thumbgate 0.9.9

package/scripts/export-dpo-pairs.js

@@ -0,0 +1,344 @@
+#!/usr/bin/env node
+/**
+ * DPO Preference Pair Exporter
+ *
+ * Transforms error + learning memories into DPO JSONL triples.
+ */
+
+const fs = require('fs');
+const path = require('path');
+const { traceForDpoPair, aggregateTraces } = require('./code-reasoning');
+const { resolveFeedbackDir } = require('./feedback-paths');
+
+const DEFAULT_LOCAL_MEMORY_LOG = path.join(resolveFeedbackDir(), 'memory-log.jsonl');
+
+function readJSONL(filePath) {
+  if (!fs.existsSync(filePath)) return [];
+  const raw = fs.readFileSync(filePath, 'utf-8').trim();
+  if (!raw) return [];
+  return raw
+    .split('\n')
+    .map((line) => {
+      try {
+        return JSON.parse(line);
+      } catch {
+        return null;
+      }
+    })
+    .filter(Boolean);
+}
+
+function extractDomainKeys(memory) {
+  const keys = new Set();
+  const genericTags = new Set(['feedback', 'positive', 'negative']);
+
+  if (Array.isArray(memory.tags)) {
+    for (const tag of memory.tags) {
+      if (!genericTags.has(tag)) keys.add(tag);
+    }
+  }
+
+  const titleWords = (memory.title || '')
+    .replace(/^(MISTAKE|SUCCESS|ERROR|LEARNING|PREFERENCE):\s*/i, '')
+    .toLowerCase()
+    .split(/\s+/)
+    .filter((w) => w.length > 3)
+    .slice(0, 3);
+
+  for (const word of titleWords) keys.add(word);
+
+  return Array.from(keys);
+}
+
+function domainOverlap(keysA, keysB) {
+  const setB = new Set(keysB);
+  let overlap = 0;
+  for (const key of keysA) {
+    if (setB.has(key)) overlap++;
+  }
+  return overlap;
+}
+
+function inferPrompt(error, learning) {
+  const shared = (error.tags || []).filter((t) => (learning.tags || []).includes(t));
+  if (shared.length > 0) {
+    return `Task domain: ${shared.join(', ')}. How should the agent handle this scenario?`;
+  }
+
+  const scenario = (error.title || '').replace(/^(MISTAKE|ERROR):\s*/i, '').trim();
+  if (scenario) return `Scenario: ${scenario}. What is the better response?`;
+  return 'How should the agent respond in this situation?';
+}
+
+function getRubricWeightedScore(memory) {
+  if (!memory || !memory.rubricSummary) return null;
+  const score = Number(memory.rubricSummary.weightedScore);
+  if (!Number.isFinite(score)) return null;
+  return score;
+}
+
+function buildRubricDelta(error, learning) {
+  const errorScore = getRubricWeightedScore(error);
+  const learningScore = getRubricWeightedScore(learning);
+  if (errorScore == null && learningScore == null) return null;
+  const delta = (learningScore != null && errorScore != null)
+    ? Math.round((learningScore - errorScore) * 1000) / 1000
+    : null;
+  return {
+    learningWeightedScore: learningScore,
+    errorWeightedScore: errorScore,
+    weightedDelta: delta,
+    errorFailingCriteria: error && error.rubricSummary ? error.rubricSummary.failingCriteria || [] : [],
+    learningFailingCriteria: learning && learning.rubricSummary ? learning.rubricSummary.failingCriteria || [] : [],
+  };
+}
+
+/**
+ * Find distractor errors for a learning: same-domain errors that are NOT
+ * the best match but look similar (high overlap, low rubric delta).
+ * These are harder negatives that train retrieval precision.
