panopticon-cli 0.5.1 → 0.5.4

package/scripts/recover-costs-deep.mjs

@@ -0,0 +1,209 @@
+#!/usr/bin/env node
+/**
+ * Deep cost recovery: scans ALL Claude transcripts, including non-workspace ones.
+ * For transcripts not in a workspace dir, infers issue ID from conversation content.
+ */
+
+import { readdirSync, readFileSync, statSync } from 'fs';
+import { join, basename } from 'path';
+import { homedir } from 'os';
+import Database from 'better-sqlite3';
+
+const CLAUDE_PROJECTS = join(homedir(), '.claude', 'projects');
+const DB_PATH = join(homedir(), '.panopticon', 'panopticon.db');
+
+const PRICING = [
+  { provider: 'anthropic', model: 'claude-opus-4.6', inputPer1k: 5e-3, outputPer1k: 0.025, cacheReadPer1k: 5e-4, cacheWrite5mPer1k: 625e-5 },
+  { provider: 'anthropic', model: 'claude-opus-4-6', inputPer1k: 5e-3, outputPer1k: 0.025, cacheReadPer1k: 5e-4, cacheWrite5mPer1k: 625e-5 },
+  { provider: 'anthropic', model: 'claude-opus-4-1', inputPer1k: 0.015, outputPer1k: 0.075, cacheReadPer1k: 15e-4, cacheWrite5mPer1k: 0.01875 },
+  { provider: 'anthropic', model: 'claude-opus-4', inputPer1k: 0.015, outputPer1k: 0.075, cacheReadPer1k: 15e-4, cacheWrite5mPer1k: 0.01875 },
+  { provider: 'anthropic', model: 'claude-sonnet-4.5', inputPer1k: 3e-3, outputPer1k: 0.015, cacheReadPer1k: 3e-4, cacheWrite5mPer1k: 375e-5 },
+  { provider: 'anthropic', model: 'claude-sonnet-4-6', inputPer1k: 3e-3, outputPer1k: 0.015, cacheReadPer1k: 3e-4, cacheWrite5mPer1k: 375e-5 },
+  { provider: 'anthropic', model: 'claude-sonnet-4', inputPer1k: 3e-3, outputPer1k: 0.015, cacheReadPer1k: 3e-4, cacheWrite5mPer1k: 375e-5 },
+  { provider: 'anthropic', model: 'claude-haiku-4.5', inputPer1k: 1e-3, outputPer1k: 5e-3, cacheReadPer1k: 1e-4, cacheWrite5mPer1k: 125e-5 },
+  { provider: 'anthropic', model: 'claude-haiku-4', inputPer1k: 1e-3, outputPer1k: 5e-3, cacheReadPer1k: 1e-4, cacheWrite5mPer1k: 125e-5 },
+  { provider: 'anthropic', model: 'claude-haiku-3', inputPer1k: 25e-5, outputPer1k: 125e-5, cacheReadPer1k: 3e-5, cacheWrite5mPer1k: 3e-4 },
+  { provider: 'custom', model: 'kimi-k2.5', inputPer1k: 6e-4, outputPer1k: 2e-3, cacheReadPer1k: 6e-5, cacheWrite5mPer1k: 75e-5 },
+  { provider: 'custom', model: 'kimi-for-coding', inputPer1k: 6e-4, outputPer1k: 2e-3, cacheReadPer1k: 6e-5, cacheWrite5mPer1k: 75e-5 },
+];
+
+function getPricing(model) {
+  return PRICING.find(p => model.startsWith(p.model)) || PRICING.find(p => model.includes(p.model)) || null;
+}
+
+function calculateCost(usage, pricing) {
+  let cost = 0;
+  let inputMul = 1, outputMul = 1;
+  const totalInput = usage.input + (usage.cacheRead || 0) + (usage.cacheWrite || 0);
+  if (pricing.model.includes('sonnet-4') && totalInput > 200000) { inputMul = 2; outputMul = 1.5; }
+  cost += usage.input / 1000 * pricing.inputPer1k * inputMul;
+  cost += usage.output / 1000 * pricing.outputPer1k * outputMul;
+  if (usage.cacheRead && pricing.cacheReadPer1k) cost += usage.cacheRead / 1000 * pricing.cacheReadPer1k;
+  if (usage.cacheWrite && pricing.cacheWrite5mPer1k) cost += usage.cacheWrite / 1000 * pricing.cacheWrite5mPer1k;
+  return Math.round(cost * 1e6) / 1e6;
+}
+
+// Issue ID pattern: PAN-123, MIN-456, KRUX-1, CLI-1, AUR-1, etc.
