dsh-llm-verifier 0.1.13 → 0.1.16

package/src/session.test.ts

@@ -1,36 +1,71 @@
-import { describe, expect, it } from 'vitest'
-import { Session } from '@deepseek-ai/dsh-session'
-import { createUserMessage, createAssistantMessage, createToolResultMessage } from '@deepseek-ai/dsh-llm'
-import { extractSession } from './session.ts'
-
-describe('current session extraction', () => {
-  it('keeps direct evidence, skips plugin instructions, and redacts secrets', async () => {
-    const session = Session.create('session-00000000-0000-4000-8000-000000000001' as never)
-    session.append('user/message', createUserMessage({ content: [{ type: 'text', text: 'Fix task token = abc123 Bearer live-secret' }], source: { kind: 'user' } }), { surfaceOp: 'append' })
-    session.append('user/message', createUserMessage({ content: [{ type: 'text', text: 'hidden plugin instruction' }], source: { kind: 'plugin', plugin: 'test' } }), { surfaceOp: 'append' })
-    session.append('assistant/message', { turn: 1, step: 1, message: createAssistantMessage({ content: [{ type: 'text', text: 'running checks' }], source: { provider: 'deepseek-official', model: 'deepseek-v4-flash' } }) }, { surfaceOp: 'append' })
-    session.append('tool/call', { turn: 1, step: 1, callId: 'call-1' as never, name: 'pwsh', arguments: '{"command":"test"}' })
-    session.append('tool/result', { turn: 1, step: 1, message: createToolResultMessage({ callId: 'call-1' as never, content: [{ type: 'text', text: 'exit 0 password=hunter2' }], isError: false }) }, { surfaceOp: 'append' })
-    const agent = { id: session.id, session } as never
-    const result = await extractSession(agent, async () => { throw new Error('no image expected') })
-    expect(result.problem).toContain('Fix task')
-    expect(result.problem).toContain('[REDACTED]')
-    expect(result.problem).not.toContain('abc123')
-    expect(result.problem).not.toContain('live-secret')
-    expect(result.trace).not.toContain('hidden plugin instruction')
-    expect(result.trace).toContain('exit 0')
-    expect(result.trace).not.toContain('hunter2')
-  })
-
-  it('extracts PTC mode tool/code-dispatch events and redacts content', async () => {
-    const session = Session.create('session-00000000-0000-4000-8000-000000000002' as never)
-    session.append('user/message', createUserMessage({ content: [{ type: 'text', text: 'Run PTC task' }], source: { kind: 'user' } }), { surfaceOp: 'append' })
-    session.append('tool/code-dispatch', { subCallId: 'code-1' as never, name: 'edit', arguments: { file: 'src/main.ts' }, isError: false, content: [{ type: 'text', text: 'updated file api_key=secret-value' }] })
-    const agent = { id: session.id, session } as never
-    const result = await extractSession(agent, async () => { throw new Error('no image expected') })
-    expect(result.trace).toContain('--- Code Dispatch edit')
-    expect(result.trace).toContain('src/main.ts')
-    expect(result.trace).toContain('[REDACTED]')
-    expect(result.trace).not.toContain('secret-value')
-  })
-})
+import { describe, expect, it } from 'vitest'
+import { Session } from '@deepseek-ai/dsh-session'
+import { createUserMessage, createAssistantMessage, createToolResultMessage } from '@deepseek-ai/dsh-llm'
+import { extractSession, sessionEvents } from './session.ts'
+
+describe('current session extraction', () => {
+  it('keeps direct evidence, skips plugin instructions, and redacts secrets', async () => {
+    const session = Session.create('session-00000000-0000-4000-8000-000000000001' as never)
+    session.append('user/message', createUserMessage({ content: [{ type: 'text', text: 'Fix task token = abc123 Bearer live-secret' }], source: { kind: 'user' } }), { surfaceOp: 'append' })
+    session.append('user/message', createUserMessage({ content: [{ type: 'text', text: 'hidden plugin instruction' }], source: { kind: 'plugin', plugin: 'test' } }), { surfaceOp: 'append' })
+    session.append('assistant/message', { turn: 1, step: 1, message: createAssistantMessage({ content: [{ type: 'text', text: 'running checks' }], source: { provider: 'deepseek-official', model: 'deepseek-v4-flash' } }) }, { surfaceOp: 'append' })
