observability-toolkit 2.0.0 → 2.1.0

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

@@ -1,816 +0,0 @@
-/**
- * Tests for Agent-as-Judge Implementation
- */
-import { describe, it, beforeEach } from 'node:test';
-import assert from 'node:assert/strict';
-import { 
-// Constants
-MAX_TRAJECTORY_LENGTH, MAX_CONCURRENT_EVALUATORS, MAX_CONSENSUS_ROUNDS, DEFAULT_CONVERGENCE_THRESHOLD, 
-// Error classes
-AgentEvalTimeoutError, 
-// Utilities
-withAgentTimeout, validateEvaluand, validateStepScore, validateToolVerification, verifyToolCall, verifyToolCalls, scoreStep, aggregateStepScores, analyzeTrajectory, calculateVariance, calculateMedian, collectiveConsensus, 
-// Classes
-ProceduralJudge, ReactiveJudge, } from './agent-as-judge.js';
-import { InputValidationError } from './input-validator.js';
-describe('agent-as-judge', () => {
-    // ============================================================================
-    // Timeout Protection Tests
-    // ============================================================================
-    describe('withAgentTimeout', () => {
-        it('should return result when function completes in time', async () => {
-            const result = await withAgentTimeout(() => Promise.resolve(42), 1000);
-            assert.equal(result, 42);
-        });
-        it('should throw AgentEvalTimeoutError when function times out', async () => {
-            await assert.rejects(withAgentTimeout(() => new Promise((resolve) => setTimeout(resolve, 200)), 50), AgentEvalTimeoutError);
-        });
-        it('should include timeout duration in error', async () => {
-            try {
-                await withAgentTimeout(() => new Promise((resolve) => setTimeout(resolve, 200)), 50);
-                assert.fail('Should have thrown');
-            }
-            catch (error) {
-                assert.ok(error instanceof AgentEvalTimeoutError);
-                assert.equal(error.timeoutMs, 50);
-                assert.ok(error.message.includes('50'));
-            }
-        });
-        it('should propagate function errors', async () => {
-            await assert.rejects(withAgentTimeout(() => Promise.reject(new Error('Test error')), 1000), { message: 'Test error' });
-        });
-    });
-    // ============================================================================
-    // Validation Tests
-    // ============================================================================
-    describe('validateEvaluand', () => {
-        it('should pass for valid evaluand', () => {
-            assert.doesNotThrow(() => validateEvaluand({ input: 'test input', output: 'test output' }));
-        });
-        it('should throw for empty input', () => {
-            assert.throws(() => validateEvaluand({ input: '', output: 'test' }), InputValidationError);
-        });
-        it('should throw for whitespace-only input', () => {
-            assert.throws(() => validateEvaluand({ input: '   ', output: 'test' }), InputValidationError);
-        });
-        it('should throw for empty output', () => {
-            assert.throws(() => validateEvaluand({ input: 'test', output: '' }), InputValidationError);
-        });
-        it('should throw for actions exceeding MAX_TRAJECTORY_LENGTH', () => {
-            const actions = Array(MAX_TRAJECTORY_LENGTH + 1).fill({
-                type: 'tool_call',
-                tool: 'test',
-            });
-            assert.throws(() => validateEvaluand({ input: 'test', output: 'test', actions }), InputValidationError);
-        });
-        it('should pass for actions at MAX_TRAJECTORY_LENGTH', () => {
-            const actions = Array(MAX_TRAJECTORY_LENGTH).fill({
-                type: 'tool_call',
-                tool: 'test',
-            });
-            assert.doesNotThrow(() => validateEvaluand({ input: 'test', output: 'test', actions }));
-        });
-    });
-    describe('validateStepScore', () => {
-        it('should pass for valid step score with number step', () => {
-            assert.doesNotThrow(() => validateStepScore({ step: 0, score: 0.5 }));
-        });
-        it('should pass for valid step score with string step', () => {
-            assert.doesNotThrow(() => validateStepScore({ step: 'step_1', score: 0.5 }));
-        });
-        it('should throw for step string exceeding 256 characters', () => {
-            assert.throws(() => validateStepScore({ step: 'a'.repeat(257), score: 0.5 }), InputValidationError);
-        });