+const ISSUE_RE = /\b(PAN|MIN|AUR|KRUX|CLI)-(\d+)\b/gi;
+
+function inferIssueFromPath(dirName) {
+  const match = dirName.match(/(pan|min|aud|krux|cli)[-](\d+)/i);
+  if (match) return `${match[1].toUpperCase()}-${match[2]}`;
+  return null;
+}
+
+/**
+ * Infer the primary issue from transcript content by counting mentions.
+ * Only considers user and assistant messages, not system/tool content.
+ */
+function inferIssueFromContent(lines) {
+  const counts = {};
+  for (const line of lines) {
+    if (!line.trim()) continue;
+    try {
+      const entry = JSON.parse(line);
+      // Only look at human and assistant messages for issue mentions
+      if (entry.type !== 'human' && entry.type !== 'assistant') continue;
+      const text = JSON.stringify(entry.message || '');
+      let match;
+      const re = new RegExp(ISSUE_RE.source, 'gi');
+      while ((match = re.exec(text)) !== null) {
+        const id = `${match[1].toUpperCase()}-${match[2]}`;
+        counts[id] = (counts[id] || 0) + 1;
+      }
+    } catch {}
+  }
+
+  // Return the most-mentioned issue (if any has 2+ mentions)
+  const sorted = Object.entries(counts).sort((a, b) => b[1] - a[1]);
+  if (sorted.length > 0 && sorted[0][1] >= 2) {
+    return sorted[0][0];
+  }
+  return null;
+}
+
+function findTranscriptFiles(dir) {
+  const files = [];
+  try {
+    for (const entry of readdirSync(dir)) {
+      if (entry.endsWith('.jsonl')) {
+        const full = join(dir, entry);
+        try { if (statSync(full).isFile()) files.push(full); } catch {}
+      }
+    }
+  } catch {}
+  return files;
+}
+
+// Main
+const db = new Database(DB_PATH);
+db.pragma('journal_mode = WAL');
+
+const insert = db.prepare(`
+  INSERT OR IGNORE INTO cost_events (
+    ts, agent_id, issue_id, session_type, provider, model,
+    input, output, cache_read, cache_write, cost, request_id, source_file
+  ) VALUES (?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?)
