAAE,MAAM,EAAE,EAClB,OAAO,CAAC,EAAE;IAAE,aAAa,CAAC,EAAE,SAAS,GAAG,MAAM,GAAG,KAAK,CAAA;CAAE,GACvD,MAAM,CA8BR;AAMD;;;;;;;;;GASG;AACH,wBAAgB,iBAAiB,CAC/B,QAAQ,EAAE,QAAQ,EAClB,aAAa,CAAC,EAAE,MAAM,GACrB;IACD,MAAM,EAAE,MAAM,CAAC;IACf,aAAa,EAAE,MAAM,CAAC;IACtB,WAAW,EAAE,MAAM,CAAC;IACpB,eAAe,EAAE,MAAM,CAAC;IACxB,gBAAgB,EAAE,MAAM,CAAC;CAC1B,CAoCA"}

package/dist/lib/agent-judge/agent-judge-verification.js

@@ -0,0 +1,235 @@
+/**
+ * Agent Judge Verification & Scoring Utilities
+ *
+ * Input validation, tool call verification, step scoring,
+ * and trajectory analysis.
+ *
+ * @see agent-as-judge.ts for core types
+ */
+import { stepScoreSchema, } from '../../backends/index.js';
+import { MAX_TOOL_VERIFICATIONS } from '../core/constants.js';
+import { InputValidationError } from '../core/input-validator.js';
+import { TOOL_VERIFICATION_WEIGHTS } from './agent-eval-metrics.js';
+import equal from 'fast-deep-equal';
+// ============================================================================
+// Validation Utilities
+// ============================================================================
+/**
+ * Validate a StepScore object using the canonical Zod schema.
+ * @throws {InputValidationError} If validation fails
+ */
+export function validateStepScore(stepScore) {
+    const result = stepScoreSchema.safeParse(stepScore);
+    if (!result.success) {
+        const issue = result.error.issues[0];
+        throw new InputValidationError(issue.message, issue.path.join('.') || 'stepScore', issue.code);
+    }
+}
+// ============================================================================
+// Tool Verification Utilities
+// ============================================================================
+/**
+ * Verify a single tool call against expected behavior.
+ *
+ * @param action - The agent action containing the tool call
+ * @param expectedTool - Expected tool name (optional)
+ * @param expectedArgs - Expected arguments (optional)
+ * @param actualResult - Actual result to compare against expected (optional)
+ * @param expectedResult - Expected result for comparison (optional)
+ * @returns ToolVerification result with weighted score:
+ *   - Tool selection: 40% weight
+ *   - Arguments: 30% weight (if expectedArgs provided)
+ *   - Result: 30% weight (if expectedResult provided)
+ * @throws {InputValidationError} If action is null, undefined, or not an object
+ */
+export function verifyToolCall(action, expectedTool, expectedArgs, actualResult, expectedResult) {
+    // H1: Validate action parameter
+    if (!action || typeof action !== 'object') {
+        throw new InputValidationError('Action is required and must be an object', 'action', 'required');
+    }
+    const toolCorrect = expectedTool ? action.tool === expectedTool : true;
+    const argsCorrect = expectedArgs
+        ? equal(action.arguments, expectedArgs)
+        : true;
+    const resultCorrect = expectedResult !== undefined
+        ? equal(actualResult ?? action.result, expectedResult)
+        : undefined;
+    // Calculate weighted score
+    let score = 0;
+    let weights = 0;
+    // Tool selection: 40% weight
+    score += (toolCorrect ? TOOL_VERIFICATION_WEIGHTS.TOOL_CORRECT : 0);
+    weights += TOOL_VERIFICATION_WEIGHTS.TOOL_CORRECT;
+    // Arguments: 30% weight
+    if (expectedArgs !== undefined) {
+        score += (argsCorrect ? TOOL_VERIFICATION_WEIGHTS.ARGS_CORRECT : 0);
+        weights += TOOL_VERIFICATION_WEIGHTS.ARGS_CORRECT;
+    }
+    // Result: 30% weight (only if expected result provided)
+    if (expectedResult !== undefined) {
+        score += (resultCorrect ? TOOL_VERIFICATION_WEIGHTS.RESULT_CORRECT : 0);
+        weights += TOOL_VERIFICATION_WEIGHTS.RESULT_CORRECT;
+    }
+    const normalizedScore = score / weights;
+    return {
+        toolName: action.tool || 'unknown',
+        toolCallId: action.toolCallId,
+        toolCorrect,
+        argsCorrect,
+        resultCorrect,
+        score: normalizedScore,
+        expectedTool,
+        evidence: {
+            actualTool: action.tool,
+            actualArgs: action.arguments,
+            actualResult: actualResult ?? action.result,
+            expectedTool,
+            expectedArgs,
+            expectedResult,
+        },
+    };
+}
+/**
+ * Verify all tool calls in an agent trajectory.