-        it('should throw for negative step index', () => {
-            assert.throws(() => validateStepScore({ step: -1, score: 0.5 }), InputValidationError);
-        });
-        it('should throw for non-integer step index', () => {
-            assert.throws(() => validateStepScore({ step: 1.5, score: 0.5 }), InputValidationError);
-        });
-        it('should throw for score below 0', () => {
-            assert.throws(() => validateStepScore({ step: 0, score: -0.1 }), InputValidationError);
-        });
-        it('should throw for score above 1', () => {
-            assert.throws(() => validateStepScore({ step: 0, score: 1.1 }), InputValidationError);
-        });
-        it('should throw for NaN score', () => {
-            assert.throws(() => validateStepScore({ step: 0, score: NaN }), InputValidationError);
-        });
-        it('should throw for Infinity score', () => {
-            assert.throws(() => validateStepScore({ step: 0, score: Infinity }), InputValidationError);
-        });
-    });
-    describe('validateToolVerification', () => {
-        it('should pass for valid verification', () => {
-            assert.doesNotThrow(() => validateToolVerification({
-                toolName: 'search',
-                toolCorrect: true,
-                argsCorrect: true,
-                score: 0.8,
-            }));
-        });
-        it('should throw for empty tool name', () => {
-            assert.throws(() => validateToolVerification({
-                toolName: '',
-                toolCorrect: true,
-                argsCorrect: true,
-                score: 0.8,
-            }), InputValidationError);
-        });
-        it('should throw for non-boolean toolCorrect', () => {
-            assert.throws(() => validateToolVerification({
-                toolName: 'test',
-                toolCorrect: 'yes',
-                argsCorrect: true,
-                score: 0.8,
-            }), InputValidationError);
-        });
-        it('should throw for invalid score', () => {
-            assert.throws(() => validateToolVerification({
-                toolName: 'test',
-                toolCorrect: true,
-                argsCorrect: true,
-                score: 1.5,
-            }), InputValidationError);
-        });
-    });
-    // ============================================================================
-    // Tool Verification Tests
-    // ============================================================================
-    describe('verifyToolCall', () => {
-        it('should throw for null action (H1)', () => {
-            assert.throws(() => verifyToolCall(null), InputValidationError);
-        });
-        it('should throw for non-object action (H1)', () => {
-            assert.throws(() => verifyToolCall('string'), InputValidationError);
-        });
-        it('should verify correct tool selection', () => {
-            const action = { type: 'tool_call', tool: 'search' };
-            const result = verifyToolCall(action, 'search');
-            assert.equal(result.toolCorrect, true);
-            assert.equal(result.score, 1.0);
-        });
-        it('should detect incorrect tool selection', () => {
-            const action = { type: 'tool_call', tool: 'read' };
-            const result = verifyToolCall(action, 'search');
-            assert.equal(result.toolCorrect, false);
-            assert.equal(result.score, 0);
-        });
-        it('should verify correct arguments', () => {
-            const action = {
-                type: 'tool_call',
-                tool: 'search',
-                arguments: { query: 'test' },
-            };
-            const result = verifyToolCall(action, 'search', { query: 'test' });
-            assert.equal(result.argsCorrect, true);
-            assert.ok(result.score > 0.9); // tool + args correct
-        });
-        it('should detect incorrect arguments', () => {
-            const action = {
-                type: 'tool_call',
-                tool: 'search',
-                arguments: { query: 'wrong' },
-            };
-            const result = verifyToolCall(action, 'search', { query: 'test' });
-            assert.equal(result.argsCorrect, false);
-        });
-        it('should verify result correctness when provided', () => {
-            const action = {
-                type: 'tool_call',
-                tool: 'calc',
-                result: 42,
-            };
-            const result = verifyToolCall(action, 'calc', undefined, undefined, 42);
