test: refactor accuracy benchmark generation

benchmarks/results/accuracy/report.md

@@ -28,7 +28,7 @@ claude-haiku-4-5
 ##### Uniform employee records (TOON optimal format)
 
 | Format | Accuracy | Tokens | Correct/Total |
-|--------|----------|--------|---------------|
+| ------ | -------- | ------ | ------------- |
 | `toon` | 86.2% | 2.483 | 100/116 |
 | `csv` | 80.2% | 2.337 | 93/116 |
 | `yaml` | 82.8% | 4.969 | 96/116 |
@@ -38,7 +38,7 @@ claude-haiku-4-5
 ##### E-commerce orders with nested structures
 
 | Format | Accuracy | Tokens | Correct/Total |
-|--------|----------|--------|---------------|
+| ------ | -------- | ------ | ------------- |
 | `toon` | 90.9% | 5.967 | 80/88 |
 | `csv` | 90.9% | 6.735 | 80/88 |
 | `yaml` | 89.8% | 7.328 | 79/88 |
@@ -48,17 +48,17 @@ claude-haiku-4-5
 ##### Time-series analytics data
 
 | Format | Accuracy | Tokens | Correct/Total |
-|--------|----------|--------|---------------|
+| ------ | -------- | ------ | ------------- |
 | `csv` | 87.9% | 1.393 | 51/58 |
 | `toon` | 86.2% | 1.515 | 50/58 |
 | `yaml` | 86.2% | 2.938 | 50/58 |
 | `json` | 87.9% | 3.665 | 51/58 |
 | `markdown-kv` | 86.2% | 3.779 | 50/58 |
 
-##### Popular GitHub repositories
+##### Top 100 GitHub repositories
 
 | Format | Accuracy | Tokens | Correct/Total |
-|--------|----------|--------|---------------|
+| ------ | -------- | ------ | ------------- |
 | `csv` | 80.4% | 8.513 | 45/56 |
 | `toon` | 80.4% | 8.745 | 45/56 |
 | `yaml` | 78.6% | 13.129 | 44/56 |
@@ -70,7 +70,7 @@ claude-haiku-4-5
 ##### gpt-5-nano
 
 | Format | Accuracy | Correct/Total |
-|--------|----------|---------------|
+| ------ | -------- | ------------- |
 | `toon` | 97.5% | 155/159 |
 | `markdown-kv` | 95.6% | 152/159 |
 | `yaml` | 94.3% | 150/159 |
@@ -80,7 +80,7 @@ claude-haiku-4-5
 ##### claude-haiku-4-5
 
 | Format | Accuracy | Correct/Total |
-|--------|----------|---------------|
+| ------ | -------- | ------------- |
 | `markdown-kv` | 76.7% | 122/159 |
 | `toon` | 75.5% | 120/159 |
 | `json` | 75.5% | 120/159 |

benchmarks/results/accuracy/summary.json

@@ -61,7 +61,7 @@
     },
     {
       "name": "github",
-      "description": "Popular GitHub repositories"
+      "description": "Top 100 GitHub repositories"
     }
   ],
   "tokenCounts": {
@@ -86,5 +86,5 @@
     "yaml-analytics": 2938,
     "yaml-github": 13129
   },
-  "timestamp": "2025-10-27T12:43:38.288Z"
+  "timestamp": "2025-10-27T13:04:50.634Z"
 }

benchmarks/scripts/accuracy-benchmark.ts

@@ -81,6 +81,7 @@ else {
 
   // Format datasets once (reuse for all questions)
   const formattedDatasets: Record<string, Record<string, string>> = {}
+
   for (const [formatName, formatter] of Object.entries(formatters)) {
     formattedDatasets[formatName] ??= {}
 
@@ -91,6 +92,7 @@ else {
 
   // Generate evaluation tasks
   const tasks: { question: Question, formatName: string, modelName: string }[] = []
+
   for (const question of questions) {
     for (const [formatName] of Object.entries(formatters)) {
       for (const [modelName] of Object.entries(activeModels)) {
@@ -100,7 +102,6 @@ else {
   }
 
   const total = tasks.length
-
   consola.start(`Running ${total} evaluations with concurrency: ${DEFAULT_CONCURRENCY}`)
 
   // Evaluate all tasks in parallel
@@ -110,16 +111,15 @@ else {
       const formattedData = formattedDatasets[task.formatName]![task.question.dataset]!
       const model = activeModels[task.modelName as keyof typeof activeModels]!
 
