| 11 | * Reporter class for formatting and displaying evaluation results |
| 12 | */ |
| 13 | export class EvaluationReporter { |
| 14 | /** |
| 15 | * Print comparison results |
| 16 | */ |
| 17 | printComparison(results: FocusedEvaluationResults[]): void { |
| 18 | console.log('\n' + '='.repeat(80)); |
| 19 | console.log('🏆 MODEL COMPARISON - TOOL USAGE EFFECTIVENESS'); |
| 20 | console.log('='.repeat(80)); |
| 21 | |
| 22 | // Sort by overall score |
| 23 | const sorted = [...results].sort((a, b) => b.overallScore - a.overallScore); |
| 24 | |
| 25 | console.log('\n📊 OVERALL RANKINGS:'); |
| 26 | sorted.forEach((result, index) => { |
| 27 | const rank = index + 1; |
| 28 | const emoji = rank === 1 ? '🥇' : rank === 2 ? '🥈' : rank === 3 ? '🥉' : ' '; |
| 29 | console.log( |
| 30 | `${emoji} ${rank}. ${result.model.padEnd(20)} ${(result.overallScore * 100).toFixed(1)}%` |
| 31 | ); |
| 32 | }); |
| 33 | |
| 34 | console.log('\n📋 DETAILED BREAKDOWN:'); |
| 35 | console.log( |
| 36 | 'Model'.padEnd(20) + |
| 37 | 'Overall'.padEnd(10) + |
| 38 | 'Tools'.padEnd(10) + |
| 39 | 'Order'.padEnd(10) + |
| 40 | 'Efficiency' |
| 41 | ); |
| 42 | console.log('-'.repeat(60)); |
| 43 | |
| 44 | sorted.forEach(result => { |
| 45 | const overall = (result.overallScore * 100).toFixed(1) + '%'; |
| 46 | const tools = (result.correctToolUsage * 100).toFixed(1) + '%'; |
| 47 | const order = (result.correctOrderUsage * 100).toFixed(1) + '%'; |
| 48 | const efficiency = (result.efficiencyScore * 100).toFixed(1) + '%'; |
| 49 | |
| 50 | console.log( |
| 51 | result.model.padEnd(20) + |
| 52 | overall.padEnd(10) + |
| 53 | tools.padEnd(10) + |
| 54 | order.padEnd(10) + |
| 55 | efficiency |
| 56 | ); |
| 57 | }); |
| 58 | |
| 59 | // Find best and worst scenarios |
| 60 | console.log('\n🎯 KEY INSIGHTS:'); |
| 61 | |
| 62 | const allResults = results.flatMap(r => r.results); |
| 63 | const bestScenarios = allResults.filter(r => r.score === 1.0); |
| 64 | const worstScenarios = allResults.filter(r => r.score < 0.5); |
| 65 | |
| 66 | if (bestScenarios.length > 0) { |
| 67 | const commonBest = this.findMostCommon(bestScenarios.map(r => r.scenarioId)); |
| 68 | console.log(`✅ Best performing scenario: ${commonBest} (consistent across models)`); |
| 69 | } |
| 70 |
nothing calls this directly
no outgoing calls
no test coverage detected