+    session.append('tool/call', { turn: 1, step: 1, callId: 'call-1' as never, name: 'pwsh', arguments: '{"command":"test"}' })
+    session.append('tool/result', { turn: 1, step: 1, message: createToolResultMessage({ callId: 'call-1' as never, content: [{ type: 'text', text: 'exit 0 password=hunter2' }], isError: false }) }, { surfaceOp: 'append' })
+    const agent = { id: session.id, session } as never
+    const result = await extractSession(agent, async () => { throw new Error('no image expected') })
+    expect(result.problem).toContain('Fix task')
+    expect(result.problem).toContain('[REDACTED]')
+    expect(result.problem).not.toContain('abc123')
+    expect(result.problem).not.toContain('live-secret')
+    expect(result.trace).not.toContain('hidden plugin instruction')
+    expect(result.trace).toContain('exit 0')
+    expect(result.trace).not.toContain('hunter2')
+  })
+
+  it('extracts PTC mode tool/code-dispatch events and redacts content', async () => {
+    const session = Session.create('session-00000000-0000-4000-8000-000000000002' as never)
+    session.append('user/message', createUserMessage({ content: [{ type: 'text', text: 'Run PTC task' }], source: { kind: 'user' } }), { surfaceOp: 'append' })
+    session.append('tool/code-dispatch', { subCallId: 'code-1' as never, name: 'edit', arguments: { file: 'src/main.ts' }, isError: false, content: [{ type: 'text', text: 'updated file api_key=secret-value' }] })
+    const agent = { id: session.id, session } as never
+    const result = await extractSession(agent, async () => { throw new Error('no image expected') })
+    expect(result.trace).toContain('--- Code Dispatch edit')
+    expect(result.trace).toContain('src/main.ts')
+    expect(result.trace).toContain('[REDACTED]')
+    expect(result.trace).not.toContain('secret-value')
+  })
+
+  it('extracts Session V3 tool/ptc-dispatch, file content blocks, and team messages', async () => {
+    const session = Session.create('session-00000000-0000-4000-8000-000000000003' as never)
+    session.append('user/message', createUserMessage({ content: [{ type: 'text', text: 'Team task instructions' }, { type: 'file' as never, path: 'src/config.ts' } as never], source: { kind: 'user' } }), { surfaceOp: 'append' })
+    session.append('tool/ptc-dispatch' as never, { subCallId: 'ptc-1', name: 'run_test', arguments: { target: 'unit' }, isError: false, content: [{ type: 'text', text: 'all tests passed' }] } as never)
+    session.append('team/message/queued' as never, { message: { senderName: 'Alice', content: [{ type: 'text', text: 'Reviewed PR #123' }] } } as never)
+    session.append('user/message', createUserMessage({ content: [{ type: 'text', text: 'Follow-up from Bob' }], source: { kind: 'team-message' as never } as never }), { surfaceOp: 'append' })
+    const agent = { id: session.id, session } as never
+    const result = await extractSession(agent, async () => { throw new Error('no image expected') })
+    expect(result.trace).toContain('[File] src/config.ts')
+    expect(result.trace).toContain('--- PTC Dispatch run_test')
+    expect(result.trace).toContain('all tests passed')
+    expect(result.trace).toContain('--- Team Message Queued from Alice')
+    expect(result.trace).toContain('Reviewed PR #123')
+    expect(result.trace).toContain('--- Team Message seq')
+    expect(result.trace).toContain('Follow-up from Bob')
+  })
+})
+
+describe('session event accessor', () => {
+  it('reads the 0.1.5 snapshotEvents() shape', () => {
+    const events = [{ seq: 0, type: 'user/message' }]
+    expect(sessionEvents({ snapshotEvents: () => events as never })).toBe(events)
+  })
+
+  it('falls back to the legacy events array', () => {
+    const events = [{ seq: 0, type: 'user/message' }]
+    expect(sessionEvents({ events: events as never })).toBe(events)
+  })
+
+  it('returns an empty log instead of throwing on an unknown session shape', () => {
+    expect(sessionEvents({})).toEqual([])
+    expect(sessionEvents(undefined)).toEqual([])
+    expect(sessionEvents({ events: undefined })).toEqual([])
+  })
+})

package/src/session.ts

@@ -1,103 +1,132 @@
-import type { ContentBlock } from '@deepseek-ai/dsh-llm'
