@bendyline/gilde 0.1.39 → 0.1.41

package/authoring/tactical/evals/dependency-upgrade.json

@@ -0,0 +1,235 @@
+{
+  "schemaVersion": 1,
+  "title": "Vendored dependency upgrade with staged receipts",
+  "objective": "The craftbook drives a staged upgrade of a vendored library across a breaking major: a recorded green suite before any change, the v2 API adopted at both real call sites with behavior preserved against the v1 reference, an honest stage record, and an enforced review - all verified mechanically, including a resolved-module check proving the call sites genuinely run v2.",
+  "tags": [
+    "code",
+    "dependency-upgrade",
+    "tactical-fleet"
+  ],
+  "prompt": "Use the Upgrade a Dependency craftbook for this: upgrade our vendored markdownlite library from 1.4.2 to the staged 2.0.0 copy. The vendoring convention and the v2 changelog are in VENDORING.md, and the request is in source/upgrade-request.md. Rendering behavior must not change - adapt the call sites to the new API rather than accepting the new defaults.",
+  "setup": {
+    "projectName": "Relay Notes",
+    "about": "A tiny note-rendering module for a team relay tool. Dependencies are vendored under lib/vendored/ per VENDORING.md. Tests run with `npm run test` (node --test).",
+    "files": [
+      {
+        "path": "package.json",
+        "content": "{\n  \"name\": \"relay-notes\",\n  \"private\": true,\n  \"version\": \"1.0.0\",\n  \"type\": \"module\",\n  \"scripts\": {\n    \"test\": \"node --test\"\n  }\n}\n"
+      },
+      {
+        "path": "lib/vendored/markdownlite.js",
+        "content": "// markdownlite 1.4.2 (vendored - see VENDORING.md)\nexport function parse(text) {\n  const lines = String(text).split(/\\r?\\n/);\n  const out = [];\n  for (const line of lines) {\n    const t = line.trim();\n    if (t === '') continue;\n    const h = /^(#{1,3})\\s+(.*)$/.exec(t);\n    if (h) {\n      out.push(`<h${h[1].length}>${inline(h[2])}</h${h[1].length}>`);\n    } else {\n      out.push(`<p>${inline(t)}</p>`);\n    }\n  }\n  return out.join('\\n');\n}\n\nfunction inline(text) {\n  return text.replace(/\\*([^*]+)\\*/g, '<em>$1</em>');\n}\n"
+      },
+      {
+        "path": "lib/vendored/v2/markdownlite.js",
+        "content": "// markdownlite 2.0.0 (vendored staging copy - see VENDORING.md)\nexport function parseDoc(text, options = {}) {\n  const wrapParagraphs = options.wrapParagraphs ?? false;\n  const lines = String(text).split(/\\r?\\n/);\n  const out = [];\n  for (const line of lines) {\n    const t = line.trim();\n    if (t === '') continue;\n    const h = /^(#{1,3})\\s+(.*)$/.exec(t);\n    if (h) {\n      out.push(`<h${h[1].length}>${inline(h[2])}</h${h[1].length}>`);\n    } else {\n      out.push(wrapParagraphs ? `<p>${inline(t)}</p>` : inline(t));\n    }\n  }\n  return out.join('\\n');\n}\n\nfunction inline(text) {\n  return text.replace(/\\*([^*]+)\\*/g, '<em>$1</em>');\n}\n"
+      },
+      {
+        "path": "src/render-note.js",
+        "content": "import { parse } from '../lib/vendored/markdownlite.js';\n\nexport function renderNote(note) {\n  return `<h2>${note.title}</h2>\\n${parse(note.body)}`;\n}\n"
+      },
+      {
+        "path": "src/render-digest.js",
+        "content": "import { parse } from '../lib/vendored/markdownlite.js';\n\nexport function renderDigest(notes) {\n  return notes\n    .map((note) => `<section>\\n<h2>${note.title}</h2>\\n${parse(note.body)}\\n</section>`)\n    .join('\\n');\n}\n"
+      },
+      {
+        "path": "tests/render-note.test.mjs",