+ */
+function findDistractors(learning, errorKeys, usedErrorId, maxDistractors) {
+  const learningKeys = extractDomainKeys(learning);
+  return errorKeys
+    .filter((err) => err.memory.id !== usedErrorId)
+    .map((err) => {
+      const overlap = domainOverlap(err.keys, learningKeys);
+      return { memory: err.memory, overlap };
+    })
+    .filter((d) => d.overlap > 0)
+    .sort((a, b) => b.overlap - a.overlap)
+    .slice(0, maxDistractors)
+    .map((d) => ({
+      id: d.memory.id,
+      title: d.memory.title,
+      content: d.memory.content,
+      overlap: d.overlap,
+    }));
+}
+
+function buildDpoPairs(errors, learnings, { maxDistractors = 2 } = {}) {
+  const pairs = [];
+  const usedErrors = new Set();
+  const usedLearnings = new Set();
+
+  const errorKeys = errors.map((e) => ({ memory: e, keys: extractDomainKeys(e) }));
+  const learningKeys = learnings.map((l) => ({ memory: l, keys: extractDomainKeys(l) }));
+
+  for (const err of errorKeys) {
+    let best = null;
+    let bestScore = 0;
+    let bestOverlap = 0;
+
+    for (const learn of learningKeys) {
+      if (usedLearnings.has(learn.memory.id)) continue;
+      const overlap = domainOverlap(err.keys, learn.keys);
+      const rubric = buildRubricDelta(err.memory, learn.memory);
+      const rubricDelta = rubric && rubric.weightedDelta != null ? rubric.weightedDelta : 0;
+      const score = overlap + Math.max(0, rubricDelta);
+      if (score > bestScore) {
+        best = learn;
+        bestScore = score;
+        bestOverlap = overlap;
+      }
+    }
+
+    if (best && bestScore > 0 && bestOverlap > 0) {
+      const distractors = findDistractors(best.memory, errorKeys, err.memory.id, maxDistractors);
+      pairs.push({
+        prompt: inferPrompt(err.memory, best.memory),
+        chosen: best.memory.content,
+        rejected: err.memory.content,
+        distractors: distractors.length > 0 ? distractors : undefined,
+        metadata: {
+          errorId: err.memory.id,
+          learningId: best.memory.id,
+          matchScore: bestScore,
+          overlapScore: domainOverlap(err.keys, best.keys),
+          matchedKeys: err.keys.filter((k) => best.keys.includes(k)),
+          errorTitle: err.memory.title,
+          learningTitle: best.memory.title,
+          rubric: buildRubricDelta(err.memory, best.memory),
+          distractorCount: distractors.length,
+        },
+      });
+      usedErrors.add(err.memory.id);
+      usedLearnings.add(best.memory.id);
+    }
+  }
+
+  return {
+    pairs,
+    unpairedErrors: errors.filter((e) => !usedErrors.has(e.id)),
+    unpairedLearnings: learnings.filter((l) => !usedLearnings.has(l.id)),
+  };
+}
+
+function toJSONL(pairs) {
+  return `${pairs.map((p) => JSON.stringify(p)).join('\n')}\n`;
+}
+
+function exportDpoFromMemories(memories) {
+  const errors = memories.filter((m) => m.category === 'error');
+  const learnings = memories.filter((m) => m.category === 'learning');
+  const result = buildDpoPairs(errors, learnings);
+
+  const traces = result.pairs.map((pair) => traceForDpoPair(pair));
+  const reasoning = aggregateTraces(traces);
+
+  const pairsWithTraces = result.pairs.map((pair, i) => ({
+    ...pair,
+    metadata: {
+      ...pair.metadata,
+      reasoningTrace: {
+        traceId: traces[i].traceId,
+        confidence: traces[i].summary.confidence,
+        passed: traces[i].summary.passed,
+        verified: traces[i].summary.verified,
+        refuted: traces[i].summary.refuted,
+        edgeCases: traces[i].edgeCases,
+      },
+    },
+  }));
+
+  return {
+    pairs: pairsWithTraces,
+    unpairedErrors: result.unpairedErrors,
+    unpairedLearnings: result.unpairedLearnings,
+    errors,
+    learnings,
+    reasoning,
+    jsonl: toJSONL(pairsWithTraces),
+  };
+}
+
+function parseArgs(argv) {