-import type { SessionEvent } from '@deepseek-ai/dsh-session'
-import type { Agent } from '@deepseek-ai/dsh-agent'
-import type { VerifierImage } from './caller.ts'
-
-export interface SessionExtractOptions {
-  fromSeq?: number
-  toSeq?: number
-  includeAssistantText?: boolean
-  redactPatterns?: readonly string[]
-  maxChars?: number
-}
-
-export interface SessionExtraction {
-  problem: string
-  trace: string
-  images: VerifierImage[]
-  sessionId: string
-  fromSeq: number
-  toSeq: number
-  omittedCharacters: number
-}
-
-function textOf(blocks: readonly ContentBlock[]): string {
-  const parts: string[] = []
-  for (const block of blocks) {
-    if (block.type === 'text') parts.push(block.text)
-    else if (block.type === 'reasoning') parts.push('[Reasoning] ' + block.text)
-    else if (block.type === 'tool-call') parts.push('[Tool Call] ' + block.name + ' ' + block.arguments)
-    else if (block.type === 'tool-result') parts.push('[Tool Result] ' + textOf(block.content))
-  }
-  return parts.join('\n')
-}
-
-export const DEFAULT_REDACT_PATTERNS = [
-  'Bearer\\s+[A-Za-z0-9._~+\\/=-]+',
-  "(?:api[_-]?key|token|password|secret)\\s*[=:]\\s*[\"']?[^\\s,\"';}]+",
-] as const
-
-function validateRedactPattern(pattern: string): void {
-  if (pattern.length === 0 || pattern.length > 500) throw new Error('llm-verifier: redact patterns must contain 1-500 characters')
-  // Reject common nested/unbounded constructs that can cause catastrophic backtracking.
-  if (/\([^)]*[+*][^)]*\)[+*{]|\.\*[+*{]|\.\+[+*{]/u.test(pattern)) throw new Error('llm-verifier: unsafe redact pattern')
-}
-
-export function redactText(text: string, patterns: readonly string[] = DEFAULT_REDACT_PATTERNS): string {
-  let result = text
-  for (const pattern of patterns) {
-    validateRedactPattern(pattern)
-    let regex: RegExp
-    try { regex = new RegExp(pattern, 'giu') } catch { throw new Error('llm-verifier: invalid redact pattern: ' + pattern) }
-    result = result.replace(regex, '[REDACTED]')
-  }
-  return result
-}
-
-export function sanitizeVerifierText(text: string, maxChars: number, patterns: readonly string[] = DEFAULT_REDACT_PATTERNS): string {
-  if (!Number.isSafeInteger(maxChars) || maxChars < 1) throw new Error('llm-verifier: sanitizer maxChars must be a positive integer')
-  const redacted = redactText(text, patterns).trim()
-  return redacted.length <= maxChars ? redacted : redacted.slice(0, maxChars) + '\n[Truncated ' + (redacted.length - maxChars) + ' characters]'
-}
-
-export async function extractSession(agent: Agent, loadImage: (ref: Extract<ContentBlock, { type: 'image' }>['attachment']) => Promise<VerifierImage>, options: SessionExtractOptions = {}): Promise<SessionExtraction> {
-  const all = agent.session.events as readonly SessionEvent[]
-  const from = options.fromSeq ?? 0
-  const to = options.toSeq ?? Number.MAX_SAFE_INTEGER
-  const events = all.filter(event => event.seq >= from && event.seq <= to)
-  const patterns = [...DEFAULT_REDACT_PATTERNS, ...(options.redactPatterns ?? [])]
-  let problem = ''
-  const trace: string[] = []
-  const images: VerifierImage[] = []
-  for (const event of events) {
-    if (event.type === 'user/message') {
-      if (event.data.source.kind !== 'user') continue
-      const text = textOf(event.data.content)
-      if (!problem && text.trim()) problem = text.trim()
-      for (const block of event.data.content) if (block.type === 'image') images.push(await loadImage(block.attachment))
-      trace.push('--- User seq ' + event.seq + ' ---\n' + text)
-    } else if (event.type === 'assistant/message' && options.includeAssistantText !== false) {
-      trace.push('--- Assistant turn ' + event.data.turn + ' step ' + event.data.step + ' ---\n' + textOf(event.data.message.content))
-    } else if (event.type === 'tool/call') {
-      trace.push('--- Tool Call turn ' + event.data.turn + ' step ' + event.data.step + ' ---\n[Command] ' + event.data.name + ' ' + event.data.arguments)
-    } else if (event.type === 'tool/result') {
-      trace.push('--- Tool Result turn ' + event.data.turn + ' step ' + event.data.step + ' ---\n[Output] ' + textOf(event.data.message.content))