+`);
+
+let totalInserted = 0;
+let totalDuplicates = 0;
+let totalUnattributed = 0;
+const issueStats = {};
+
+const projectDirs = readdirSync(CLAUDE_PROJECTS);
+
+for (const dirName of projectDirs) {
+  const projectDir = join(CLAUDE_PROJECTS, dirName);
+  try { if (!statSync(projectDir).isDirectory()) continue; } catch { continue; }
+
+  // Try to get issue from path first
+  const pathIssueId = inferIssueFromPath(dirName);
+
+  const transcripts = findTranscriptFiles(projectDir);
+  if (transcripts.length === 0) continue;
+
+  for (const transcript of transcripts) {
+    let content;
+    try { content = readFileSync(transcript, 'utf-8'); } catch { continue; }
+    const lines = content.split('\n');
+
+    // Determine issue ID: path first, then content inference
+    let issueId = pathIssueId;
+    if (!issueId) {
+      issueId = inferIssueFromContent(lines);
+    }
+
+    if (!issueId) {
+      // Count usage events we're skipping
+      for (const line of lines) {
+        try {
+          const entry = JSON.parse(line);
+          if (entry.type === 'assistant' && entry.message?.usage) {
+            const u = entry.message.usage;
+            if ((u.input_tokens || 0) + (u.output_tokens || 0) > 0) totalUnattributed++;
+          }
+        } catch {}
+      }
+      continue;
+    }
+
+    for (const line of lines) {
+      if (!line.trim()) continue;
+      try {
+        const entry = JSON.parse(line);
+        if (entry.type !== 'assistant' || !entry.message?.usage) continue;
+
+        const usage = entry.message.usage;
+        const model = entry.message.model || 'claude-sonnet-4';
+        const requestId = entry.requestId;
+        if (!requestId) continue;
+
+        const input = usage.input_tokens || 0;
+        const output = usage.output_tokens || 0;
+        const cacheRead = usage.cache_read_input_tokens || 0;
+        const cacheWrite = usage.cache_creation_input_tokens || 0;
+        if (input === 0 && output === 0 && cacheRead === 0 && cacheWrite === 0) continue;
+
+        let provider = 'anthropic';
+        if (model.includes('gpt')) provider = 'openai';
+        else if (model.includes('gemini')) provider = 'google';
+        else if (model.includes('kimi')) provider = 'custom';
+
+        const pricing = getPricing(model);
+        if (!pricing) continue;
+
+        const cost = calculateCost({ input, output, cacheRead, cacheWrite }, pricing);
+        const ts = entry.timestamp || new Date(statSync(transcript).mtime).toISOString();
+
+        const result = insert.run(
+          ts, 'recovered-deep', issueId, 'interactive', provider, model,
+          input, output, cacheRead, cacheWrite, cost, requestId, basename(transcript)
+        );
+
+        if (result.changes > 0) {
+          totalInserted++;
+          if (!issueStats[issueId]) issueStats[issueId] = { inserted: 0, cost: 0 };
+          issueStats[issueId].inserted++;
+          issueStats[issueId].cost += cost;
+        } else {
+          totalDuplicates++;
+        }
+      } catch {}
+    }
+  }
+}
+
+db.close();
+
+console.log(`\nDeep Cost Recovery Complete`);
+console.log(`  NEW events inserted: ${totalInserted}`);
+console.log(`  Duplicates skipped: ${totalDuplicates}`);
+console.log(`  Unattributable events: ${totalUnattributed}`);
+console.log(`\nNewly recovered costs by issue:`);
+const sorted = Object.entries(issueStats).sort((a, b) => b[1].cost - a[1].cost);
+for (const [id, stats] of sorted) {
+  console.log(`  ${id.padEnd(12)} ${String(stats.inserted).padStart(5)} events  $${stats.cost.toFixed(2)}`);
+}
+const totalCost = sorted.reduce((sum, [, s]) => sum + s.cost, 0);
+console.log(`\n  TOTAL NEWLY RECOVERED: $${totalCost.toFixed(2)}`);

package/scripts/recover-costs-proportional.mjs

@@ -0,0 +1,206 @@
+#!/usr/bin/env node
+/**
+ * Proportional cost recovery: for multi-issue sessions, splits costs
+ * based on which issue was being discussed at each point in the conversation.
+ *
+ * Strategy: for each assistant response with usage, look at the surrounding
+ * context (the human message before it and assistant message itself) to determine
+ * which issue is being worked on at that moment.