+ *
+ * @param actions - List of agent actions
+ * @param expectedTools - Map of expected tool calls by index or toolCallId
+ * @returns Array of ToolVerification results
+ */
+export function verifyToolCalls(actions, expectedTools) {
+    const verifications = [];
+    for (let i = 0; i < actions.length && verifications.length < MAX_TOOL_VERIFICATIONS; i++) {
+        const action = actions[i];
+        if (action.type !== 'tool_call' || !action.tool)
+            continue;
+        // Check for expected tool by index or toolCallId
+        const expected = expectedTools?.get(i) ??
+            (action.toolCallId ? expectedTools?.get(action.toolCallId) : undefined);
+        const verification = verifyToolCall(action, expected?.tool, expected?.args, undefined, expected?.result);
+        verifications.push(verification);
+    }
+    return verifications;
+}
+// ============================================================================
+// Step Scoring Utilities
+// ============================================================================
+/**
+ * Score a single step in the agent trajectory.
+ *
+ * @param action - The agent action to score
+ * @param stepIndex - Index of this step (non-negative integer)
+ * @param evaluation - Score and explanation from LLM judge
+ * @returns StepScore result with score clamped to [0, 1]
+ * @throws {InputValidationError} If evaluation.score is not a finite number
+ * @throws {InputValidationError} If stepIndex is invalid (validated via validateStepScore)
+ */
+export function scoreStep(action, stepIndex, evaluation) {
+    // H4: Validate score is a finite number before clamping
+    if (typeof evaluation.score !== 'number' || !Number.isFinite(evaluation.score)) {
+        throw new InputValidationError('Evaluation score must be a finite number', 'score', 'type');
+    }
+    const stepScore = {
+        step: stepIndex,
+        score: Math.max(0, Math.min(1, evaluation.score)), // Clamp to [0, 1]
+        explanation: evaluation.explanation,
+        evidence: {
+            actionType: action.type,
+            tool: action.tool,
+            reasoning: action.reasoning,
+        },
+    };
+    validateStepScore(stepScore);
+    return stepScore;
+}
+/**
+ * Calculate aggregate step scores from individual scores.
+ *
+ * @param stepScores - Array of step scores
+ * @param aggregation - Aggregation method ('average', 'weighted', 'min')
+ * @param weights - Weights for weighted aggregation (must match stepScores length, finite non-negative)
+ * @param options - Additional options
+ * @param options.emptyBehavior - What to return for empty arrays: 'vacuous' (1.0, default), 'zero' (0.0), or 'NaN'
+ * @returns Aggregated score (0-1)
+ * @throws {Error} If aggregation is 'weighted' but weights array is missing or wrong length
+ * @throws {Error} If any weight is negative, NaN, or Infinity
+ */
+export function aggregateStepScores(stepScores, aggregation = 'average', weights, options) {
+    if (stepScores.length === 0) {
+        const behavior = options?.emptyBehavior ?? 'vacuous';
+        if (behavior === 'zero')
+            return 0;
+        if (behavior === 'NaN')
+            return NaN;
+        return 1; // vacuous: no steps to fail = perfect
+    }
+    switch (aggregation) {
+        case 'average':
+            return stepScores.reduce((sum, s) => sum + s.score, 0) / stepScores.length;
+        case 'weighted': {
+            if (!weights || weights.length !== stepScores.length) {
+                throw new Error('Weights required for weighted aggregation');
+            }
+            // L8/M1: Validate weights are finite non-negative numbers
+            for (let i = 0; i < weights.length; i++) {
+                if (weights[i] < 0 || !Number.isFinite(weights[i])) {
+                    throw new Error(`Invalid weight at index ${i}: ${weights[i]}. Weights must be finite non-negative numbers.`);
+                }
+            }
+            const totalWeight = weights.reduce((sum, w) => sum + w, 0);
+            if (totalWeight === 0)
+                return 0;
+            return stepScores.reduce((sum, s, i) => sum + s.score * weights[i], 0) / totalWeight;
+        }
+        case 'min':
+            return Math.min(...stepScores.map(s => s.score));
+    }
+}
+// ============================================================================
+// Trajectory Analysis Utilities
+// ============================================================================
+/**
+ * Calculate trajectory efficiency metrics.