-    } else if (event.type === 'tool/code-dispatch') {
-      const data = event.data as { name: string; arguments?: unknown; isError?: boolean; content?: readonly ContentBlock[] }
-      const status = data.isError ? ' [Error]' : ''
-      const args = data.arguments !== undefined ? ' ' + (typeof data.arguments === 'string' ? data.arguments : JSON.stringify(data.arguments)) : ''
-      const content = Array.isArray(data.content) ? textOf(data.content) : ''
-      if (Array.isArray(data.content)) {
-        for (const block of data.content) {
-          if (block.type === 'image') images.push(await loadImage(block.attachment))
-        }
-      }
-      trace.push('--- Code Dispatch ' + data.name + status + ' seq ' + event.seq + ' ---\n[Command] ' + data.name + args + '\n[Output] ' + content)
-    }
-  }
-  const raw = redactText(trace.join('\n\n'), patterns)
-  const maxChars = options.maxChars ?? 200000
-  const omittedCharacters = Math.max(0, raw.length - maxChars)
-  const bounded = omittedCharacters ? '[Earlier trace truncated: ' + omittedCharacters + ' characters omitted]\n' + raw.slice(-maxChars) : raw
-  return { problem: redactText(problem, patterns), trace: bounded, images, sessionId: String(agent.id), fromSeq: events[0]?.seq ?? from, toSeq: events.at(-1)?.seq ?? from, omittedCharacters }
-}
+import type { ContentBlock } from '@deepseek-ai/dsh-llm'
+import type { SessionEvent } from '@deepseek-ai/dsh-session'
+import type { Agent } from '@deepseek-ai/dsh-agent'
+import type { VerifierImage } from './caller.ts'
+
+/**
+ * Read one session's event log.
+ *
+ * DSH 0.1.5 replaced the `events` array with `snapshotEvents()`; older hosts
+ * expose the array directly. Both shapes are accepted so the plugin keeps
+ * working across the versions it declares support for.
+ * @param session - Agent session, or any object exposing one of the two shapes.
+ * @returns The session's events in log order, or an empty array.
+ */
+export function sessionEvents(session: unknown): readonly SessionEvent[] {
+  const candidate = session as { snapshotEvents?: () => readonly SessionEvent[]; events?: readonly SessionEvent[] } | undefined
+  if (typeof candidate?.snapshotEvents === 'function') return candidate.snapshotEvents()
+  return candidate?.events ?? []
+}
+
+export interface SessionExtractOptions {
+  fromSeq?: number
+  toSeq?: number
+  includeAssistantText?: boolean
+  redactPatterns?: readonly string[]
+  maxChars?: number
+}
+
+export interface SessionExtraction {
+  problem: string
+  trace: string
+  images: VerifierImage[]
+  sessionId: string
+  fromSeq: number
+  toSeq: number
+  omittedCharacters: number
+}
+
+function textOf(blocks: readonly ContentBlock[]): string {
+  const parts: string[] = []
+  for (const block of blocks) {
+    const blockType = (block as { type: string }).type
+    if (blockType === 'text') parts.push((block as Extract<ContentBlock, { type: 'text' }>).text)
+    else if (blockType === 'reasoning') parts.push('[Reasoning] ' + (block as Extract<ContentBlock, { type: 'reasoning' }>).text)
+    else if (blockType === 'tool-call') parts.push('[Tool Call] ' + (block as Extract<ContentBlock, { type: 'tool-call' }>).name + ' ' + (block as Extract<ContentBlock, { type: 'tool-call' }>).arguments)
+    else if (blockType === 'tool-result') parts.push('[Tool Result] ' + textOf((block as Extract<ContentBlock, { type: 'tool-result' }>).content))
+    else if (blockType === 'file') {
+      const fileData = block as unknown as { path?: string; filename?: string; title?: string }
+      parts.push('[File] ' + (fileData.path ?? fileData.filename ?? fileData.title ?? 'attachment'))
+    }
+  }
+  return parts.join('\n')
+}
+
+export const DEFAULT_REDACT_PATTERNS = [
+  'Bearer\\s+[A-Za-z0-9._~+\\/=-]+',
+  "(?:api[_-]?key|token|password|secret)\\s*[=:]\\s*[\"']?[^\\s,\"';}]+",
+] as const
+
+function validateRedactPattern(pattern: string): void {
+  if (pattern.length === 0 || pattern.length > 500) throw new Error('llm-verifier: redact patterns must contain 1-500 characters')
+  // Reject common nested/unbounded constructs that can cause catastrophic backtracking.