+        "content": "import assert from 'node:assert/strict';\nimport { test } from 'node:test';\nimport { renderNote } from '../src/render-note.js';\n\ntest('renders the body as wrapped paragraphs', () => {\n  assert.equal(\n    renderNote({ title: 'Standup', body: 'we shipped the *fast* path' }),\n    '<h2>Standup</h2>\\n<p>we shipped the <em>fast</em> path</p>',\n  );\n});\n\ntest('renders headings inside the body', () => {\n  assert.equal(\n    renderNote({ title: 'Plan', body: '# Goals\\nship it' }),\n    '<h2>Plan</h2>\\n<h1>Goals</h1>\\n<p>ship it</p>',\n  );\n});\n"
+      },
+      {
+        "path": "tests/render-digest.test.mjs",
+        "content": "import assert from 'node:assert/strict';\nimport { test } from 'node:test';\nimport { renderDigest } from '../src/render-digest.js';\n\ntest('joins every note as a wrapped section', () => {\n  const html = renderDigest([\n    { title: 'One', body: 'first' },\n    { title: 'Two', body: 'second line' },\n  ]);\n  assert.equal(\n    html,\n    '<section>\\n<h2>One</h2>\\n<p>first</p>\\n</section>\\n<section>\\n<h2>Two</h2>\\n<p>second line</p>\\n</section>',\n  );\n});\n"
+      },
+      {
+        "path": "VENDORING.md",
+        "content": "# Vendoring convention\n\nThis project vendors small dependencies directly under `lib/vendored/` instead\nof installing them from a registry. Each vendored library is a single file named\nafter the package. Upgrades arrive as a staged copy in a version subfolder\n(for example `lib/vendored/v2/`) so the switch is an explicit, reviewable\nchange: move the staged copy into place (or point the imports at it), adapt the\ncall sites, and remove the staging folder when the switch is complete.\n\n## markdownlite\n\nCurrent vendored version: 1.4.2 at `lib/vendored/markdownlite.js`.\nStaged upgrade: 2.0.0 at `lib/vendored/v2/markdownlite.js`.\n\n### markdownlite 2.0.0 changelog\n\n- BREAKING: `parse(text)` is renamed to `parseDoc(text, options)`.\n- BREAKING: plain lines are no longer wrapped in `<p>` by default. Paragraph\n  wrapping is now opt-in: pass `{ wrapParagraphs: true }` to keep the 1.x\n  output.\n- Heading and emphasis handling are unchanged.\n"
+      },
+      {
+        "path": "source/upgrade-request.md",
+        "content": "# Upgrade request\n\nPlease move us from vendored markdownlite 1.4.2 to the staged 2.0.0 copy.\nThe v2 changelog is in VENDORING.md. Rendering behavior must not change:\nnotes and digests must produce exactly the same HTML as today, so adapt the\ncall sites to the new API rather than accepting the new defaults.\n"
+      },
+      {
+        "path": "tests/verify-upgrade.mjs",
+        "surface": "harness",
+        "content": "import assert from 'node:assert/strict';\nimport { execFileSync } from 'node:child_process';\nimport { readFileSync } from 'node:fs';\nimport { join } from 'node:path';\nimport { pathToFileURL } from 'node:url';\n\nconst root = process.cwd();\n\n// 1) Both call sites moved off the removed v1 API and resolve to a v2 module.\n//    Accepted either way: v2 copied into place at lib/vendored/markdownlite.js,\n//    or imports repointed at the staged v2 file.\nconst callSites = ['src/render-note.js', 'src/render-digest.js'];\nfor (const file of callSites) {\n  const source = readFileSync(join(root, file), 'utf8');\n  assert.ok(/\\bparseDoc\\s*\\(/.test(source), `${file} must call the v2 parseDoc API`);\n  assert.ok(!