+ */
+
+import { readdirSync, readFileSync, statSync } from 'fs';
+import { join, basename } from 'path';
+import { homedir } from 'os';
+import Database from 'better-sqlite3';
+
+const CLAUDE_PROJECTS = join(homedir(), '.claude', 'projects');
+const DB_PATH = join(homedir(), '.panopticon', 'panopticon.db');
+
+const PRICING = [
+  { provider: 'anthropic', model: 'claude-opus-4.6', inputPer1k: 5e-3, outputPer1k: 0.025, cacheReadPer1k: 5e-4, cacheWrite5mPer1k: 625e-5 },
+  { provider: 'anthropic', model: 'claude-opus-4-6', inputPer1k: 5e-3, outputPer1k: 0.025, cacheReadPer1k: 5e-4, cacheWrite5mPer1k: 625e-5 },
+  { provider: 'anthropic', model: 'claude-opus-4-1', inputPer1k: 0.015, outputPer1k: 0.075, cacheReadPer1k: 15e-4, cacheWrite5mPer1k: 0.01875 },
+  { provider: 'anthropic', model: 'claude-opus-4', inputPer1k: 0.015, outputPer1k: 0.075, cacheReadPer1k: 15e-4, cacheWrite5mPer1k: 0.01875 },
+  { provider: 'anthropic', model: 'claude-sonnet-4.5', inputPer1k: 3e-3, outputPer1k: 0.015, cacheReadPer1k: 3e-4, cacheWrite5mPer1k: 375e-5 },
+  { provider: 'anthropic', model: 'claude-sonnet-4-6', inputPer1k: 3e-3, outputPer1k: 0.015, cacheReadPer1k: 3e-4, cacheWrite5mPer1k: 375e-5 },
+  { provider: 'anthropic', model: 'claude-sonnet-4', inputPer1k: 3e-3, outputPer1k: 0.015, cacheReadPer1k: 3e-4, cacheWrite5mPer1k: 375e-5 },
+  { provider: 'anthropic', model: 'claude-haiku-4.5', inputPer1k: 1e-3, outputPer1k: 5e-3, cacheReadPer1k: 1e-4, cacheWrite5mPer1k: 125e-5 },
+  { provider: 'anthropic', model: 'claude-haiku-4', inputPer1k: 1e-3, outputPer1k: 5e-3, cacheReadPer1k: 1e-4, cacheWrite5mPer1k: 125e-5 },
+  { provider: 'anthropic', model: 'claude-haiku-3', inputPer1k: 25e-5, outputPer1k: 125e-5, cacheReadPer1k: 3e-5, cacheWrite5mPer1k: 3e-4 },
+  { provider: 'custom', model: 'kimi-k2.5', inputPer1k: 6e-4, outputPer1k: 2e-3, cacheReadPer1k: 6e-5, cacheWrite5mPer1k: 75e-5 },
+  { provider: 'custom', model: 'kimi-for-coding', inputPer1k: 6e-4, outputPer1k: 2e-3, cacheReadPer1k: 6e-5, cacheWrite5mPer1k: 75e-5 },
+];
+
+function getPricing(model) {
+  return PRICING.find(p => model.startsWith(p.model)) || PRICING.find(p => model.includes(p.model)) || null;
+}
+
+function calculateCost(usage, pricing) {
+  let cost = 0;
+  let inputMul = 1, outputMul = 1;
+  const totalInput = usage.input + (usage.cacheRead || 0) + (usage.cacheWrite || 0);
+  if (pricing.model.includes('sonnet-4') && totalInput > 200000) { inputMul = 2; outputMul = 1.5; }
+  cost += usage.input / 1000 * pricing.inputPer1k * inputMul;
+  cost += usage.output / 1000 * pricing.outputPer1k * outputMul;
+  if (usage.cacheRead && pricing.cacheReadPer1k) cost += usage.cacheRead / 1000 * pricing.cacheReadPer1k;
+  if (usage.cacheWrite && pricing.cacheWrite5mPer1k) cost += usage.cacheWrite / 1000 * pricing.cacheWrite5mPer1k;
+  return Math.round(cost * 1e6) / 1e6;
+}
+
+const ISSUE_RE = /\b(PAN|MIN|AUR|KRUX|CLI)-(\d+)\b/gi;
+
+function extractIssues(text) {
+  const counts = {};
+  let match;
+  const re = new RegExp(ISSUE_RE.source, 'gi');
+  while ((match = re.exec(text)) !== null) {
+    const id = `${match[1].toUpperCase()}-${match[2]}`;
+    counts[id] = (counts[id] || 0) + 1;
+  }
+  return counts;
+}
+
+function inferIssueFromPath(dirName) {
+  const match = dirName.match(/(pan|min|aud|krux|cli)[-](\d+)/i);
+  if (match) return `${match[1].toUpperCase()}-${match[2]}`;
+  return null;
+}
+
+// Main
+const db = new Database(DB_PATH);
+db.pragma('journal_mode = WAL');
+
+const insert = db.prepare(`
+  INSERT OR IGNORE INTO cost_events (
+    ts, agent_id, issue_id, session_type, provider, model,
+    input, output, cache_read, cache_write, cost, request_id, source_file
+  ) VALUES (?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?)