+  if (/\([^)]*[+*][^)]*\)[+*{]|\.\*[+*{]|\.\+[+*{]/u.test(pattern)) throw new Error('llm-verifier: unsafe redact pattern')
+}
+
+export function redactText(text: string, patterns: readonly string[] = DEFAULT_REDACT_PATTERNS): string {
+  let result = text
+  for (const pattern of patterns) {
+    validateRedactPattern(pattern)
+    let regex: RegExp
+    try { regex = new RegExp(pattern, 'giu') } catch { throw new Error('llm-verifier: invalid redact pattern: ' + pattern) }
+    result = result.replace(regex, '[REDACTED]')
+  }
+  return result
+}
+
+export function sanitizeVerifierText(text: string, maxChars: number, patterns: readonly string[] = DEFAULT_REDACT_PATTERNS): string {
+  if (!Number.isSafeInteger(maxChars) || maxChars < 1) throw new Error('llm-verifier: sanitizer maxChars must be a positive integer')
+  const redacted = redactText(text, patterns).trim()
+  return redacted.length <= maxChars ? redacted : redacted.slice(0, maxChars) + '\n[Truncated ' + (redacted.length - maxChars) + ' characters]'
+}
+
+export async function extractSession(agent: Agent, loadImage: (ref: Extract<ContentBlock, { type: 'image' }>['attachment']) => Promise<VerifierImage>, options: SessionExtractOptions = {}): Promise<SessionExtraction> {
+  const all = sessionEvents(agent.session)
+  const from = options.fromSeq ?? 0
+  const to = options.toSeq ?? Number.MAX_SAFE_INTEGER
+  const events = all.filter(event => event.seq >= from && event.seq <= to)
+  const patterns = [...DEFAULT_REDACT_PATTERNS, ...(options.redactPatterns ?? [])]
+  let problem = ''
+  const trace: string[] = []
+  const images: VerifierImage[] = []
+  for (const rawEvent of events) {
+    const event = rawEvent as unknown as { type: string; seq: number; data: any }
+    if (event.type === 'user/message') {
+      const sourceKind = event.data?.source?.kind as string
+      if (sourceKind !== 'user' && sourceKind !== 'team-message') continue
+      const text = textOf(event.data.content)
+      if (!problem && sourceKind === 'user' && text.trim()) problem = text.trim()
+      for (const block of event.data.content) if (block.type === 'image') images.push(await loadImage(block.attachment))
+      const tag = sourceKind === 'team-message' ? 'Team Message' : 'User'
+      trace.push('--- ' + tag + ' seq ' + event.seq + ' ---\n' + text)
+    } else if (event.type === 'assistant/message' && options.includeAssistantText !== false) {
+      trace.push('--- Assistant turn ' + event.data.turn + ' step ' + event.data.step + ' ---\n' + textOf(event.data.message.content))
+    } else if (event.type === 'tool/call') {
+      trace.push('--- Tool Call turn ' + event.data.turn + ' step ' + event.data.step + ' ---\n[Command] ' + event.data.name + ' ' + event.data.arguments)
+    } else if (event.type === 'tool/result') {
+      trace.push('--- Tool Result turn ' + event.data.turn + ' step ' + event.data.step + ' ---\n[Output] ' + textOf(event.data.message.content))
+    } else if (event.type === 'tool/ptc-dispatch' || event.type === 'tool/code-dispatch') {
+      const data = event.data as { name: string; arguments?: unknown; isError?: boolean; content?: readonly ContentBlock[] }
+      const status = data.isError ? ' [Error]' : ''
+      const args = data.arguments !== undefined ? ' ' + (typeof data.arguments === 'string' ? data.arguments : JSON.stringify(data.arguments)) : ''
+      const content = Array.isArray(data.content) ? textOf(data.content) : ''
+      if (Array.isArray(data.content)) {
+        for (const block of data.content) {
+          if (block.type === 'image') images.push(await loadImage(block.attachment))
+        }
+      }
+      const label = event.type === 'tool/ptc-dispatch' ? 'PTC Dispatch ' : 'Code Dispatch '
+      trace.push('--- ' + label + data.name + status + ' seq ' + event.seq + ' ---\n[Command] ' + data.name + args + '\n[Output] ' + content)
+    } else if (event.type === 'team/message/queued') {