+  const args = {};
+  argv.forEach((arg) => {
+    if (!arg.startsWith('--')) return;
+    const [key, ...rest] = arg.slice(2).split('=');
+    args[key] = rest.length ? rest.join('=') : true;
+  });
+  return args;
+}
+
+function runCli() {
+  const args = parseArgs(process.argv.slice(2));
+
+  if (args.test) {
+    runTests();
+    return;
+  }
+
+  let memories = [];
+
+  if (args.input) {
+    const raw = fs.readFileSync(args.input, 'utf-8');
+    const parsed = JSON.parse(raw);
+    memories = Array.isArray(parsed) ? parsed : parsed.memories || [];
+  } else if (args['from-local']) {
+    memories = readJSONL(DEFAULT_LOCAL_MEMORY_LOG);
+  } else {
+    console.error('Provide --input=<path-to-json> or --from-local');
+    process.exit(1);
+  }
+
+  const result = exportDpoFromMemories(memories);
+  const jsonl = result.jsonl;
+
+  if (args.output) {
+    fs.writeFileSync(args.output, jsonl);
+    console.error(`Wrote ${result.pairs.length} DPO pairs to ${args.output}`);
+  } else {
+    process.stdout.write(jsonl);
+  }
+
+  console.error(`Errors=${result.errors.length} Learnings=${result.learnings.length} Pairs=${result.pairs.length}`);
+  console.error(`Unpaired errors=${result.unpairedErrors.length} Unpaired learnings=${result.unpairedLearnings.length}`);
+}
+
+function runTests() {
+  let passed = 0;
+  let failed = 0;
+
+  function assert(condition, name) {
+    if (condition) {
+      passed++;
+      console.log(`  PASS ${name}`);
+    } else {
+      failed++;
+      console.log(`  FAIL ${name}`);
+    }
+  }
+
+  console.log('\nexport-dpo-pairs.js tests\n');
+
+  const errors = [
+    {
+      id: 1,
+      title: 'MISTAKE: Claimed done with no test proof',
+      content: 'Claimed completion without running tests.',
+      category: 'error',
+      tags: ['verification', 'feedback'],
+      rubricSummary: {
+        weightedScore: 0.32,
+        failingCriteria: ['verification_evidence'],
+        failingGuardrails: ['testsPassed'],
+      },
+    },
+    {
+      id: 2,
+      title: 'MISTAKE: Generic mismatch',
+      content: 'No matching learning memory for this domain.',
+      category: 'error',
+      tags: ['unique-tag'],
+    },
+  ];
+
+  const learnings = [
+    {
+      id: 10,
+      title: 'SUCCESS: Always run tests before completion claims',
+      content: 'Run tests and include output before saying complete.',
+      category: 'learning',
+      tags: ['verification', 'feedback'],
+      rubricSummary: {
+        weightedScore: 0.89,
+        failingCriteria: [],
+        failingGuardrails: [],
+      },
+    },
+  ];
+
+  const result = buildDpoPairs(errors, learnings);
+  assert(result.pairs.length === 1, 'one pair built from overlapping domain keys');
+  assert(result.unpairedErrors.length === 1, 'unpaired error left when no match exists');
+  assert(result.unpairedLearnings.length === 0, 'no unpaired learnings');
+
+  const jsonl = toJSONL(result.pairs);
+  assert(jsonl.endsWith('\n'), 'JSONL output ends with newline');
+
+  const parsed = JSON.parse(jsonl.trim());
+  assert(parsed.prompt.includes('verification'), 'inferred prompt includes shared domain');
+  assert(parsed.metadata.rubric.weightedDelta > 0, 'rubric delta metadata is attached');
+
+  console.log(`\nResults: ${passed} passed, ${failed} failed\n`);
+  process.exit(failed > 0 ? 1 : 0);
+}
+
+module.exports = {
+  readJSONL,
+  extractDomainKeys,
+  domainOverlap,
+  inferPrompt,
+  buildRubricDelta,
+  findDistractors,
+  buildDpoPairs,
+  toJSONL,
+  exportDpoFromMemories,
+  DEFAULT_LOCAL_MEMORY_LOG,
+};
+
+if (require.main === module) {
+  runCli();
+}

package/scripts/export-kto-pairs.js

@@ -0,0 +1,309 @@
+#!/usr/bin/env node
+/**
+ * KTO (Kahneman-Tversky Optimization) Exporter
+ *
+ * Transforms binary up/down feedback into KTO JSONL records.
