xinmi/scripts/wiki_leaf_diagnostic_continuation.test.cjs

'use strict'

const { test } = require('node:test')
const assert = require('node:assert/strict')
const {
  VERSION, checkpointHash, runDiagnosticContinuation, sha256,
} = require('./wiki_leaf_diagnostic_continuation.cjs')

function fixture() {
  const source = 'alpha\n'.repeat(20) + 'TARGET-WINDOW\n' + 'omega\n'.repeat(20)
  const targetText = 'TARGET-WINDOW\n'
  const start = source.indexOf(targetText)
  const end = start + Array.from(targetText).length
  const windows = [
    { key: checkpointHash(['wiki-full-read-v1', 'alpha\n'.repeat(20)]), start_rune: 0, end_rune: start,
      text_sha256: sha256('alpha\n'.repeat(20)), status: 'checkpointed', result: { entities: [] } },
    { key: checkpointHash(['wiki-full-read-v1', targetText]), start_rune: start, end_rune: end,
      text_sha256: sha256(targetText), status: 'missing' },
    { key: checkpointHash(['wiki-full-read-v1', 'omega\n'.repeat(20)]), start_rune: end, end_rune: Array.from(source).length,
      text_sha256: sha256('omega\n'.repeat(20)), status: 'checkpointed', result: { concepts: [] } },
  ]
  const manifest = {
    version: VERSION,
    max_model_calls: 1,
    scope: {
      tenant_id: 10059, knowledge_base_id: 'kb', knowledge_id: 'doc', model_id: 'model',
      checkpoint_fingerprint: 'a'.repeat(64), source_sha256: sha256(source),
      parent_attempt: 6, parent_pending_op_id: 4145, diagnostic_pending_op_id: 9001,
      reviewed_release: 'public-reviewed', reviewed_api_image: 'sha256:reviewed',
      target_window: { key: windows[1].key, start_rune: start, end_rune: end, text_sha256: windows[1].text_sha256 },
    },
    windows,
  }
  return { source, manifest }
}

test('reuses completed siblings and captures exactly one identity failure', () => {
  const { source, manifest } = fixture()
  let calls = 0
  let capture
  const result = runDiagnosticContinuation(manifest, source, (text, request) => {
    calls++
    assert.equal(text, 'TARGET-WINDOW\n')
    assert.equal(request.max_model_calls, 1)
    return { status: 'identity_failure', response: '{bad candidate}' }
  }, artifact => { capture = artifact })
  assert.equal(calls, 1)
  assert.equal(result.receipt.status, 'captured')
  assert.equal(capture.response_sha256, sha256('{bad candidate}'))
  assert.equal(capture.response, undefined)
  assert.deepEqual(result.receipt.completed_sibling_keys, [manifest.windows[0].key, manifest.windows[2].key])
  assert.deepEqual(result.windows, manifest.windows)
})

test('saves only a validated target result and never runs a second call', () => {
  const { source, manifest } = fixture()
  let calls = 0
  const result = runDiagnosticContinuation(manifest, source, () => {
    calls++
    return { status: 'leaf_succeeded', result: { entities: [{ name: 'Target', slug: 'entity/target' }] } }
  })
  assert.equal(calls, 1)
  assert.equal(result.receipt.status, 'leaf_succeeded')
  assert.equal(result.receipt.target_checkpoint_saved, true)
  assert.equal(result.windows.filter(window => window.status === 'checkpointed').length, 3)
  assert.deepEqual(result.windows[1].result.entities[0], { name: 'Target', slug: 'entity/target' })
})

test('rejects an already checkpointed target and source/scope drift before model call', () => {
  const { source, manifest } = fixture()
  let calls = 0
  const completedTarget = structuredClone(manifest)
  completedTarget.windows[1] = { ...completedTarget.windows[1], status: 'checkpointed', result: {} }
  assert.throws(() => runDiagnosticContinuation(completedTarget, source, () => { calls++ }), /exactly one missing target/)
  assert.equal(calls, 0)
  assert.throws(() => runDiagnosticContinuation(manifest, `${source}drift`, () => { calls++ }), /source SHA-256/)
  assert.equal(calls, 0)
  const changedScope = structuredClone(manifest)
  changedScope.scope.target_window.end_rune++
  assert.throws(() => runDiagnosticContinuation(changedScope, source, () => { calls++ }), /target window does not match/)
  assert.equal(calls, 0)
})

test('rejects a second missing leaf, a wrong key, and a budget larger than one', () => {
  const { source, manifest } = fixture()
  const secondMissing = structuredClone(manifest)
  secondMissing.windows[2] = { ...secondMissing.windows[2], status: 'missing' }
  assert.throws(() => runDiagnosticContinuation(secondMissing, source, () => ({})), /exactly one missing target/)
  const wrongKey = structuredClone(manifest)
  wrongKey.windows[1].key = 'b'.repeat(64)
  wrongKey.scope.target_window.key = wrongKey.windows[1].key
  assert.throws(() => runDiagnosticContinuation(wrongKey, source, () => ({})), /target key does not match/)
  const overBudget = structuredClone(manifest)
  overBudget.max_model_calls = 2
  assert.throws(() => runDiagnosticContinuation(overBudget, source, () => ({})), /exactly one model call/)
})