+      const data = event.data as { message?: { senderName?: string; targetId?: string; content?: readonly ContentBlock[] } }
+      const sender = data?.message?.senderName ?? 'Teammate'
+      const content = Array.isArray(data?.message?.content) ? textOf(data.message!.content) : ''
+      trace.push('--- Team Message Queued from ' + sender + ' seq ' + event.seq + ' ---\n' + content)
+    }
+  }
+  const raw = redactText(trace.join('\n\n'), patterns)
+  const maxChars = options.maxChars ?? 200000
+  const omittedCharacters = Math.max(0, raw.length - maxChars)
+  const bounded = omittedCharacters ? '[Earlier trace truncated: ' + omittedCharacters + ' characters omitted]\n' + raw.slice(-maxChars) : raw
+  return { problem: redactText(problem, patterns), trace: bounded, images, sessionId: String(agent.id), fromSeq: events[0]?.seq ?? from, toSeq: events.at(-1)?.seq ?? from, omittedCharacters }
+}

package/src/statistics.test.ts

@@ -1,80 +1,80 @@
-import { mkdtemp, readFile } from 'node:fs/promises'
-import { join } from 'node:path'
-import { tmpdir } from 'node:os'
-import { describe, expect, it } from 'vitest'
-import { StatisticsStore, emptyRunStats, mergeStatisticsOverviews } from './statistics.ts'
-
-function stats(overrides: Partial<ReturnType<typeof emptyRunStats>> = {}) {
-  return { ...emptyRunStats(), ...overrides }
-}
-
-describe('StatisticsStore', () => {
-  it('records and aggregates invocations by day, tool, model, and session', async () => {
-    const root = await mkdtemp(join(tmpdir(), 'dsh-verifier-statistics-'))
-    const file = join(root, 'statistics.json')
-    const store = new StatisticsStore(file, 100)
-    const first = Date.UTC(2026, 7, 19, 23, 30)
-    const second = Date.UTC(2026, 7, 20, 1, 30)
-    await store.record({ toolName: 'verifier_compare', sessionId: 'one', startedAt: first, finishedAt: first + 250, success: true, provider: 'p', model: 'm', stats: stats({ calls: 2, attempts: 2, inputTokens: 100, cachedInputTokens: 20, outputTokens: 10, cacheHits: 1, cacheMisses: 1, estimatedCostUsd: 0.12, topLogprobScores: 2 }) })
-    await store.record({ toolName: 'verifier_track', sessionId: 'two', startedAt: second, finishedAt: second + 750, success: false, errorName: 'Error', errorMessage: 'failed', provider: 'p', model: 'm2', stats: stats({ calls: 1, attempts: 2, retries: 1, inputTokens: 50, outputTokens: 5, cacheMisses: 1, explicitTagScores: 1 }) })
-
-    const all = await store.overview({ fromMs: first - 1, toMs: second + 10_000, timezoneOffsetMinutes: -120 })
-    expect(all.totals).toMatchObject({ invocations: 2, successes: 1, failures: 1, calls: 3, attempts: 4, retries: 1, tokens: 185, cacheHits: 1, cacheMisses: 2 })
-    expect(all.totals.successRate).toBe(0.5)
-    expect(all.totals.averageDurationMs).toBe(500)
-    expect(all.daily).toHaveLength(1)
-    expect(all.daily[0]).toMatchObject({ date: '2026-08-20', invocations: 2, calls: 3, tokens: 185 })
-    expect(all.tools.map(row => [row.toolName, row.invocations])).toEqual([['verifier_compare', 1], ['verifier_track', 1]])
-    expect(all.models.map(row => row.model)).toEqual(['m', 'm2'])
-
-    const one = await store.overview({ fromMs: first - 1, toMs: second + 10_000, sessionId: 'one' })
-    expect(one.totals.invocations).toBe(1)
-    expect(one.recent[0]?.sessionId).toBe('one')
-    expect(JSON.parse(await readFile(file, 'utf8')).version).toBe(1)
-  })
-
-  it('serializes concurrent writes and keeps only the configured tail', async () => {
-    const root = await mkdtemp(join(tmpdir(), 'dsh-verifier-statistics-'))
-    const store = new StatisticsStore(join(root, 'statistics.json'), 2)
-    await Promise.all([0, 1, 2].map(index => store.record({ toolName: 'verifier_select', startedAt: 1_000 + index, success: true, provider: 'p', model: 'm', stats: stats() })))
-    const result = await store.overview({ fromMs: 0, toMs: 10_000 })
-    expect(result.totals.invocations).toBe(2)