+`);
+
+let totalInserted = 0;
+let totalDuplicates = 0;
+let totalUnattributed = 0;
+const issueStats = {};
+
+const projectDirs = readdirSync(CLAUDE_PROJECTS);
+
+for (const dirName of projectDirs) {
+  const projectDir = join(CLAUDE_PROJECTS, dirName);
+  try { if (!statSync(projectDir).isDirectory()) continue; } catch { continue; }
+
+  const pathIssueId = inferIssueFromPath(dirName);
+
+  let transcripts;
+  try {
+    transcripts = readdirSync(projectDir)
+      .filter(f => f.endsWith('.jsonl'))
+      .map(f => join(projectDir, f))
+      .filter(f => { try { return statSync(f).isFile(); } catch { return false; } });
+  } catch { continue; }
+
+  for (const transcript of transcripts) {
+    let content;
+    try { content = readFileSync(transcript, 'utf-8'); } catch { continue; }
+    const lines = content.split('\n').filter(l => l.trim());
+
+    // Parse all entries
+    const entries = [];
+    for (const line of lines) {
+      try { entries.push(JSON.parse(line)); } catch {}
+    }
+
+    // Track the "current issue context" as we walk through the conversation
+    let currentIssue = pathIssueId || null;
+    let lastHumanText = '';
+
+    for (let i = 0; i < entries.length; i++) {
+      const entry = entries[i];
+
+      // Track human messages to build context window
+      if (entry.type === 'human') {
+        const text = JSON.stringify(entry.message || '');
+        lastHumanText = text;
+        // Update current issue if this human message mentions issues
+        const issues = extractIssues(text);
+        const sorted = Object.entries(issues).sort((a, b) => b[1] - a[1]);
+        if (sorted.length > 0) {
+          currentIssue = sorted[0][0];
+        }
+        continue;
+      }
+
+      if (entry.type !== 'assistant' || !entry.message?.usage) continue;
+
+      const usage = entry.message.usage;
+      const model = entry.message.model || 'claude-sonnet-4';
+      const requestId = entry.requestId;
+      if (!requestId) continue;
+
+      const input = usage.input_tokens || 0;
+      const output = usage.output_tokens || 0;
+      const cacheRead = usage.cache_read_input_tokens || 0;
+      const cacheWrite = usage.cache_creation_input_tokens || 0;
+      if (input === 0 && output === 0 && cacheRead === 0 && cacheWrite === 0) continue;
+
+      // Check this assistant message for issue mentions too
+      const assistantText = JSON.stringify(entry.message?.content || '');
+      const assistantIssues = extractIssues(assistantText);
+      const combinedText = lastHumanText + ' ' + assistantText;
+      const contextIssues = extractIssues(combinedText);
+      const sorted = Object.entries(contextIssues).sort((a, b) => b[1] - a[1]);
+
+      // Use the most-mentioned issue in the immediate context, falling back to running context
+      let issueId = null;
+      if (sorted.length > 0) {
+        issueId = sorted[0][0];
+        currentIssue = issueId; // Update running context
+      } else {
+        issueId = currentIssue;
+      }
+
+      if (!issueId) {
+        totalUnattributed++;