-            assert.equal(result.resultCorrect, true);
-        });
-        it('should include evidence in result', () => {
-            const action = {
-                type: 'tool_call',
-                tool: 'test',
-                arguments: { foo: 'bar' },
-            };
-            const result = verifyToolCall(action, 'expected', { baz: 'qux' });
-            assert.ok(result.evidence);
-            assert.deepEqual(result.evidence.actualTool, 'test');
-            assert.deepEqual(result.evidence.expectedTool, 'expected');
-        });
-    });
-    describe('verifyToolCalls', () => {
-        it('should verify multiple tool calls', () => {
-            const actions = [
-                { type: 'tool_call', tool: 'search', toolCallId: 'call_1' },
-                { type: 'reasoning', reasoning: 'thinking...' },
-                { type: 'tool_call', tool: 'read', toolCallId: 'call_2' },
-            ];
-            const expected = new Map([
-                ['call_1', { tool: 'search' }],
-                ['call_2', { tool: 'read' }],
-            ]);
-            const results = verifyToolCalls(actions, expected);
-            assert.equal(results.length, 2); // Only tool_call actions
-            assert.equal(results[0].toolCorrect, true);
-            assert.equal(results[1].toolCorrect, true);
-        });
-        it('should skip non-tool actions', () => {
-            const actions = [
-                { type: 'reasoning', reasoning: 'thinking' },
-                { type: 'response', reasoning: 'responding' },
-            ];
-            const results = verifyToolCalls(actions);
-            assert.equal(results.length, 0);
-        });
-        it('should respect MAX_TOOL_VERIFICATIONS limit', () => {
-            const actions = Array(1000).fill({
-                type: 'tool_call',
-                tool: 'test',
-            });
-            const results = verifyToolCalls(actions);
-            assert.ok(results.length <= 500); // MAX_TOOL_VERIFICATIONS
-        });
-    });
-    // ============================================================================
-    // Step Scoring Tests
-    // ============================================================================
-    describe('scoreStep', () => {
-        it('should create valid step score', () => {
-            const action = { type: 'tool_call', tool: 'search' };
-            const result = scoreStep(action, 0, { score: 0.8, explanation: 'Good' });
-            assert.equal(result.step, 0);
-            assert.equal(result.score, 0.8);
-            assert.equal(result.explanation, 'Good');
-        });
-        it('should clamp score to [0, 1]', () => {
-            const action = { type: 'tool_call', tool: 'test' };
-            const high = scoreStep(action, 0, { score: 1.5 });
-            assert.equal(high.score, 1);
-            const low = scoreStep(action, 0, { score: -0.5 });
-            assert.equal(low.score, 0);
-        });
-        it('should throw for NaN score (H4)', () => {
-            const action = { type: 'tool_call', tool: 'test' };
-            assert.throws(() => scoreStep(action, 0, { score: NaN }), InputValidationError);
-        });
-        it('should throw for string score (H4)', () => {
-            const action = { type: 'tool_call', tool: 'test' };
-            assert.throws(() => scoreStep(action, 0, { score: '0.5' }), InputValidationError);
-        });
-        it('should include action metadata in evidence', () => {
-            const action = {
-                type: 'tool_call',
-                tool: 'search',
-                reasoning: 'searching for data',
-            };
-            const result = scoreStep(action, 0, { score: 0.9 });
-            const evidence = result.evidence;
-            assert.equal(evidence.actionType, 'tool_call');
-            assert.equal(evidence.tool, 'search');
-            assert.equal(evidence.reasoning, 'searching for data');
-        });
-    });
-    describe('aggregateStepScores', () => {
-        it('should return 1 for empty array', () => {
-            const result = aggregateStepScores([]);
-            assert.equal(result, 1);
-        });
-        it('should calculate average correctly', () => {
-            const scores = [
-                { step: 0, score: 0.8 },
-                { step: 1, score: 0.6 },
-                { step: 2, score: 1.0 },
-            ];
-            const result = aggregateStepScores(scores, 'average');