+ * Unlike DPO (which needs paired preferences), KTO works with
+ * individual binary signals — a natural fit for thumbs-up/down data.
+ *
+ * Output format per line:
+ *   {"prompt": "...", "completion": "...", "label": true/false, "metadata": {...}}
+ */
+
+const fs = require('fs');
+const path = require('path');
+const { resolveFeedbackDir } = require('./feedback-paths');
+
+const DEFAULT_FEEDBACK_LOG = path.join(resolveFeedbackDir(), 'feedback-log.jsonl');
+const DEFAULT_MEMORY_LOG = path.join(resolveFeedbackDir(), 'memory-log.jsonl');
+
+function readJSONL(filePath) {
+  if (!fs.existsSync(filePath)) return [];
+  const raw = fs.readFileSync(filePath, 'utf-8').trim();
+  if (!raw) return [];
+  return raw
+    .split('\n')
+    .map((line) => {
+      try {
+        return JSON.parse(line);
+      } catch {
+        return null;
+      }
+    })
+    .filter(Boolean);
+}
+
+/**
+ * Infer a prompt string from a feedback entry.
+ * Uses context, tags, or domain info to reconstruct what was being asked.
+ */
+function inferPrompt(entry) {
+  if (entry.context && entry.context.trim()) {
+    return entry.context.trim();
+  }
+  if (entry.richContext && entry.richContext.domain) {
+    return `Task domain: ${entry.richContext.domain}`;
+  }
+  if (Array.isArray(entry.tags) && entry.tags.length > 0) {
+    return `Task: ${entry.tags.join(', ')}`;
+  }
+  return 'General coding task';
+}
+
+/**
+ * Infer a completion string from a feedback entry.
+ * For positive: whatWorked or content describes the good response.
+ * For negative: whatWentWrong or whatToChange describes the bad response.
+ */
+function inferCompletion(entry) {
+  const signal = normalizeSignal(entry.signal);
+  if (signal === 'positive') {
+    if (entry.whatWorked && entry.whatWorked.trim()) return entry.whatWorked.trim();
+    if (entry.content && entry.content.trim()) return entry.content.trim();
+    return 'Completed task successfully';
+  }
+  if (entry.whatWentWrong && entry.whatWentWrong.trim()) return entry.whatWentWrong.trim();
+  if (entry.whatToChange && entry.whatToChange.trim()) return entry.whatToChange.trim();
+  if (entry.content && entry.content.trim()) return entry.content.trim();
+  return 'Failed to complete task correctly';
+}
+
+function normalizeSignal(signal) {
+  const value = String(signal || '').trim().toLowerCase();
+  if (['up', 'thumbsup', 'thumbs-up', 'thumbs_up', 'positive', 'good'].includes(value)) return 'positive';
+  if (['down', 'thumbsdown', 'thumbs-down', 'thumbs_down', 'negative', 'bad'].includes(value)) return 'negative';
+  return null;
+}
+
+/**
+ * Build a single KTO record from a feedback or memory entry.
+ * Returns null if the entry lacks a valid signal.