+        continue;
+      }
+
+      let provider = 'anthropic';
+      if (model.includes('gpt')) provider = 'openai';
+      else if (model.includes('gemini')) provider = 'google';
+      else if (model.includes('kimi')) provider = 'custom';
+
+      const pricing = getPricing(model);
+      if (!pricing) continue;
+
+      const cost = calculateCost({ input, output, cacheRead, cacheWrite }, pricing);
+      const ts = entry.timestamp || new Date(statSync(transcript).mtime).toISOString();
+
+      const result = insert.run(
+        ts, 'recovered-proportional', issueId, 'interactive', provider, model,
+        input, output, cacheRead, cacheWrite, cost, requestId, basename(transcript)
+      );
+
+      if (result.changes > 0) {
+        totalInserted++;
+        if (!issueStats[issueId]) issueStats[issueId] = { inserted: 0, cost: 0 };
+        issueStats[issueId].inserted++;
+        issueStats[issueId].cost += cost;
+      } else {
+        totalDuplicates++;
+      }
+    }
+  }
+}
+
+db.close();
+
+console.log(`\nProportional Cost Recovery Complete`);
+console.log(`  NEW events inserted: ${totalInserted}`);
+console.log(`  Duplicates skipped: ${totalDuplicates}`);
+console.log(`  Unattributable: ${totalUnattributed}`);
+console.log(`\nNewly recovered costs by issue:`);
+const sorted = Object.entries(issueStats).sort((a, b) => b[1].cost - a[1].cost);
+for (const [id, stats] of sorted) {
+  console.log(`  ${id.padEnd(12)} ${String(stats.inserted).padStart(5)} events  $${stats.cost.toFixed(2)}`);
+}
+const totalCost = sorted.reduce((sum, [, s]) => sum + s.cost, 0);
+console.log(`\n  TOTAL NEWLY RECOVERED: $${totalCost.toFixed(2)}`);

package/scripts/recover-costs.mjs

@@ -0,0 +1,169 @@
+#!/usr/bin/env node
+/**
+ * Recover cost events from Claude Code transcript files.
+ *
+ * Scans ~/.claude/projects/ for transcript JSONL files,
+ * infers issue ID from directory path, extracts usage data,
+ * and inserts into the panopticon SQLite database.
+ *
+ * Deduplication is handled by the UNIQUE index on request_id.
+ */
+
+import { readdirSync, readFileSync, existsSync, statSync } from 'fs';
+import { join, basename } from 'path';
+import { homedir } from 'os';
+import Database from 'better-sqlite3';
+
+const CLAUDE_PROJECTS = join(homedir(), '.claude', 'projects');
+const DB_PATH = join(homedir(), '.panopticon', 'panopticon.db');
+
+// Pricing table (same as record-cost-event.js)
+const PRICING = [
+  { provider: 'anthropic', model: 'claude-opus-4.6', inputPer1k: 5e-3, outputPer1k: 0.025, cacheReadPer1k: 5e-4, cacheWrite5mPer1k: 625e-5, cacheWrite1hPer1k: 0.01 },
+  { provider: 'anthropic', model: 'claude-opus-4-6', inputPer1k: 5e-3, outputPer1k: 0.025, cacheReadPer1k: 5e-4, cacheWrite5mPer1k: 625e-5, cacheWrite1hPer1k: 0.01 },
+  { provider: 'anthropic', model: 'claude-opus-4-1', inputPer1k: 0.015, outputPer1k: 0.075, cacheReadPer1k: 15e-4, cacheWrite5mPer1k: 0.01875, cacheWrite1hPer1k: 0.03 },
+  { provider: 'anthropic', model: 'claude-opus-4', inputPer1k: 0.015, outputPer1k: 0.075, cacheReadPer1k: 15e-4, cacheWrite5mPer1k: 0.01875, cacheWrite1hPer1k: 0.03 },