/\\bparse\\s*\\(/.test(source), `${file} still calls the removed v1 parse API`);\n  const spec = /import\\s*\\{[^}]*\\}\\s*from\\s*['\"]([^'\"]*markdownlite[^'\"]*)['\"]/.exec(source);\n  assert.ok(spec, `${file} must import the vendored markdownlite module`);\n  const mod = await import(new URL(spec[1], pathToFileURL(join(root, file))).href);\n  assert.equal(typeof mod.parseDoc, 'function', `${file} must resolve to markdownlite v2 (parseDoc export)`);\n  assert.equal(mod.parse, undefined, `${file} resolves to a module that still exports the removed v1 parse API`);\n}\nconsole.log('DEPUPGRADE_ORACLE call sites adapted');\n\n// 2) Behavior is unchanged against the v1 reference output.\nconst { renderNote } = await import(pathToFileURL(join(root, 'src/render-note.js')).href);\nconst { renderDigest } = await import(pathToFileURL(join(root, 'src/render-digest.js')).href);\nassert.equal(\n  renderNote({ title: 'Standup', body: 'we shipped the *fast* path' }),\n  '<h2>Standup</h2>\\n<p>we shipped the <em>fast</em> path</p>',\n  'renderNote must keep wrapping paragraphs after the upgrade',\n);\nassert.equal(\n  renderNote({ title: 'Plan', body: '# Goals\\nship it' }),\n  '<h2>Plan</h2>\\n<h1>Goals</h1>\\n<p>ship it</p>',\n  'renderNote must keep heading handling after the upgrade',\n);\nassert.equal(\n  renderDigest([\n    { title: 'One', body: 'first' },\n    { title: 'Two', body: 'second line' },\n  ]),\n  '<section>\\n<h2>One</h2>\\n<p>first</p>\\n</section>\\n<section>\\n<h2>Two</h2>\\n<p>second line</p>\\n</section>',\n  'renderDigest must keep its exact output after the upgrade',\n);\nconsole.log('DEPUPGRADE_ORACLE behavior ok');\n\n// 3) The suite is green on the upgraded tree, and the seeded tests survive.\nreadFileSync(join(root, 'tests/render-note.test.mjs'));\nreadFileSync(join(root, 'tests/render-digest.test.mjs'));\nfunction runSuite(cwd) {\n  try {\n    execFileSync(process.execPath, ['--test'], { cwd, stdio: 'pipe', timeout: 30000 });\n    return 0;\n  } catch (err) {\n    return typeof err.status === 'number' ? err.status : 1;\n  }\n}\nassert.equal(runSuite(root), 0, 'the suite must be green on the upgraded tree');\nconsole.log('DEPUPGRADE_ORACLE suite green');\nconsole.log('DEPUPGRADE_ORACLE done');\n"
+      }
+    ],
+    "craftbookParams": {
+      "workPath": "tasks/eval",
+      "scope": "Upgrade the vendored markdownlite from 1.4.2 to the staged 2.0.0 copy per VENDORING.md; rendering behavior must not change."
+    }
+  },
+  "success": {
+    "summary": "The vendored library runs at v2 through both call sites (either moved into place or repointed), every adaptation preserves the exact v1 rendering behavior, the suite is green, and every fleet artifact exists with honest receipts.",
+    "deliverables": [
+      {
+        "path": "tasks/eval/survey.md",
+        "kind": "markdown-notes",
+        "artifact": true,
+        "minBytes": 600,
+        "checks": [
+          {
+            "kind": "contains",
+            "file": "tasks/eval/survey.md",
+            "pattern": "^##\\s+Current state[\\s\\S]*^##\\s+Candidates[\\s\\S]*^##\\s+Breaking-change notes[\\s\\S]*^##\\s+Stage plan",
+            "flags": "im",
+            "label": "survey sections in order"
+          },
+          {
+            "kind": "contains",
+            "file": "tasks/eval/survey.md",
+            "pattern": "parseDoc",
+            "label": "names the real v2 API change from the local changelog"
+          },
+          {
+            "kind": "citationsResolve",
+            "file": "tasks/eval/survey.md",
+            "minCitations": 2,
+            "artifact": true
+          }
+        ]
+      },
+      {
+        "path": "tasks/eval/change-notes.md",
+        "kind": "markdown-notes",
+        "artifact": true,
+        "minBytes": 700,
+        "checks": [
+          {
+            "kind": "contains",
+            "file": "tasks/eval/change-notes.md",