-            // Use approximate comparison for floating point
-            assert.ok(Math.abs(result - 0.8) < 0.0001);
-        });
-        it('should calculate weighted average correctly', () => {
-            const scores = [
-                { step: 0, score: 1.0 },
-                { step: 1, score: 0.0 },
-            ];
-            const result = aggregateStepScores(scores, 'weighted', [3, 1]);
-            assert.equal(result, 0.75);
-        });
-        it('should throw for weighted without weights', () => {
-            const scores = [{ step: 0, score: 0.5 }];
-            assert.throws(() => aggregateStepScores(scores, 'weighted'), Error);
-        });
-        it('should throw for mismatched weights length', () => {
-            const scores = [
-                { step: 0, score: 0.5 },
-                { step: 1, score: 0.5 },
-            ];
-            assert.throws(() => aggregateStepScores(scores, 'weighted', [1]), Error);
-        });
-        it('should return 0 for all-zero weights', () => {
-            const scores = [{ step: 0, score: 0.8 }];
-            const result = aggregateStepScores(scores, 'weighted', [0]);
-            assert.equal(result, 0);
-        });
-        it('should throw on negative weights (L8)', () => {
-            const scores = [
-                { step: 0, score: 0.5 },
-                { step: 1, score: 0.8 },
-            ];
-            assert.throws(() => aggregateStepScores(scores, 'weighted', [1, -1]), /Invalid weight at index 1: -1\. Weights must be finite non-negative numbers/);
-        });
-        it('should throw on negative weight at index 0', () => {
-            const scores = [
-                { step: 0, score: 0.5 },
-                { step: 1, score: 0.8 },
-            ];
-            assert.throws(() => aggregateStepScores(scores, 'weighted', [-1, 1]), /Invalid weight at index 0: -1\. Weights must be finite non-negative numbers/);
-        });
-        it('should throw on NaN weight (M1)', () => {
-            const scores = [
-                { step: 0, score: 0.5 },
-                { step: 1, score: 0.8 },
-            ];
-            assert.throws(() => aggregateStepScores(scores, 'weighted', [1, NaN]), /Invalid weight at index 1: NaN\. Weights must be finite non-negative numbers/);
-        });
-        it('should throw on Infinity weight', () => {
-            const scores = [
-                { step: 0, score: 0.5 },
-                { step: 1, score: 0.8 },
-            ];
-            assert.throws(() => aggregateStepScores(scores, 'weighted', [1, Infinity]), /Invalid weight at index 1: Infinity\. Weights must be finite non-negative numbers/);
-        });
-        it('should throw on -Infinity weight', () => {
-            const scores = [
-                { step: 0, score: 0.5 },
-                { step: 1, score: 0.8 },
-            ];
-            assert.throws(() => aggregateStepScores(scores, 'weighted', [-Infinity, 1]), /Invalid weight at index 0: -Infinity\. Weights must be finite non-negative numbers/);
-        });
-        it('should calculate min correctly', () => {
-            const scores = [
-                { step: 0, score: 0.9 },
-                { step: 1, score: 0.3 },
-                { step: 2, score: 0.7 },
-            ];
-            const result = aggregateStepScores(scores, 'min');
-            assert.equal(result, 0.3);
-        });
-    });
-    // ============================================================================
-    // Trajectory Analysis Tests
-    // ============================================================================
-    describe('analyzeTrajectory', () => {
-        it('should calculate basic metrics', () => {
-            const evaluand = {
-                input: 'test',
-                output: 'result',
-                actions: [
-                    { type: 'tool_call', tool: 'search' },
-                    { type: 'reasoning', reasoning: 'thinking' },
-                    { type: 'tool_call', tool: 'read' },
-                ],
-            };
-            const result = analyzeTrajectory(evaluand);
-            assert.equal(result.length, 3);
-            assert.equal(result.toolCallCount, 2);
-            assert.equal(result.uniqueTools, 2);
-        });
-        it('should handle empty actions', () => {
-            const evaluand = { input: 'test', output: 'result' };
-            const result = analyzeTrajectory(evaluand);
-            assert.equal(result.length, 0);