+ */
+function buildKtoRecord(entry) {
+  const signal = normalizeSignal(entry.signal);
+  if (!signal) return null;
+
+  const label = signal === 'positive';
+  const prompt = inferPrompt(entry);
+  const completion = inferCompletion(entry);
+
+  return {
+    prompt,
+    completion,
+    label,
+    metadata: {
+      sourceId: entry.id || null,
+      signal,
+      signalSource: entry.sourceFeedbackId ? 'memory-log' : 'feedback-log',
+      tags: entry.tags || [],
+      domain: (entry.richContext && entry.richContext.domain) || null,
+      outcomeCategory: (entry.richContext && entry.richContext.outcomeCategory) || null,
+      timestamp: entry.timestamp || null,
+      rubricScore: (entry.rubric && entry.rubric.weightedScore != null)
+        ? entry.rubric.weightedScore
+        : null,
+    },
+  };
+}
+
+/**
+ * Build KTO records from an array of feedback/memory entries.
+ */
+function buildKtoPairs(entries) {
+  const records = [];
+  const skipped = [];
+  for (const entry of entries) {
+    const record = buildKtoRecord(entry);
+    if (record) {
+      records.push(record);
+    } else {
+      skipped.push(entry);
+    }
+  }
+  return { records, skipped };
+}
+
+function toJSONL(records) {
+  if (records.length === 0) return '';
+  return `${records.map((r) => JSON.stringify(r)).join('\n')}\n`;
+}
+
+function exportKtoFromFeedback(feedbackEntries, memoryEntries) {
+  const all = [...feedbackEntries, ...memoryEntries];
+  // Deduplicate by id
+  const seen = new Set();
+  const unique = [];
+  for (const entry of all) {
+    const key = entry.id || JSON.stringify(entry);
+    if (!seen.has(key)) {
+      seen.add(key);
+      unique.push(entry);
+    }
+  }
+  const result = buildKtoPairs(unique);
+  return {
+    records: result.records,
+    skipped: result.skipped,
+    totalInput: unique.length,
+    jsonl: toJSONL(result.records),
+  };
+}
+
+function parseArgs(argv) {
+  const args = {};
+  argv.forEach((arg) => {
+    if (!arg.startsWith('--')) return;
+    const [key, ...rest] = arg.slice(2).split('=');
+    args[key] = rest.length ? rest.join('=') : true;
+  });
+  return args;
+}
+
+function runCli() {
+  const args = parseArgs(process.argv.slice(2));
+
+  if (args.test) {
+    runTests();
+    return;
+  }
+
+  let feedbackEntries = [];
+  let memoryEntries = [];
+
+  if (args.input) {
+    const raw = fs.readFileSync(args.input, 'utf-8');
+    const parsed = JSON.parse(raw);
+    feedbackEntries = Array.isArray(parsed) ? parsed : parsed.entries || [];
+  } else if (args['from-local']) {
+    feedbackEntries = readJSONL(DEFAULT_FEEDBACK_LOG);
+    memoryEntries = readJSONL(DEFAULT_MEMORY_LOG);
+  } else {
+    console.error('Provide --input=<path-to-json> or --from-local');
+    process.exit(1);
+  }
+
+  const result = exportKtoFromFeedback(feedbackEntries, memoryEntries);
+
+  if (args.output) {
+    fs.writeFileSync(args.output, result.jsonl);
+    console.error(`Wrote ${result.records.length} KTO records to ${args.output}`);
+  } else {
+    process.stdout.write(result.jsonl);
+  }
+
+  const positiveCount = result.records.filter((r) => r.label === true).length;
+  const negativeCount = result.records.filter((r) => r.label === false).length;
+  console.error(`Total=${result.totalInput} Exported=${result.records.length} Positive=${positiveCount} Negative=${negativeCount} Skipped=${result.skipped.length}`);
+}
+
+function runTests() {
+  let passed = 0;
+  let failed = 0;
+