+            "pattern": "^##\\s+Stages executed[\\s\\S]*^##\\s+Files touched[\\s\\S]*^##\\s+Adaptations[\\s\\S]*^##\\s+Reverted stages",
+            "flags": "im",
+            "label": "change-notes sections in order"
+          },
+          {
+            "kind": "citationsResolve",
+            "file": "tasks/eval/change-notes.md",
+            "minCitations": 2,
+            "artifact": true
+          }
+        ]
+      },
+      {
+        "path": "tasks/eval/verification.md",
+        "kind": "markdown-notes",
+        "artifact": true,
+        "minBytes": 450,
+        "checks": [
+          {
+            "kind": "contains",
+            "file": "tasks/eval/verification.md",
+            "pattern": "^##\\s+Suite[\\s\\S]*^##\\s+Adaptations verified[\\s\\S]*^##\\s+Rollback[\\s\\S]*^##\\s+Result",
+            "flags": "im",
+            "label": "verification sections in order"
+          }
+        ]
+      },
+      {
+        "path": "tasks/eval/review.md",
+        "kind": "markdown-report",
+        "artifact": true,
+        "minBytes": 400,
+        "checks": [
+          {
+            "kind": "contains",
+            "file": "tasks/eval/review.md",
+            "pattern": "Verdict:\\s*(?:PASS|REVISE)",
+            "flags": "i",
+            "label": "explicit reviewer verdict"
+          }
+        ]
+      }
+    ],
+    "checks": [
+      {
+        "kind": "notContains",
+        "file": "src/render-note.js",
+        "pattern": "\\bparse\\s*\\(|import\\s*\\{[^}]*\\bparse\\b[^}]*\\}",
+        "label": "the removed v1 parse API is gone from the note call site"
+      },
+      {
+        "kind": "notContains",
+        "file": "src/render-digest.js",
+        "pattern": "\\bparse\\s*\\(|import\\s*\\{[^}]*\\bparse\\b[^}]*\\}",
+        "label": "the removed v1 parse API is gone from the digest call site"
+      },
+      {
+        "kind": "sourceParses",
+        "file": "src/render-note.js"
+      },
+      {
+        "kind": "sourceParses",
+        "file": "src/render-digest.js"
+      },
+      {
+        "kind": "nodeRuns",
+        "file": "tests/render-note.test.mjs"
+      },
+      {
+        "kind": "nodeScriptPasses",
+        "script": "tests/verify-upgrade.mjs",
+        "timeoutMs": 60000,
+        "requiredOutput": [
+          {
+            "pattern": "DEPUPGRADE_ORACLE call sites adapted",
+            "label": "both call sites run the v2 API"
+          },
+          {
+            "pattern": "DEPUPGRADE_ORACLE behavior ok",
+            "label": "rendering behavior preserved against the v1 reference"
+          },
+          {
+            "pattern": "DEPUPGRADE_ORACLE suite green",
+            "label": "suite green on the upgraded tree"
+          }
+        ]
+      }
+    ],
+    "taskNotes": {
+      "minBytes": 120,
+      "checks": [
+        {
+          "kind": "contains",
+          "file": "task-notes.md",
+          "pattern": "\\bDONE\\b[\\s\\S]*(npm\\s+run\\s+test|node\\s+--test)",
+          "flags": "i",
+          "label": "DONE note names the real test command"
+        }
+      ]
+    },
+    "taskGraph": {
+      "requireCraftbookTask": true,
+      "requireTerminalStep": true
+    },
+    "unchangedFixtures": [
+      "source/upgrade-request.md",
+      "package.json"
+    ]
+  },
+  "rubric": {
+    "artifact": {
+      "path": "tasks/eval/change-notes.md",
+      "kind": "markdown"
+    },
+    "axes": [
+      {
+        "name": "Staged discipline",
+        "description": "The work ran as ordered stages with the suite genuinely run before the first change and between stages, and the record distinguishes executed stages from reverted or deferred ones."