+ *
+ * Uses JSON.stringify for redundancy detection. For large argument objects,
+ * consider implementing hash-based comparison for better performance.
+ *
+ * @param evaluand - The evaluand with actions
+ * @param optimalLength - Expected optimal trajectory length (optional)
+ * @returns Efficiency metrics
+ */
+export function analyzeTrajectory(evaluand, optimalLength) {
+    const actions = evaluand.actions || [];
+    const length = actions.length;
+    const toolCalls = actions.filter(a => a.type === 'tool_call');
+    const toolCallCount = toolCalls.length;
+    const uniqueTools = new Set(toolCalls.map(a => a.tool).filter(Boolean)).size;
+    // Calculate efficiency ratio (if optimal length known)
+    const efficiencyRatio = optimalLength && optimalLength > 0
+        ? Math.min(1, optimalLength / Math.max(length, 1))
+        : 1;
+    // Detect redundant actions (same tool with same args)
+    const seenToolCalls = new Set();
+    let redundantActions = 0;
+    for (const action of toolCalls) {
+        // M2: Wrap in try-catch to handle circular references or non-serializable values
+        try {
+            const key = JSON.stringify({ tool: action.tool, args: action.arguments });
+            if (seenToolCalls.has(key)) {
+                redundantActions++;
+            }
+            seenToolCalls.add(key);
+        }
+        catch {
+            // Skip this action for redundancy detection if serialization fails
+            continue;
+        }
+    }
+    return {
+        length,
+        toolCallCount,
+        uniqueTools,
+        efficiencyRatio,
+        redundantActions,
+    };
+}
+//# sourceMappingURL=agent-judge-verification.js.map

package/dist/lib/agent-judge/agent-judge-verification.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"agent-judge-verification.js","sourceRoot":"","sources":["../../../src/lib/agent-judge/agent-judge-verification.ts"],"names":[],"mappings":"AAAA;;;;;;;GAOG;AAEH,OAAO,EACL,eAAe,GAGhB,MAAM,yBAAyB,CAAC;AACjC,OAAO,EAAE,sBAAsB,EAAE,MAAM,sBAAsB,CAAC;AAC9D,OAAO,EAAE,oBAAoB,EAAE,MAAM,4BAA4B,CAAC;AAElE,OAAO,EAAE,yBAAyB,EAAE,MAAM,yBAAyB,CAAC;AACpE,OAAO,KAAK,MAAM,iBAAiB,CAAC;AAEpC,+EAA+E;AAC/E,uBAAuB;AACvB,+EAA+E