-            assert.equal(result.toolCallCount, 0);
-        });
-        it('should calculate efficiency ratio', () => {
-            const evaluand = {
-                input: 'test',
-                output: 'result',
-                actions: Array(10).fill({ type: 'tool_call', tool: 'test' }),
-            };
-            const result = analyzeTrajectory(evaluand, 5);
-            assert.equal(result.efficiencyRatio, 0.5); // 5/10
-        });
-        it('should cap efficiency ratio at 1', () => {
-            const evaluand = {
-                input: 'test',
-                output: 'result',
-                actions: [{ type: 'tool_call', tool: 'test' }],
-            };
-            const result = analyzeTrajectory(evaluand, 10);
-            assert.equal(result.efficiencyRatio, 1);
-        });
-        it('should detect redundant actions', () => {
-            const evaluand = {
-                input: 'test',
-                output: 'result',
-                actions: [
-                    { type: 'tool_call', tool: 'search', arguments: { q: 'a' } },
-                    { type: 'tool_call', tool: 'search', arguments: { q: 'a' } }, // Duplicate
-                    { type: 'tool_call', tool: 'search', arguments: { q: 'b' } }, // Different args
-                ],
-            };
-            const result = analyzeTrajectory(evaluand);
-            assert.equal(result.redundantActions, 1);
-        });
-    });
-    // ============================================================================
-    // Statistical Functions Tests
-    // ============================================================================
-    describe('calculateVariance', () => {
-        it('should return 0 for empty array', () => {
-            assert.equal(calculateVariance([]), 0);
-        });
-        it('should return 0 for single value', () => {
-            assert.equal(calculateVariance([5]), 0);
-        });
-        it('should return 0 for identical values', () => {
-            assert.equal(calculateVariance([3, 3, 3, 3]), 0);
-        });
-        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([2, 4, 4, 4, 5, 5, 7, 9]);
-            assert.ok(Math.abs(result - 32 / 7) < 0.001, `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([2, 6]);
-            assert.equal(result, 8);
-        });
-    });
-    describe('calculateMedian', () => {
-        it('should return 0 for empty array', () => {
-            assert.equal(calculateMedian([]), 0);
-        });
-        it('should return value for single element', () => {
-            assert.equal(calculateMedian([5]), 5);
-        });
-        it('should calculate median for odd count', () => {
-            assert.equal(calculateMedian([1, 3, 5]), 3);
-        });
-        it('should calculate median for even count', () => {
-            assert.equal(calculateMedian([1, 2, 3, 4]), 2.5);
-        });
-        it('should handle unsorted input', () => {
-            assert.equal(calculateMedian([5, 1, 3]), 3);
-        });
-    });
-    // ============================================================================
-    // Consensus Tests
-    // ============================================================================
-    describe('collectiveConsensus', () => {
-        const validEvaluand = { input: 'test', output: 'result' };
-        const config = {
-            rounds: 3,
-            convergenceThreshold: DEFAULT_CONVERGENCE_THRESHOLD,
-        };
-        it('should throw for invalid evaluand', async () => {
-            await assert.rejects(collectiveConsensus({ input: '', output: 'test' }, [], config), InputValidationError);
-        });
-        it('should throw for too many judges (H2)', async () => {
-            const judges = Array(MAX_CONCURRENT_EVALUATORS + 1).fill({
-                id: 'judge',
-                evaluate: async () => 0.5,
-            });
-            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 () => 0.5 }];
-            await assert.rejects(collectiveConsensus(validEvaluand, judges, { ...config, rounds: 0 }), InputValidationError);
-        });
-        it('should reach consensus with agreeing judges', async () => {
-            const judges = [
-                { id: 'j1', evaluate: async () => 0.8 },
-                { id: 'j2', evaluate: async () => 0.82 },
-                { id: 'j3', evaluate: async () => 0.79 },
-            ];
-            const result = await collectiveConsensus(validEvaluand, judges, config);