+      },
+      {
+        "name": "Local evidence honesty",
+        "description": "Breaking-change notes come from sources actually read in the repository (VENDORING.md, the staged v2 file), with nothing presented as a release note that was invented."
+      },
+      {
+        "name": "Adaptation fidelity",
+        "description": "Call sites were adapted at their real locations to preserve exact rendering behavior under the v2 API, rather than accepting changed defaults or weakening tests."
+      }
+    ]
+  },
+  "qualityFocus": [
+    "staged upgrades with receipts",
+    "local breaking-change evidence",
+    "honest reverts"
+  ]
+}

package/authoring/tactical/evals/flaky-test-fix.json

@@ -0,0 +1,239 @@
+{
+  "schemaVersion": 1,
+  "title": "Deflake an order-dependent test with receipts",
+  "objective": "The craftbook drives a real deflake of a seeded order-dependence flake: statistical reproduction with recorded failing runs, a mechanism-level diagnosis, cleanup at the level every test shares (never skipping or weakening the test), and a stability proof - verified mechanically, including a five-run consecutive-green oracle, a late-probe order-robustness check, and a mutant restoring the original leaking order.",
+  "tags": [
+    "code",
+    "flaky-test",
+    "tests",
+    "tactical-fleet"
+  ],
+  "prompt": "Use the Fix a Flaky Test craftbook: our suite is red in CI but the failing test passes when people run it by itself, so it keeps getting waved through. The report is in source/flake-report.md. Find the nondeterminism, remove it for real, and prove the suite stays green.",
+  "setup": {
+    "projectName": "Flagship Flags",
+    "about": "A tiny feature-flags module and its test suite. Tests run with `npm run test` (node --test).",
+    "files": [
+      {
+        "path": "package.json",
+        "content": "{\n  \"name\": \"flagship\",\n  \"private\": true,\n  \"version\": \"1.0.0\",\n  \"type\": \"module\",\n  \"scripts\": {\n    \"test\": \"node --test\"\n  }\n}\n",
+        "modelInput": false
+      },
+      {
+        "path": "src/flags.js",
+        "content": "const flags = new Map();\n\nexport function enable(name) {\n  flags.set(name, true);\n}\n\nexport function disable(name) {\n  flags.delete(name);\n}\n\nexport function isEnabled(name) {\n  return flags.get(name) === true;\n}\n\nexport function enabledCount() {\n  return flags.size;\n}\n\nexport function resetFlags() {\n  flags.clear();\n}\n"
+      },
+      {
+        "path": "tests/flags.test.mjs",
+        "content": "import assert from 'node:assert/strict';\nimport { test } from 'node:test';\nimport { enable, disable, isEnabled, enabledCount } from '../src/flags.js';\n\ntest('enable turns a flag on', () => {\n  enable('beta-banner');\n  assert.equal(isEnabled('beta-banner'), true);\n});\n\ntest('disable turns a flag off again', () => {\n  enable('holiday-theme');\n  disable('holiday-theme');\n  assert.equal(isEnabled('holiday-theme'), false);\n});\n\ntest('unknown flags read as disabled', () => {\n  assert.equal(isEnabled('dark-mode'), false);\n});\n\ntest('a fresh session starts with no flags enabled', () => {\n  assert.equal(enabledCount(), 0);\n});\n"
+      },
+      {
+        "path": "source/flake-report.md",
+        "content": "# Flake report\n\nThe suite has been red in CI all week, and every developer who ran the failing test by itself saw it pass - so it kept getting waved through.\n\n    npm run test\n    not ok 4 - a fresh session starts with no flags enabled\n      AssertionError: 1 !== 0\n\nRunning the full suite fails on that test consistently; running just that one test with `--test-name-pattern \"fresh session\"` is green every time. Nobody has touched `src/flags.js` in weeks.\n\nConstraints:\n\n- The pristine-session test guards a real invariant (feature sessions must start clean). Do not delete or weaken it.\n- The app depends on `src/flags.js` exactly as it is - fix the tests' hygiene, not the module's API.\n",