;AAE/E;;;GAGG;AACH,MAAM,UAAU,iBAAiB,CAAC,SAAoB;IACpD,MAAM,MAAM,GAAG,eAAe,CAAC,SAAS,CAAC,SAAS,CAAC,CAAC;IACpD,IAAI,CAAC,MAAM,CAAC,OAAO,EAAE,CAAC;QACpB,MAAM,KAAK,GAAG,MAAM,CAAC,KAAK,CAAC,MAAM,CAAC,CAAC,CAAC,CAAC;QACrC,MAAM,IAAI,oBAAoB,CAC5B,KAAK,CAAC,OAAO,EACb,KAAK,CAAC,IAAI,CAAC,IAAI,CAAC,GAAG,CAAC,IAAI,WAAW,EACnC,KAAK,CAAC,IAAI,CACX,CAAC;IACJ,CAAC;AACH,CAAC;AAED,+EAA+E;AAC/E,8BAA8B;AAC9B,+EAA+E;AAE/E;;;;;;;;;;;;;GAaG;AACH,MAAM,UAAU,cAAc,CAC5B,MAAmB,EACnB,YAAqB,EACrB,YAAsC,EACtC,YAAsB,EACtB,cAAwB;IAExB,gCAAgC;IAChC,IAAI,CAAC,MAAM,IAAI,OAAO,MAAM,KAAK,QAAQ,EAAE,CAAC;QAC1C,MAAM,IAAI,oBAAoB,CAAC,0CAA0C,EAAE,QAAQ,EAAE,UAAU,CAAC,CAAC;IACnG,CAAC;IAED,MAAM,WAAW,GAAG,YAAY,CAAC,CAAC,CAAC,MAAM,CAAC,IAAI,KAAK,YAAY,CAAC,CAAC,CAAC,IAAI,CAAC;IACvE,MAAM,WAAW,GAAG,YAAY;QAC9B,CAAC,CAAC,KAAK,CAAC,MAAM,CAAC,SAAS,EAAE,YAAY,CAAC;QACvC,CAAC,CAAC,IAAI,CAAC;IACT,MAAM,aAAa,GAAG,cAAc,KAAK,SAAS;QAChD,CAAC,CAAC,KAAK,CAAC,YAAY,IAAI,MAAM,CAAC,MAAM,EAAE,cAAc,CAAC;QACtD,CAAC,CAAC,SAAS,CAAC;IAEd,2BAA2B;IAC3B,IAAI,KAAK,GAAG,CAAC,CAAC;IACd,IAAI,OAAO,GAAG,CAAC,CAAC;IAEhB,6BAA6B;IAC7B,KAAK,IAAI,CAAC,WAAW,CAAC,CAAC,CAAC,yBAAyB,CAAC,YAAY,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC;IACpE,OAAO,IAAI,yBAAyB,CAAC,YAAY,CAAC;IAElD,wBAAwB;IACxB,IAAI,YAAY,KAAK,SAAS,EAAE,CAAC;QAC/B,KAAK,IAAI,CAAC,WAAW,CAAC,CAAC,CAAC,yBAAyB,CAAC,YAAY,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC;QACpE,OAAO,IAAI,yBAAyB,CAAC,YAAY,CAAC;IACpD,CAAC;IAED,wDAAwD;IACxD,IAAI,cAAc,KAAK,SAAS,EAAE,CAAC;QACjC,KAAK,IAAI,CAAC,aAAa,CAAC,CAAC,CAAC,yBAAyB,CAAC,cAAc,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC;QACxE,OAAO,IAAI,yBAAyB,CAAC,cAAc,CAAC;IACtD,CAAC;IAED,MAAM,eAAe,GAAG,KAAK,GAAG,OAAO,CAAC;IAExC,OAAO;QACL,QAAQ,EAAE,MAAM,CAAC,IAAI,IAAI,SAAS;QAClC,UAAU,EAAE,MAAM,CAAC,UAAU;QAC7B,WAAW;QACX,WAAW;QACX,aAAa;QACb,KAAK,EAAE,eAAe;QACtB,YAAY;QACZ,QAAQ,EAAE;YACR,UAAU,EAAE,MAAM,CAAC,IAAI;YACvB,UAAU,EAAE,MAAM,CAAC,SAAS;YAC5B,YAAY,EAAE,YAAY,IAAI,MAAM,CAAC,MAAM;YAC3C,YAAY;YACZ,YAAY;YACZ,cAAc;SACf;KACF,CAAC;AACJ,CAAC;AAED;;;;;;GAMG;AACH,MAAM,UAAU,eAAe,CAC7B,OAAsB,EACtB,aAAwG;IAExG,MAAM,aAAa,GAAuB,EAAE,CAAC;IAE7C,KAAK,IAAI,CAAC,GAAG,CAAC,EAAE,CAAC,GAAG,OAAO,CAAC,MAAM,IAAI,aAAa,CAAC,MAAM,GAAG,sBAAsB,EAAE,CAAC,EAAE,EAAE,CAAC;QACzF,MAAM,