-            assert.ok(result.converged);
-            assert.ok(result.finalScore > 0.7);
-        });
-        it('should handle disagreeing judges', async () => {
-            const judges = [
-                { id: 'j1', evaluate: async () => 0.2 },
-                { id: 'j2', evaluate: async () => 0.8 },
-            ];
-            const result = await collectiveConsensus(validEvaluand, judges, {
-                rounds: 2,
-                convergenceThreshold: 0.01, // Very tight
-            });
-            assert.equal(result.converged, false);
-        });
-        it('should handle judge failures gracefully (H3)', async () => {
-            const judges = [
-                { id: 'j1', evaluate: async () => 0.8 },
-                {
-                    id: 'j2',
-                    evaluate: async () => {
-                        throw new Error('Judge failed');
-                    },
-                },
-                { id: 'j3', evaluate: async () => 0.75 },
-            ];
-            const result = await collectiveConsensus(validEvaluand, judges, config);
-            // Should succeed with 2 working judges
-            assert.ok(result.finalScore > 0);
-        });
-        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 respect MAX_CONSENSUS_ROUNDS', async () => {
-            let callCount = 0;
-            const judges = [
-                {
-                    id: 'j1',
-                    evaluate: async () => {
-                        callCount++;
-                        return 0.5;
-                    },
-                },
-            ];
-            await collectiveConsensus(validEvaluand, judges, {
-                rounds: 100, // Exceeds max
-                convergenceThreshold: 0.0001,
-            });
-            assert.ok(callCount <= MAX_CONSENSUS_ROUNDS);
-        });
-    });
-    // ============================================================================
-    // ProceduralJudge Tests
-    // ============================================================================
-    describe('ProceduralJudge', () => {
-        it('should execute all stages in order', async () => {
-            const executionOrder = [];
-            const stages = [
-                {
-                    name: 'stage1',
-                    evaluate: async () => {
-                        executionOrder.push('stage1');
-                        return { score: 0.8, explanation: 'Good' };
-                    },
-                },
-                {
-                    name: 'stage2',
-                    evaluate: async () => {
-                        executionOrder.push('stage2');
-                        return { score: 0.9, explanation: 'Great' };
-                    },
-                },
-            ];
-            const judge = new ProceduralJudge(stages);
-            const result = await judge.evaluate({
-                input: 'test',
-                output: 'result',
-            });
-            assert.deepEqual(executionOrder, ['stage1', 'stage2']);
-            assert.equal(result.stepScores.length, 2);
-        });
-        it('should support early termination', async () => {
-            const executionOrder = [];
-            const stages = [
-                {
-                    name: 'safety',
-                    evaluate: async () => {
-                        executionOrder.push('safety');
-                        return { score: 0.2, explanation: 'Failed safety' };
-                    },
-                },
-                {
-                    name: 'quality',
-                    evaluate: async () => {
-                        executionOrder.push('quality');
-                        return { score: 0.9, explanation: 'Good' };
-                    },
-                },
-            ];
-            const judge = new ProceduralJudge(stages, 'safety');
-            const result = await judge.evaluate({
-                input: 'test',
-                output: 'result',
-            });
-            assert.deepEqual(executionOrder, ['safety']);
-            assert.equal(result.overallScore, 0);
-            assert.ok(result.explanation.includes('Early termination'));
-        });
-        it('should pass context between stages', async () => {
-            let stage2Context = {};
-            const stages = [
-                {
-                    name: 'stage1',
-                    evaluate: async (_, ctx) => {
-                        ctx['data'] = 'from_stage1';