+        "modelInput": true
+      },
+      {
+        "path": "tests/verify-deflake.mjs",
+        "surface": "harness",
+        "content": "import assert from 'node:assert/strict';\nimport { execFileSync } from 'node:child_process';\nimport { cpSync, mkdirSync, readFileSync, readdirSync, rmSync, writeFileSync } from 'node:fs';\nimport { join } from 'node:path';\nimport { enable, isEnabled, resetFlags, enabledCount } from '../src/flags.js';\n\nconst root = process.cwd();\nconst ORACLE_FILE = 'verify-deflake.mjs';\n\n// 1) The flags module still means what it meant.\nresetFlags();\nenable('oracle-probe');\nassert.equal(isEnabled('oracle-probe'), true, 'enable must still work');\nassert.equal(enabledCount(), 1, 'enabledCount must still count');\nresetFlags();\nassert.equal(enabledCount(), 0, 'resetFlags must still clear');\nconsole.log('DEFLAKE_ORACLE flags semantics ok');\n\nfunction runSuite(cwd) {\n  try {\n    execFileSync(process.execPath, ['--test'], { cwd, stdio: 'pipe', timeout: 30000 });\n    return 0;\n  } catch (err) {\n    return typeof err.status === 'number' ? err.status : 1;\n  }\n}\n\nfunction copyTree(dest) {\n  rmSync(dest, { recursive: true, force: true });\n  mkdirSync(dest, { recursive: true });\n  cpSync(join(root, 'src'), join(dest, 'src'), { recursive: true });\n  cpSync(join(root, 'tests'), join(dest, 'tests'), { recursive: true });\n  rmSync(join(dest, 'tests', ORACLE_FILE), { force: true });\n  cpSync(join(root, 'package.json'), join(dest, 'package.json'));\n  return dest;\n}\n\n// 2) Five consecutive green runs of the real suite.\nfor (let run = 1; run <= 5; run += 1) {\n  assert.equal(runSuite(root), 0, `full-suite run ${run} of 5 must be green`);\n}\nconsole.log('DEFLAKE_ORACLE stable 5x');\n\n// 3) The pristine-session assertion survives, unweakened and unskipped.\nconst testFiles = readdirSync(join(root, 'tests')).filter(\n  (name) => name !== ORACLE_FILE && /\\.(mjs|cjs|js)$/.test(name),\n);\nconst combined = testFiles\n  .map((name) => readFileSync(join(root, 'tests', name), 'utf8'))\n  .join('\\n');\nassert.match(\n  combined,\n  /enabledCount\\s*\\(\\s*\\)\\s*(?:,\\s*0|===\\s*0)/,\n  'the pristine-session assertion (enabledCount() must be 0) has to survive',\n);\nassert.ok(\n  !/\\.skip\\s*\\(|\\.todo\\s*\\(|\\.only\\s*\\(|\\{\\s*skip\\s*:|\\{\\s*todo\\s*:/.test(combined),\n  'no test may be skipped, stubbed as todo, or narrowed with only',\n);\nconsole.log('DEFLAKE_ORACLE assertion intact');\n\n// 4) Order robustness: tests appended at the END of the suite must inherit\n// the cleanup. The first probe enables a flag and never cleans up; the\n// second expects a pristine session anyway. Only hygiene that covers every\n// test (beforeEach/afterEach-style) passes both - a one-off reset inside\n// the seeded failing test, or a reorder, fails here. Aliased imports keep\n// the probe valid whatever the model did to the file's own imports.