-      const result = await evaluateQuestion(
-        task.question,
-        task.formatName,
+      const result = await evaluateQuestion({
+        question: task.question,
+        formatName: task.formatName,
         formattedData,
         model,
-        task.modelName,
-      )
+      })
 
-      // Progress update
-      if ((index + 1) % 10 === 0) {
+      // Progress update after task completes
+      if ((index + 1) % 10 === 0 || (index + 1) === total) {
         const percent = (((index + 1) / total) * 100).toFixed(1)
         consola.start(`Progress: ${index + 1}/${total} (${percent}%)`)
       }
@@ -133,6 +133,7 @@ else {
 }
 
 // Generate/regenerate markdown report
+consola.start('Generating report and saving results…')
 const formatResults = calculateFormatResults(results, tokenCounts)
 await saveResults(results, formatResults, questions, tokenCounts)
 

benchmarks/scripts/token-efficiency-benchmark.ts

@@ -46,7 +46,7 @@ const BENCHMARK_EXAMPLES = [
   {
     name: 'E-commerce Order',
     emoji: '🛒',
-    description: 'Nested order with customer and items',
+    description: 'Single nested order with customer and items',
     getData: generateOrder,
     showDetailed: false,
   },

benchmarks/src/constants.ts

@@ -5,8 +5,9 @@ export const ROOT_DIR: string = url.fileURLToPath(new URL('../../', import.meta.
 export const BENCHMARKS_DIR: string = url.fileURLToPath(new URL('../', import.meta.url))
 
 /**
- * Benchmark execution configuration
+ * Default concurrency for parallel evaluations
  */
+export const DEFAULT_CONCURRENCY = 20
 
 /**
  * Enable dry run mode for quick testing with limited AI requests
@@ -27,13 +28,3 @@ export const DRY_RUN_LIMITS = {
   /** Models to use in dry run */
   allowedModels: [] as string[],
 }
-
-/**
- * Default concurrency for parallel evaluations
- */
-export const DEFAULT_CONCURRENCY = 20
-
-/**
- * Delay between API requests to avoid rate limiting (in milliseconds)
- */
-export const RATE_LIMIT_DELAY_MS = 100

benchmarks/src/datasets.ts

@@ -122,16 +122,16 @@ const analyticsDataset: Dataset = {
 }
 
 /**
- * GitHub dataset: Popular repositories
+ * Real-world dataset: Top 100 starred GitHub repositories
  *
  * @remarks
- * Tests TOON's tabular format with real-world data
+ * Tests TOON's tabular format
  */
 const githubDataset: Dataset = {
   name: 'github',
-  description: 'Popular GitHub repositories',
+  description: 'Top 100 GitHub repositories',
   data: {
-    repositories: githubRepos.slice(0, 200),
+    repositories: githubRepos,
   },
 }
 

benchmarks/src/evaluate.ts

@@ -9,12 +9,10 @@
 
 import type { LanguageModelV2 } from '@ai-sdk/provider'
 import type { EvaluationResult, Question } from './types'
-import { setTimeout } from 'node:timers/promises'
 import { anthropic } from '@ai-sdk/anthropic'
 import { openai } from '@ai-sdk/openai'
 import { generateText } from 'ai'
 import { consola } from 'consola'
-import { RATE_LIMIT_DELAY_MS } from './constants'
 