MAAM,GAAG,OAAO,CAAC,CAAC,CAAC,CAAC;QAC1B,IAAI,MAAM,CAAC,IAAI,KAAK,WAAW,IAAI,CAAC,MAAM,CAAC,IAAI;YAAE,SAAS;QAE1D,iDAAiD;QACjD,MAAM,QAAQ,GAAG,aAAa,EAAE,GAAG,CAAC,CAAC,CAAC;YACpC,CAAC,MAAM,CAAC,UAAU,CAAC,CAAC,CAAC,aAAa,EAAE,GAAG,CAAC,MAAM,CAAC,UAAU,CAAC,CAAC,CAAC,CAAC,SAAS,CAAC,CAAC;QAE1E,MAAM,YAAY,GAAG,cAAc,CACjC,MAAM,EACN,QAAQ,EAAE,IAAI,EACd,QAAQ,EAAE,IAAI,EACd,SAAS,EACT,QAAQ,EAAE,MAAM,CACjB,CAAC;QAEF,aAAa,CAAC,IAAI,CAAC,YAAY,CAAC,CAAC;IACnC,CAAC;IAED,OAAO,aAAa,CAAC;AACvB,CAAC;AAED,+EAA+E;AAC/E,yBAAyB;AACzB,+EAA+E;AAE/E;;;;;;;;;GASG;AACH,MAAM,UAAU,SAAS,CACvB,MAAmB,EACnB,SAAiB,EACjB,UAAmD;IAEnD,wDAAwD;IACxD,IAAI,OAAO,UAAU,CAAC,KAAK,KAAK,QAAQ,IAAI,CAAC,MAAM,CAAC,QAAQ,CAAC,UAAU,CAAC,KAAK,CAAC,EAAE,CAAC;QAC/E,MAAM,IAAI,oBAAoB,CAC5B,0CAA0C,EAC1C,OAAO,EACP,MAAM,CACP,CAAC;IACJ,CAAC;IAED,MAAM,SAAS,GAAc;QAC3B,IAAI,EAAE,SAAS;QACf,KAAK,EAAE,IAAI,CAAC,GAAG,CAAC,CAAC,EAAE,IAAI,CAAC,GAAG,CAAC,CAAC,EAAE,UAAU,CAAC,KAAK,CAAC,CAAC,EAAE,kBAAkB;QACrE,WAAW,EAAE,UAAU,CAAC,WAAW;QACnC,QAAQ,EAAE;YACR,UAAU,EAAE,MAAM,CAAC,IAAI;YACvB,IAAI,EAAE,MAAM,CAAC,IAAI;YACjB,SAAS,EAAE,MAAM,CAAC,SAAS;SAC5B;KACF,CAAC;IAEF,iBAAiB,CAAC,SAAS,CAAC,CAAC;IAC7B,OAAO,SAAS,CAAC;AACnB,CAAC;AAED;;;;;;;;;;;GAWG;AACH,MAAM,UAAU,mBAAmB,CACjC,UAAuB,EACvB,cAA8C,SAAS,EACvD,OAAkB,EAClB,OAAwD;IAExD,IAAI,UAAU,CAAC,MAAM,KAAK,CAAC,EAAE,CAAC;QAC5B,MAAM,QAAQ,GAAG,OAAO,EAAE,aAAa,IAAI,SAAS,CAAC;QACrD,IAAI,QAAQ,KAAK,MAAM;YAAE,OAAO,CAAC,CAAC;QAClC,IAAI,QAAQ,KAAK,KAAK;YAAE,OAAO,GAAG,CAAC;QACnC,OAAO,CAAC,CAAC,CAAC,sCAAsC;IAClD,CAAC;IAED,QAAQ,WAAW,EAAE,CAAC;QACpB,KAAK,SAAS;YACZ,OAAO,UAAU,CAAC,MAAM,CAAC,CAAC,GAAG,EAAE,CAAC,EAAE,EAAE,CAAC,GAAG,GAAG,CAAC,CAAC,KAAK,EAAE,CAAC,CAAC,GAAG,UAAU,CAAC,MAAM,CAAC;QAE7E,KAAK,UAAU,CAAC,CAAC,CAAC;YAChB,IAAI,CAAC,OAAO,IAAI,OAAO,CAAC,MAAM,KAAK,UAAU,CAAC,MAAM,EAAE,CAAC;gBACrD,MAAM,IAAI,KAAK,CAAC,2CAA2C,CAAC,CAAC;YAC/D,CAAC;YACD,0DAA0D;YAC1D,KAAK,IAAI,CAAC,GAAG,CAAC,EAAE,CAAC,GAAG,OAAO,CAAC,MAAM,EAAE,CAAC,EAAE,EAAE,CAAC;gBACxC,IAAI,OAAO,CAAC,CAAC,CAAC,GAAG,CAAC,IAAI,CAAC,MAAM,CAAC,QAAQ,CAAC,OAAO,CAAC,CAAC,CAAC,CAAC,EAAE,CAAC;oBACnD,MAAM,IAAI,KAAK,CAAC,2BAA2B,CAAC,KAAK,OAAO,CAAC,CAAC,CAAC,gDAAgD,CAAC,CAAC;gBAC/G,CAAC;YACH,CAAC;YACD,MAAM,WAAW,GAAG,OAAO,CAAC,MAAM,CAAC,CAAC,GAAG,EAAE,CAAC,EAAE,EAAE,CAAC,GAAG,GAAG,