+  { provider: 'anthropic', model: 'claude-sonnet-4.5', inputPer1k: 3e-3, outputPer1k: 0.015, cacheReadPer1k: 3e-4, cacheWrite5mPer1k: 375e-5, cacheWrite1hPer1k: 6e-3 },
+  { provider: 'anthropic', model: 'claude-sonnet-4-6', inputPer1k: 3e-3, outputPer1k: 0.015, cacheReadPer1k: 3e-4, cacheWrite5mPer1k: 375e-5, cacheWrite1hPer1k: 6e-3 },
+  { provider: 'anthropic', model: 'claude-sonnet-4', inputPer1k: 3e-3, outputPer1k: 0.015, cacheReadPer1k: 3e-4, cacheWrite5mPer1k: 375e-5, cacheWrite1hPer1k: 6e-3 },
+  { provider: 'anthropic', model: 'claude-haiku-4.5', inputPer1k: 1e-3, outputPer1k: 5e-3, cacheReadPer1k: 1e-4, cacheWrite5mPer1k: 125e-5, cacheWrite1hPer1k: 2e-3 },
+  { provider: 'anthropic', model: 'claude-haiku-4', inputPer1k: 1e-3, outputPer1k: 5e-3, cacheReadPer1k: 1e-4, cacheWrite5mPer1k: 125e-5, cacheWrite1hPer1k: 2e-3 },
+  { provider: 'anthropic', model: 'claude-haiku-3', inputPer1k: 25e-5, outputPer1k: 125e-5, cacheReadPer1k: 3e-5, cacheWrite5mPer1k: 3e-4, cacheWrite1hPer1k: 5e-4 },
+  { provider: 'custom', model: 'kimi-k2.5', inputPer1k: 6e-4, outputPer1k: 2e-3, cacheReadPer1k: 6e-5, cacheWrite5mPer1k: 75e-5 },
+  { provider: 'custom', model: 'kimi-for-coding', inputPer1k: 6e-4, outputPer1k: 2e-3, cacheReadPer1k: 6e-5, cacheWrite5mPer1k: 75e-5 },
+];
+
+function getPricing(model) {
+  return PRICING.find(p => model.startsWith(p.model)) || PRICING.find(p => model.includes(p.model)) || null;
+}
+
+function calculateCost(usage, pricing) {
+  let cost = 0;
+  let inputMul = 1, outputMul = 1;
+  const totalInput = usage.input + (usage.cacheRead || 0) + (usage.cacheWrite || 0);
+  if ((pricing.model.includes('sonnet-4')) && totalInput > 200000) {
+    inputMul = 2; outputMul = 1.5;
+  }
+  cost += usage.input / 1000 * pricing.inputPer1k * inputMul;
+  cost += usage.output / 1000 * pricing.outputPer1k * outputMul;
+  if (usage.cacheRead && pricing.cacheReadPer1k) cost += usage.cacheRead / 1000 * pricing.cacheReadPer1k;
+  if (usage.cacheWrite && pricing.cacheWrite5mPer1k) cost += usage.cacheWrite / 1000 * pricing.cacheWrite5mPer1k;
+  return Math.round(cost * 1e6) / 1e6;
+}
+
+function inferIssueId(dirName) {
+  const match = dirName.match(/(pan|min|aud|krux|cli)[-](\d+)/i);
+  if (match) return `${match[1].toUpperCase()}-${match[2]}`;
+  return null;
+}
+
+function findTranscripts(projectDir) {
+  const transcripts = [];
+  try {
+    const entries = readdirSync(projectDir, { recursive: true });
+    for (const entry of entries) {
+      if (entry.endsWith('.jsonl')) {
+        const full = join(projectDir, entry);
+        try { if (statSync(full).isFile()) transcripts.push(full); } catch {}
+      }
+    }
+  } catch {}
+  return transcripts;
+}
+
+// Main
+const db = new Database(DB_PATH);
+db.pragma('journal_mode = WAL');
+
+const insert = db.prepare(`
+  INSERT OR IGNORE INTO cost_events (
+    ts, agent_id, issue_id, session_type, provider, model,
+    input, output, cache_read, cache_write, cost, request_id, source_file
+  ) VALUES (?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?)