-                        return { score: 0.8, explanation: 'OK' };
-                    },
-                },
-                {
-                    name: 'stage2',
-                    evaluate: async (_, ctx) => {
-                        stage2Context = { ...ctx };
-                        return { score: 0.9, explanation: 'OK' };
-                    },
-                },
-            ];
-            const judge = new ProceduralJudge(stages);
-            await judge.evaluate({ input: 'test', output: 'result' });
-            assert.ok(stage2Context['stage1']);
-        });
-        it('should validate evaluand', async () => {
-            const judge = new ProceduralJudge([
-                { name: 'test', evaluate: async () => ({ score: 1, explanation: '' }) },
-            ]);
-            await assert.rejects(judge.evaluate({ input: '', output: 'test' }), InputValidationError);
-        });
-        describe('constructor validation (M8)', () => {
-            it('should throw on empty stages array', () => {
-                assert.throws(() => new ProceduralJudge([]), InputValidationError);
-            });
-            it('should throw on stage with empty name', () => {
-                assert.throws(() => new ProceduralJudge([
-                    { name: '', evaluate: async () => ({ score: 1, explanation: '' }) },
-                ]), InputValidationError);
-            });
-            it('should throw on stage with whitespace-only name', () => {
-                assert.throws(() => new ProceduralJudge([
-                    { name: '   ', evaluate: async () => ({ score: 1, explanation: '' }) },
-                ]), InputValidationError);
-            });
-            it('should throw on stage without evaluate function', () => {
-                assert.throws(
-                // @ts-expect-error - testing runtime validation
-                () => new ProceduralJudge([{ name: 'test' }]), InputValidationError);
-            });
-            it('should throw on invalid earlyTerminationOn stage name', () => {
-                assert.throws(() => new ProceduralJudge([{ name: 'stage1', evaluate: async () => ({ score: 1, explanation: '' }) }], 'nonexistent'), InputValidationError);
-            });
-            it('should accept valid stages with earlyTerminationOn', () => {
-                const judge = new ProceduralJudge([{ name: 'safety', evaluate: async () => ({ score: 1, explanation: '' }) }], 'safety');
-                assert.ok(judge);
-            });
-        });
-    });
-    // ============================================================================
-    // ReactiveJudge Tests
-    // ============================================================================
-    describe('ReactiveJudge', () => {
-        it('should route to selected specialists', async () => {
-            const executedSpecialists = [];
-            const router = async () => ['quality', 'safety'];
-            const specialists = new Map([
-                [
-                    'quality',
-                    async () => {
-                        executedSpecialists.push('quality');
-                        return { score: 0.9, explanation: 'Good quality' };
-                    },
-                ],
-                [
-                    'safety',
-                    async () => {
-                        executedSpecialists.push('safety');
-                        return { score: 1.0, explanation: 'Safe' };
-                    },
-                ],
-                [
-                    'style',
-                    async () => {
-                        executedSpecialists.push('style');
-                        return { score: 0.8, explanation: 'OK style' };
-                    },
-                ],
-            ]);
-            const judge = new ReactiveJudge(router, specialists);
-            const result = await judge.evaluate({
-                input: 'test',
-                output: 'result',
-            });
-            assert.deepEqual(executedSpecialists, ['quality', 'safety']);
-            assert.equal(result.stepScores.length, 2);
-        });
-        it('should trigger deep dive when needed', async () => {
-            const deepDiveTriggered = { value: false };
-            const router = async () => ['quality'];
-            const specialists = new Map([
-                [
-                    'quality',
-                    async () => ({
-                        score: 0.5,
-                        explanation: 'Needs investigation',