 /**
  * Models used for evaluation
@@ -28,11 +26,8 @@ export const models: Record<string, LanguageModelV2> = {
  * Evaluate a single question with a specific format and model
  */
 export async function evaluateQuestion(
-  question: Question,
-  formatName: string,
-  formattedData: string,
-  model: LanguageModelV2,
-  modelName: string,
+  { question, formatName, formattedData, model}:
+  { question: Question, formatName: string, formattedData: string, model: LanguageModelV2 },
 ): Promise<EvaluationResult> {
   const prompt = `Given the following data in ${formatName} format:
 
@@ -51,10 +46,8 @@ Provide only the direct answer, without any additional explanation or formatting
     temperature: model.modelId.startsWith('gpt-') ? undefined : 0,
   })
 
-  await setTimeout(RATE_LIMIT_DELAY_MS)
-
   const latencyMs = performance.now() - startTime
-  const correct = await validateAnswer({
+  const isCorrect = await validateAnswer({
     actual: text.trim(),
     expected: question.groundTruth,
     question: question.prompt,
@@ -63,10 +56,10 @@ Provide only the direct answer, without any additional explanation or formatting
   return {
     questionId: question.id,
     format: formatName,
-    model: modelName,
+    model: model.modelId,
     expected: question.groundTruth,
     actual: text.trim(),
-    correct,
+    isCorrect,
     inputTokens: usage.inputTokens,
     outputTokens: usage.outputTokens,
     latencyMs,
@@ -105,8 +98,6 @@ Respond with only "YES" or "NO".`
       temperature: 0,
     })
 
-    await setTimeout(RATE_LIMIT_DELAY_MS)
-
     return text.trim().toUpperCase() === 'YES'
   }
   catch (error) {

benchmarks/src/report.ts

@@ -3,7 +3,7 @@
  *
  * Handles:
  * - Statistical analysis
- * - Twitter-ready markdown report generation with visual elements
+ * - Markdown report generation with visual elements
  * - Per-dataset breakdowns
  * - Cost analysis
  * - Result file saving
@@ -28,7 +28,7 @@ export function calculateFormatResults(
 
   return formatNames.map((formatName) => {
     const formatResults = results.filter(r => r.format === formatName)
-    const correctCount = formatResults.filter(r => r.correct).length
+    const correctCount = formatResults.filter(r => r.isCorrect).length
     const totalCount = formatResults.length
     const accuracy = correctCount / totalCount
 
@@ -59,24 +59,17 @@ export function generateMarkdownReport(
   questions: Question[],
   tokenCounts: Record<string, number>,
 ): string {
-  const lines: string[] = [
-    '### Retrieval Accuracy',
-    '',
-  ]
-
   const toon = formatResults.find(r => r.format === 'toon')
   const json = formatResults.find(r => r.format === 'json')
 
-  // Model-by-model breakdown with ASCII bars
+  // Build model-by-model breakdown with ASCII bars
   const modelCount = Object.keys(models).length
-  lines.push(`Tested across **${modelCount} ${modelCount === 1 ? 'LLM' : 'LLMs'}** with data retrieval tasks:`, '', '```')
-
   const modelNames = Object.keys(models)
-  for (let i = 0; i < modelNames.length; i++) {
-    const modelName = modelNames[i]!
+
+  const modelBreakdown = modelNames.map((modelName, i) => {
     const modelResults = formatResults.map((fr) => {
       const modelFormatResults = results.filter(r => r.model === modelName && r.format === fr.format)
-      const correctCount = modelFormatResults.filter(r => r.correct).length
+      const correctCount = modelFormatResults.filter(r => r.isCorrect).length
       const totalCount = modelFormatResults.length
       const accuracy = totalCount > 0 ? correctCount / totalCount : 0
 
@@ -88,34 +81,24 @@ export function generateMarkdownReport(
       }
     }).sort((a, b) => b.accuracy - a.accuracy)
 