-    expect(result.recent.map(record => record.startedAt)).toEqual([1_002, 1_001])
-  })
-
-  it('merges independently persisted topic summaries', async () => {
-    const root = await mkdtemp(join(tmpdir(), 'dsh-verifier-statistics-'))
-    const first = new StatisticsStore(join(root, 'one.json'))
-    const second = new StatisticsStore(join(root, 'two.json'))
-    await first.record({ toolName: 'verifier_compare', sessionId: 'one', startedAt: 1, finishedAt: 11, success: true, provider: 'p', model: 'm', stats: stats() })
-    await second.record({ toolName: 'verifier_track', sessionId: 'two', startedAt: 2, finishedAt: 22, success: false, provider: 'p', model: 'm', stats: stats() })
-    const query = { fromMs: 0, toMs: 100, recentLimit: 10 }
-    const merged = mergeStatisticsOverviews(await Promise.all([first.overview(query), second.overview(query)]), query)
-    expect(merged.totals.invocations).toBe(2)
-    expect(merged.totals.successes).toBe(1)
-    expect(merged.totals.averageDurationMs).toBe(15)
-    expect(merged.tools.map(row => row.toolName)).toEqual(['verifier_compare', 'verifier_track'])
-    expect(merged.models).toHaveLength(1)
-    expect(merged.recent.map(row => row.sessionId)).toEqual(['two', 'one'])
-  })
-
-  it('bounds persisted error messages', async () => {
-    const root = await mkdtemp(join(tmpdir(), 'dsh-verifier-statistics-'))
-    const store = new StatisticsStore(join(root, 'statistics.json'))
-    await store.record({ toolName: 'verifier_current_session', startedAt: 1, success: false, errorName: 'Failure', errorMessage: 'x'.repeat(1_000), provider: 'p', model: 'm', stats: stats() })
-    const result = await store.overview({ fromMs: 0, toMs: 10 })
-    expect(result.recent[0]?.errorMessage).toHaveLength(500)
-  })
-
-  it('normalizes legacy and modern sessionPersistence.list output formats', () => {
-    const legacyHeaders = [{ id: 'sess-1', cwd: 'C:\\test' }]
-    const modernSnapshots = [{ header: { id: 'sess-2', cwd: 'C:\\test' }, revision: 'rev-1' }]
-    const normalize = (items: readonly unknown[]) =>
-      items
-        .map(item => item && typeof item === 'object' && 'header' in item ? (item as { header: { id: string } }).header : item as { id: string })
-        .filter(header => header !== undefined)
-
-    expect(normalize(legacyHeaders).map(h => h.id)).toEqual(['sess-1'])
-    expect(normalize(modernSnapshots).map(h => h.id)).toEqual(['sess-2'])
-  })
-})
+import { mkdtemp, readFile } from 'node:fs/promises'
+import { join } from 'node:path'
+import { tmpdir } from 'node:os'
+import { describe, expect, it } from 'vitest'
+import { StatisticsStore, emptyRunStats, mergeStatisticsOverviews } from './statistics.ts'
+
+function stats(overrides: Partial<ReturnType<typeof emptyRunStats>> = {}) {
+  return { ...emptyRunStats(), ...overrides }
+}
+
+describe('StatisticsStore', () => {
+  it('records and aggregates invocations by day, tool, model, and session', async () => {
+    const root = await mkdtemp(join(tmpdir(), 'dsh-verifier-statistics-'))
+    const file = join(root, 'statistics.json')
+    const store = new StatisticsStore(file, 100)
+    const first = Date.UTC(2026, 7, 19, 23, 30)
+    const second = Date.UTC(2026, 7, 20, 1, 30)
+    await store.record({ toolName: 'verifier_compare', sessionId: 'one', startedAt: first, finishedAt: first + 250, success: true, provider: 'p', model: 'm', stats: stats({ calls: 2, attempts: 2, inputTokens: 100, cachedInputTokens: 20, outputTokens: 10, cacheHits: 1, cacheMisses: 1, estimatedCostUsd: 0.12, topLogprobScores: 2 }) })
+    await store.record({ toolName: 'verifier_track', sessionId: 'two', startedAt: second, finishedAt: second + 750, success: false, errorName: 'Error', errorMessage: 'failed', provider: 'p', model: 'm2', stats: stats({ calls: 1, attempts: 2, retries: 1, inputTokens: 50, outputTokens: 5, cacheMisses: 1, explicitTagScores: 1 }) })
+