-                        needsDeepDive: true,
-                    }),
-                ],
-            ]);
-            const deepDiveSpecialists = new Map([
-                [
-                    'quality',
-                    async () => {
-                        deepDiveTriggered.value = true;
-                        return { score: 0.4, explanation: 'Deep analysis' };
-                    },
-                ],
-            ]);
-            const judge = new ReactiveJudge(router, specialists, deepDiveSpecialists);
-            const result = await judge.evaluate({
-                input: 'test',
-                output: 'result',
-            });
-            assert.equal(deepDiveTriggered.value, true);
-            assert.equal(result.stepScores.length, 2); // Regular + deep dive
-        });
-        it('should skip missing specialists', async () => {
-            const router = async () => ['missing', 'existing'];
-            const specialists = new Map([
-                ['existing', async () => ({ score: 0.8, explanation: 'OK' })],
-            ]);
-            const judge = new ReactiveJudge(router, specialists);
-            const result = await judge.evaluate({
-                input: 'test',
-                output: 'result',
-            });
-            assert.equal(result.stepScores.length, 1);
-        });
-        it('should validate evaluand', async () => {
-            const judge = new ReactiveJudge(async () => [], new Map());
-            await assert.rejects(judge.evaluate({ input: '', output: 'test' }), InputValidationError);
-        });
-    });
-    // ============================================================================
-    // AgentJudge Memory Tests (H5)
-    // ============================================================================
-    describe('AgentJudge memory management', () => {
-        class TestJudge extends ProceduralJudge {
-            // Expose protected methods for testing
-            testStore(key, value) {
-                this.storeInMemory(key, value);
-            }
-            testGet(key) {
-                return this.getFromMemory(key);
-            }
-            getMemorySize() {
-                return this.memory.size;
-            }
-        }
-        let judge;
-        beforeEach(() => {
-            judge = new TestJudge([
-                { name: 'test', evaluate: async () => ({ score: 1, explanation: '' }) },
-            ]);
-        });
-        it('should store and retrieve values', () => {
-            judge.testStore('key1', 'value1');
-            assert.equal(judge.testGet('key1'), 'value1');
-        });
-        it('should implement LRU - reading moves to end (H5)', () => {
-            // Fill nearly to capacity
-            for (let i = 0; i < 997; i++) {
-                judge.testStore(`fill_${i}`, i);
-            }
-            // Store 3 more items we care about
-            judge.testStore('key1', 'value1');
-            judge.testStore('key2', 'value2');
-            judge.testStore('key3', 'value3');
-            // Memory is now at 1000 - order is fill_0...fill_996, key1, key2, key3
-            // Read key1 - should move it to end (after key3)
-            judge.testGet('key1');
-            // Order is now: fill_0...fill_996, key2, key3, key1
-            // Store 1 more to trigger eviction of fill_0
-            judge.testStore('new_item', 'new');
-            // Order: fill_1...fill_996, key2, key3, key1, new_item
-            // key1 should still exist (was accessed recently), key2/key3 should exist
-            assert.equal(judge.testGet('key1'), 'value1');
-            assert.equal(judge.testGet('key2'), 'value2');
-            assert.equal(judge.testGet('key3'), 'value3');
-            // fill_0 should be evicted
-            assert.equal(judge.testGet('fill_0'), undefined);
-        });
-        it('should evict oldest on overflow', () => {
-            // Fill to capacity
-            for (let i = 0; i < 1000; i++) {
-                judge.testStore(`key_${i}`, i);
-            }
-            assert.equal(judge.getMemorySize(), 1000);
-            // Add one more
-            judge.testStore('new_key', 'new_value');
-            assert.equal(judge.getMemorySize(), 1000);
-            assert.equal(judge.testGet('key_0'), undefined); // First one evicted
-            assert.equal(judge.testGet('new_key'), 'new_value');
-        });
-    });
-});
-//# sourceMappingURL=agent-as-judge.test.js.map