CAAC,EAAE,CAAC,CAAC,CAAC;YAC3D,IAAI,WAAW,KAAK,CAAC;gBAAE,OAAO,CAAC,CAAC;YAChC,OAAO,UAAU,CAAC,MAAM,CAAC,CAAC,GAAG,EAAE,CAAC,EAAE,CAAC,EAAE,EAAE,CAAC,GAAG,GAAG,CAAC,CAAC,KAAK,GAAG,OAAO,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,GAAG,WAAW,CAAC;QACvF,CAAC;QAED,KAAK,KAAK;YACR,OAAO,IAAI,CAAC,GAAG,CAAC,GAAG,UAAU,CAAC,GAAG,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC,KAAK,CAAC,CAAC,CAAC;IACrD,CAAC;AACH,CAAC;AAED,+EAA+E;AAC/E,gCAAgC;AAChC,+EAA+E;AAE/E;;;;;;;;;GASG;AACH,MAAM,UAAU,iBAAiB,CAC/B,QAAkB,EAClB,aAAsB;IAQtB,MAAM,OAAO,GAAG,QAAQ,CAAC,OAAO,IAAI,EAAE,CAAC;IACvC,MAAM,MAAM,GAAG,OAAO,CAAC,MAAM,CAAC;IAC9B,MAAM,SAAS,GAAG,OAAO,CAAC,MAAM,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC,IAAI,KAAK,WAAW,CAAC,CAAC;IAC9D,MAAM,aAAa,GAAG,SAAS,CAAC,MAAM,CAAC;IACvC,MAAM,WAAW,GAAG,IAAI,GAAG,CAAC,SAAS,CAAC,GAAG,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC,IAAI,CAAC,CAAC,MAAM,CAAC,OAAO,CAAC,CAAC,CAAC,IAAI,CAAC;IAE7E,uDAAuD;IACvD,MAAM,eAAe,GAAG,aAAa,IAAI,aAAa,GAAG,CAAC;QACxD,CAAC,CAAC,IAAI,CAAC,GAAG,CAAC,CAAC,EAAE,aAAa,GAAG,IAAI,CAAC,GAAG,CAAC,MAAM,EAAE,CAAC,CAAC,CAAC;QAClD,CAAC,CAAC,CAAC,CAAC;IAEN,sDAAsD;IACtD,MAAM,aAAa,GAAG,IAAI,GAAG,EAAU,CAAC;IACxC,IAAI,gBAAgB,GAAG,CAAC,CAAC;IACzB,KAAK,MAAM,MAAM,IAAI,SAAS,EAAE,CAAC;QAC/B,iFAAiF;QACjF,IAAI,CAAC;YACH,MAAM,GAAG,GAAG,IAAI,CAAC,SAAS,CAAC,EAAE,IAAI,EAAE,MAAM,CAAC,IAAI,EAAE,IAAI,EAAE,MAAM,CAAC,SAAS,EAAE,CAAC,CAAC;YAC1E,IAAI,aAAa,CAAC,GAAG,CAAC,GAAG,CAAC,EAAE,CAAC;gBAC3B,gBAAgB,EAAE,CAAC;YACrB,CAAC;YACD,aAAa,CAAC,GAAG,CAAC,GAAG,CAAC,CAAC;QACzB,CAAC;QAAC,MAAM,CAAC;YACP,mEAAmE;YACnE,SAAS;QACX,CAAC;IACH,CAAC;IAED,OAAO;QACL,MAAM;QACN,aAAa;QACb,WAAW;QACX,eAAe;QACf,gBAAgB;KACjB,CAAC;AACJ,CAAC"}

package/dist/lib/agent-judge/agent-judge-verification.test.d.ts

@@ -0,0 +1,5 @@
+/**
+ * Tests for Agent-as-Judge: verification, validation, scoring, and trajectory analysis.
+ */
+export {};
+//# sourceMappingURL=agent-judge-verification.test.d.ts.map

package/dist/lib/agent-judge/agent-judge-verification.test.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"agent-judge-verification.test.d.ts","sourceRoot":"","sources":["../../../src/lib/agent-judge/agent-judge-verification.test.ts"],"names":[],"mappings":"AAAA;;GAEG"}