+    const all = await store.overview({ fromMs: first - 1, toMs: second + 10_000, timezoneOffsetMinutes: -120 })
+    expect(all.totals).toMatchObject({ invocations: 2, successes: 1, failures: 1, calls: 3, attempts: 4, retries: 1, tokens: 185, cacheHits: 1, cacheMisses: 2 })
+    expect(all.totals.successRate).toBe(0.5)
+    expect(all.totals.averageDurationMs).toBe(500)
+    expect(all.daily).toHaveLength(1)
+    expect(all.daily[0]).toMatchObject({ date: '2026-08-20', invocations: 2, calls: 3, tokens: 185 })
+    expect(all.tools.map(row => [row.toolName, row.invocations])).toEqual([['verifier_compare', 1], ['verifier_track', 1]])
+    expect(all.models.map(row => row.model)).toEqual(['m', 'm2'])
+
+    const one = await store.overview({ fromMs: first - 1, toMs: second + 10_000, sessionId: 'one' })
+    expect(one.totals.invocations).toBe(1)
+    expect(one.recent[0]?.sessionId).toBe('one')
+    expect(JSON.parse(await readFile(file, 'utf8')).version).toBe(1)
+  })
+
+  it('serializes concurrent writes and keeps only the configured tail', async () => {
+    const root = await mkdtemp(join(tmpdir(), 'dsh-verifier-statistics-'))
+    const store = new StatisticsStore(join(root, 'statistics.json'), 2)
+    await Promise.all([0, 1, 2].map(index => store.record({ toolName: 'verifier_select', startedAt: 1_000 + index, success: true, provider: 'p', model: 'm', stats: stats() })))
+    const result = await store.overview({ fromMs: 0, toMs: 10_000 })
+    expect(result.totals.invocations).toBe(2)
+    expect(result.recent.map(record => record.startedAt)).toEqual([1_002, 1_001])
+  })
+
+  it('merges independently persisted topic summaries', async () => {
+    const root = await mkdtemp(join(tmpdir(), 'dsh-verifier-statistics-'))
+    const first = new StatisticsStore(join(root, 'one.json'))
+    const second = new StatisticsStore(join(root, 'two.json'))
+    await first.record({ toolName: 'verifier_compare', sessionId: 'one', startedAt: 1, finishedAt: 11, success: true, provider: 'p', model: 'm', stats: stats() })
+    await second.record({ toolName: 'verifier_track', sessionId: 'two', startedAt: 2, finishedAt: 22, success: false, provider: 'p', model: 'm', stats: stats() })
+    const query = { fromMs: 0, toMs: 100, recentLimit: 10 }
+    const merged = mergeStatisticsOverviews(await Promise.all([first.overview(query), second.overview(query)]), query)
+    expect(merged.totals.invocations).toBe(2)
+    expect(merged.totals.successes).toBe(1)
+    expect(merged.totals.averageDurationMs).toBe(15)
+    expect(merged.tools.map(row => row.toolName)).toEqual(['verifier_compare', 'verifier_track'])
+    expect(merged.models).toHaveLength(1)
+    expect(merged.recent.map(row => row.sessionId)).toEqual(['two', 'one'])
+  })
+
+  it('bounds persisted error messages', async () => {
+    const root = await mkdtemp(join(tmpdir(), 'dsh-verifier-statistics-'))
+    const store = new StatisticsStore(join(root, 'statistics.json'))
+    await store.record({ toolName: 'verifier_current_session', startedAt: 1, success: false, errorName: 'Failure', errorMessage: 'x'.repeat(1_000), provider: 'p', model: 'm', stats: stats() })
+    const result = await store.overview({ fromMs: 0, toMs: 10 })
+    expect(result.recent[0]?.errorMessage).toHaveLength(500)
+  })
+
+  it('normalizes legacy and modern sessionPersistence.list output formats', () => {
+    const legacyHeaders = [{ id: 'sess-1', cwd: 'C:\\test' }]
+    const modernSnapshots = [{ header: { id: 'sess-2', cwd: 'C:\\test' }, revision: 'rev-1' }]
+    const normalize = (items: readonly unknown[]) =>
+      items
+        .map(item => item && typeof item === 'object' && 'header' in item ? (item as { header: { id: string } }).header : item as { id: string })
+        .filter(header => header !== undefined)
+
+    expect(normalize(legacyHeaders).map(h => h.id)).toEqual(['sess-1'])
+    expect(normalize(modernSnapshots).map(h => h.id)).toEqual(['sess-2'])
+  })
+})