-    // Add blank line before model name, except for first model
-    if (i > 0)
-      lines.push('')
-    lines.push(modelName)
-    for (const result of modelResults) {
+    const formatLines = modelResults.map((result) => {
       const bar = createProgressBar(result.accuracy, 1, 20)
       const accuracyStr = `${(result.accuracy * 100).toFixed(1)}%`.padStart(6)
       const countStr = `(${result.correctCount}/${result.totalCount})`
-      lines.push(`  ${result.format.padEnd(12)} ${bar} ${accuracyStr} ${countStr}`)
-    }
-  }
+      return `  ${result.format.padEnd(12)} ${bar} ${accuracyStr} ${countStr}`
+    }).join('\n')
 
-  lines.push('```', '')
+    // Add blank line before model name, except for first model
+    return `${i > 0 ? '\n' : ''}${modelName}\n${formatLines}`
+  }).join('\n')
 
-  // Summary comparison
-  if (toon && json) {
-    const tokenSavings = ((1 - toon.totalTokens / json.totalTokens) * 100).toFixed(1)
-    lines.push(
-      `**Tradeoff:** TOON achieves ${(toon.accuracy * 100).toFixed(1)}% accuracy (vs JSON's ${(json.accuracy * 100).toFixed(1)}%) while using ${tokenSavings}% fewer tokens.`,
-      '',
-    )
-  }
-
-  lines.push('<details>', '<summary><strong>View detailed breakdown by dataset and model</strong></summary>', '', '#### Performance by Dataset', '')
-
-  for (const dataset of datasets) {
-    lines.push(`##### ${dataset.description}`, '')
+  // Build summary comparison
+  const summaryComparison = toon && json
+    ? `**Tradeoff:** TOON achieves ${(toon.accuracy * 100).toFixed(1)}% accuracy (vs JSON's ${(json.accuracy * 100).toFixed(1)}%) while using ${((1 - toon.totalTokens / json.totalTokens) * 100).toFixed(1)}% fewer tokens.`
+    : ''
 
+  // Build performance by dataset
+  const datasetBreakdown = datasets.map((dataset) => {
     const datasetResults = formatResults.map((fr) => {
       const datasetFormatResults = results.filter(r => r.questionId.includes(dataset.name) || questions.find(q => q.id === r.questionId)?.dataset === dataset.name)
       if (datasetFormatResults.length === 0)
@@ -125,7 +108,7 @@ export function generateMarkdownReport(
       if (formatDatasetResults.length === 0)
         return undefined
 
-      const correctCount = formatDatasetResults.filter(r => r.correct).length
+      const correctCount = formatDatasetResults.filter(r => r.isCorrect).length
       const totalCount = formatDatasetResults.length
       const accuracy = totalCount > 0 ? correctCount / totalCount : 0
 
@@ -143,7 +126,7 @@ export function generateMarkdownReport(
     }).filter(Boolean) as { format: string, accuracy: number, tokens: number, correctCount: number, totalCount: number }[]
 
     if (datasetResults.length === 0)
-      continue
+      return ''
 
     // Sort by efficiency
     datasetResults.sort((a, b) => {
@@ -152,29 +135,24 @@ export function generateMarkdownReport(
       return effB - effA
     })
 
-    lines.push(
-      '| Format | Accuracy | Tokens | Correct/Total |',
-      '|--------|----------|--------|---------------|',
-    )
+    const tableRows = datasetResults.slice(0, 6).map(result =>
+      `| \`${result.format}\` | ${(result.accuracy * 100).toFixed(1)}% | ${result.tokens.toLocaleString()} | ${result.correctCount}/${result.totalCount} |`,
+    ).join('\n')
 
-    for (const result of datasetResults.slice(0, 6)) {
-      lines.push(
-        `| \`${result.format}\` | ${(result.accuracy * 100).toFixed(1)}% | ${result.tokens.toLocaleString()} | ${result.correctCount}/${result.totalCount} |`,
-      )
-    }
+    return `
+##### ${dataset.description}
 
-    lines.push('')
-  }
-
-  // Model breakdown
-  lines.push('#### Performance by Model', '')
-
-  for (const modelName of Object.keys(models)) {
-    lines.push(`##### ${modelName}`, '')
+| Format | Accuracy | Tokens | Correct/Total |
+| ------ | -------- | ------ | ------------- |
+${tableRows}
+`.trimStart()
+  }).filter(Boolean).join('\n')
 