+`);
+
+let totalInserted = 0;
+let totalDuplicates = 0;
+let totalErrors = 0;
+const issueStats = {};
+
+const projectDirs = readdirSync(CLAUDE_PROJECTS);
+for (const dirName of projectDirs) {
+  const issueId = inferIssueId(dirName);
+  if (!issueId) continue;
+
+  const projectDir = join(CLAUDE_PROJECTS, dirName);
+  if (!statSync(projectDir).isDirectory()) continue;
+
+  const transcripts = findTranscripts(projectDir);
+  if (transcripts.length === 0) continue;
+
+  for (const transcript of transcripts) {
+    let content;
+    try { content = readFileSync(transcript, 'utf-8'); } catch { continue; }
+
+    const lines = content.split('\n');
+    for (const line of lines) {
+      if (!line.trim()) continue;
+      try {
+        const entry = JSON.parse(line);
+        if (entry.type !== 'assistant' || !entry.message?.usage) continue;
+
+        const usage = entry.message.usage;
+        const model = entry.message.model || 'claude-sonnet-4';
+        const requestId = entry.requestId;
+        if (!requestId) continue;
+
+        const input = usage.input_tokens || 0;
+        const output = usage.output_tokens || 0;
+        const cacheRead = usage.cache_read_input_tokens || 0;
+        const cacheWrite = usage.cache_creation_input_tokens || 0;
+        if (input === 0 && output === 0 && cacheRead === 0 && cacheWrite === 0) continue;
+
+        let provider = 'anthropic';
+        if (model.includes('gpt')) provider = 'openai';
+        else if (model.includes('gemini')) provider = 'google';
+        else if (model.includes('kimi')) provider = 'custom';
+
+        const pricing = getPricing(model);
+        if (!pricing) continue;
+
+        const cost = calculateCost({ input, output, cacheRead, cacheWrite }, pricing);
+
+        // Use timestamp from the entry if available, otherwise from transcript modification time
+        const ts = entry.timestamp || new Date(statSync(transcript).mtime).toISOString();
+
+        const result = insert.run(
+          ts, 'recovered', issueId, 'interactive', provider, model,
+          input, output, cacheRead, cacheWrite, cost, requestId, basename(transcript)
+        );
+
+        if (result.changes > 0) {
+          totalInserted++;
+          if (!issueStats[issueId]) issueStats[issueId] = { inserted: 0, cost: 0 };
+          issueStats[issueId].inserted++;
+          issueStats[issueId].cost += cost;
+        } else {
+          totalDuplicates++;
+        }
+      } catch {
+        totalErrors++;
+      }
+    }
+  }
+}
+
+db.close();
+
+// Report
+console.log(`\nCost Recovery Complete`);
+console.log(`  Inserted: ${totalInserted} new events`);
+console.log(`  Duplicates skipped: ${totalDuplicates}`);
+console.log(`  Errors: ${totalErrors}`);
+console.log(`\nRecovered costs by issue:`);
+const sorted = Object.entries(issueStats).sort((a, b) => b[1].cost - a[1].cost);
+for (const [id, stats] of sorted) {
+  console.log(`  ${id.padEnd(12)} ${stats.inserted} events  $${stats.cost.toFixed(2)}`);
+}
+const totalCost = sorted.reduce((sum, [, s]) => sum + s.cost, 0);
+console.log(`\n  TOTAL RECOVERED: $${totalCost.toFixed(2)}`);