+  function assert(condition, name) {
+    if (condition) {
+      passed++;
+      console.log(`  PASS ${name}`);
+    } else {
+      failed++;
+      console.log(`  FAIL ${name}`);
+    }
+  }
+
+  console.log('\nexport-kto-pairs.js tests\n');
+
+  // Test 1: positive signal produces label true
+  const pos = buildKtoRecord({
+    id: 'fb_1',
+    signal: 'positive',
+    context: 'Implemented auth',
+    whatWorked: 'JWT tokens with refresh rotation',
+    tags: ['auth'],
+    timestamp: '2025-01-01T00:00:00Z',
+  });
+  assert(pos !== null, 'positive signal produces a record');
+  assert(pos.label === true, 'positive signal produces label: true');
+
+  // Test 2: negative signal produces label false
+  const neg = buildKtoRecord({
+    id: 'fb_2',
+    signal: 'negative',
+    context: 'Tried to deploy',
+    whatWentWrong: 'Missing env vars',
+    tags: ['deploy'],
+    timestamp: '2025-01-01T00:00:00Z',
+  });
+  assert(neg !== null, 'negative signal produces a record');
+  assert(neg.label === false, 'negative signal produces label: false');
+
+  // Test 3: missing context handled gracefully
+  const noCtx = buildKtoRecord({
+    id: 'fb_3',
+    signal: 'up',
+    tags: ['testing'],
+  });
+  assert(noCtx !== null, 'entry with missing context still produces record');
+  assert(noCtx.prompt === 'Task: testing', 'missing context falls back to tags');
+
+  // Test 4: invalid signal returns null
+  const invalid = buildKtoRecord({ id: 'fb_4', signal: 'maybe' });
+  assert(invalid === null, 'invalid signal returns null');
+
+  // Test 5: JSONL output is valid
+  const records = [pos, neg];
+  const jsonl = toJSONL(records);
+  const lines = jsonl.trim().split('\n');
+  let allValid = true;
+  for (const line of lines) {
+    try {
+      JSON.parse(line);
+    } catch {
+      allValid = false;
+    }
+  }
+  assert(allValid, 'JSONL output is valid JSON per line');
+  assert(jsonl.endsWith('\n'), 'JSONL output ends with newline');
+
+  // Test 6: metadata includes signal source and timestamp
+  assert(pos.metadata.signalSource === 'feedback-log', 'metadata includes signal source');
+  assert(pos.metadata.timestamp === '2025-01-01T00:00:00Z', 'metadata includes timestamp');
+  assert(pos.metadata.signal === 'positive', 'metadata includes normalized signal');
+
+  // Test 7: empty context with richContext domain
+  const richCtx = buildKtoRecord({
+    id: 'fb_5',
+    signal: 'up',
+    richContext: { domain: 'security', outcomeCategory: 'quick-success' },
+  });
+  assert(richCtx.prompt === 'Task domain: security', 'richContext domain used as prompt fallback');
+  assert(richCtx.metadata.domain === 'security', 'metadata captures domain');
+
+  // Test 8: buildKtoPairs filters bad entries
+  const result = buildKtoPairs([
+    { id: 'a', signal: 'up', context: 'good' },
+    { id: 'b', signal: 'invalid' },
+    { id: 'c', signal: 'down', context: 'bad', whatWentWrong: 'broke it' },
+  ]);
+  assert(result.records.length === 2, 'buildKtoPairs keeps valid entries');
+  assert(result.skipped.length === 1, 'buildKtoPairs tracks skipped entries');
+
+  console.log(`\nResults: ${passed} passed, ${failed} failed\n`);
+  process.exit(failed > 0 ? 1 : 0);
+}
+
+module.exports = {
+  readJSONL,
+  normalizeSignal,
+  inferPrompt,
+  inferCompletion,
+  buildKtoRecord,
+  buildKtoPairs,
+  toJSONL,
+  exportKtoFromFeedback,
+  DEFAULT_FEEDBACK_LOG,
+  DEFAULT_MEMORY_LOG,
+};
+
+if (require.main === module) {
+  runCli();
+}