+  // Build performance by model
+  const modelPerformance = modelNames.map((modelName) => {
     const modelResults = formatResults.map((fr) => {
       const modelFormatResults = results.filter(r => r.model === modelName && r.format === fr.format)
-      const correctCount = modelFormatResults.filter(r => r.correct).length
+      const correctCount = modelFormatResults.filter(r => r.isCorrect).length
       const totalCount = modelFormatResults.length
       const accuracy = correctCount / totalCount
 
@@ -186,36 +164,55 @@ export function generateMarkdownReport(
       }
     }).sort((a, b) => b.accuracy - a.accuracy)
 
-    lines.push('| Format | Accuracy | Correct/Total |', '|--------|----------|---------------|')
+    const tableRows = modelResults.map(result =>
+      `| \`${result.format}\` | ${(result.accuracy * 100).toFixed(1)}% | ${result.correctCount}/${result.totalCount} |`,
+    ).join('\n')
 
-    for (const result of modelResults) {
-      lines.push(`| \`${result.format}\` | ${(result.accuracy * 100).toFixed(1)}% | ${result.correctCount}/${result.totalCount} |`)
-    }
+    return `
+##### ${modelName}
 
-    lines.push('')
-  }
+| Format | Accuracy | Correct/Total |
+| ------ | -------- | ------------- |
+${tableRows}
+`.trimStart()
+  }).join('\n')
 
-  // Methodology
-  lines.push(
-    '#### Methodology',
-    '',
-    '- **Semantic validation**: LLM-as-judge validates responses semantically (not exact string matching).',
-    '- **Token counting**: Using `gpt-tokenizer` with `o200k_base` encoding.',
-    '- **Question types**: Field retrieval, aggregation, and filtering tasks.',
-    '- **Real data**: Faker.js-generated datasets + GitHub repositories.',
-    '',
-    '</details>',
-    '',
-  )
+  return `
+### Retrieval Accuracy
 
-  return lines.join('\n')
+Tested across **${modelCount} ${modelCount === 1 ? 'LLM' : 'LLMs'}** with data retrieval tasks:
+
+\`\`\`
+${modelBreakdown}
+\`\`\`
+
+${summaryComparison}
+
+<details>
+<summary><strong>View detailed breakdown by dataset and model</strong></summary>
+
+#### Performance by Dataset
+
+${datasetBreakdown}
+#### Performance by Model
+
+${modelPerformance}
+#### Methodology
+
+- **Semantic validation**: LLM-as-judge validates responses semantically (not exact string matching).
+- **Token counting**: Using \`gpt-tokenizer\` with \`o200k_base\` encoding.
+- **Question types**: Field retrieval, aggregation, and filtering tasks.
+- **Real data**: Faker.js-generated datasets + GitHub repositories.
+
+</details>
+`.trimStart()
 }
 
 /**
  * Calculate token counts for all format+dataset combinations
  */
 export function calculateTokenCounts(
-  formatters: Record<string, (data: any) => string>,
+  formatters: Record<string, (data: unknown) => string>,
 ): Record<string, number> {
   const tokenCounts: Record<string, number> = {}
 
@@ -272,7 +269,7 @@ export async function saveResults(
 }
 
 /**
- * Generate visual progress bar using ASCII characters (█ for filled, ░ for empty)
+ * Generate visual progress bar using ASCII characters (`█` for filled, `░` for empty)
  */
 function createProgressBar(tokens: number, maxTokens: number, width = 30): string {
   const filled = Math.round((tokens / maxTokens) * width)

benchmarks/src/types.ts

@@ -18,7 +18,7 @@ export interface EvaluationResult {
   model: string
   expected: string
   actual: string
-  correct: boolean
+  isCorrect: boolean
   inputTokens?: number
   outputTokens?: number
   latencyMs: number