\nconst hostFile = testFiles.find((name) =>\n  /enabledCount/.test(readFileSync(join(root, 'tests', name), 'utf8')),\n);\nassert.ok(hostFile, 'no test file references enabledCount');\nconst probeRoot = copyTree(join(root, '.oracle-probe'));\nconst probeTests =\n  readFileSync(join(probeRoot, 'tests', hostFile), 'utf8') +\n  \"\\n\\nimport { test as zzTest } from 'node:test';\\nimport zzAssert from 'node:assert/strict';\\nimport { enable as zzEnable, isEnabled as zzIsEnabled, enabledCount as zzEnabledCount } from '../src/flags.js';\\n\\nzzTest('zz probe enables a flag of its own', () => {\\n  zzEnable('zz-probe-flag');\\n  zzAssert.equal(zzIsEnabled('zz-probe-flag'), true);\\n});\\n\\nzzTest('zz probe still gets a pristine session', () => {\\n  zzAssert.equal(zzEnabledCount(), 0);\\n});\\n\";\nwriteFileSync(join(probeRoot, 'tests', hostFile), probeTests);\nassert.equal(\n  runSuite(probeRoot),\n  0,\n  'a test appended after the suite must still see a pristine session - cleanup has to cover every test, not only the seeded order',\n);\nconsole.log('DEFLAKE_ORACLE order robust');\n\n// 5) Mutant: restore the ORIGINAL leaking test file - the suite must fail\n// again, proving the seeded defect is still detectable and the green suite\n// comes from removed nondeterminism, not from a neutered module or dodge.\nconst ORIGINAL_TEST_FILE = `import assert from 'node:assert/strict';\nimport { test } from 'node:test';\nimport { enable, disable, isEnabled, enabledCount } from '../src/flags.js';\n\ntest('enable turns a flag on', () => {\n  enable('beta-banner');\n  assert.equal(isEnabled('beta-banner'), true);\n});\n\ntest('disable turns a flag off again', () => {\n  enable('holiday-theme');\n  disable('holiday-theme');\n  assert.equal(isEnabled('holiday-theme'), false);\n});\n\ntest('unknown flags read as disabled', () => {\n  assert.equal(isEnabled('dark-mode'), false);\n});\n\ntest('a fresh session starts with no flags enabled', () => {\n  assert.equal(enabledCount(), 0);\n});\n`;\nconst mutantRoot = copyTree(join(root, '.oracle-mutant'));\nwriteFileSync(join(mutantRoot, 'tests', 'flags.test.mjs'), ORIGINAL_TEST_FILE);\nassert.notEqual(\n  runSuite(mutantRoot),\n  0,\n  'restoring the original leaking order must make the suite fail again - the seeded defect stopped being detectable',\n);\nconsole.log('DEFLAKE_ORACLE mutant caught');\nconsole.log('DEFLAKE_ORACLE done');\n"
+      }
+    ],
+    "craftbookParams": {
+      "workPath": "tasks/eval",
+      "scope": "The test 'a fresh session starts with no flags enabled' in tests/flags.test.mjs fails whenever the whole suite runs and passes by itself; see source/flake-report.md. Make the suite deterministic without weakening that test."
+    }
+  },
+  "success": {
+    "summary": "The order-dependence mechanism is removed with hygiene that covers every test, the pristine-session assertion survives unweakened, five consecutive suite runs are green, a late-appended probe still sees a pristine session, and restoring the original test file makes the suite fail again.",
+    "deliverables": [
+      {
+        "path": "tasks/eval/flake-evidence.md",
+        "kind": "markdown-notes",
+        "artifact": true,
+        "minBytes": 500,
+        "checks": [
+          {
+            "kind": "contains",
+            "file": "tasks/eval/flake-evidence.md",
+            "pattern": "^##\\s+Symptom[\\s\\S]*^##\\s+Runs observed[\\s\\S]*^##\\s+Failure signature[\\s\\S]*^##\\s+Suspected class",
+            "flags": "im",
+            "label": "flake-evidence sections in order"
+          },
+          {
+            "kind": "citationsResolve",
+            "file": "tasks/eval/flake-evidence.md",
+            "minCitations": 2,
+            "artifact": true
+          }
+        ]
+      },
+      {
+        "path": "tasks/eval/diagnosis.md",
+        "kind": "markdown-notes",
+        "artifact": true,
+        "minBytes": 600,
+        "checks": [
+          {
+            "kind": "contains",
+            "file": "tasks/eval/diagnosis.md",
+            "pattern": "^##\\s+Mechanism[\\s\\S]*^##\\s+Experiment[\\s\\S]*^##\\s+Defect site[\\s\\S]*^##\\s+Siblings checked",
+            "flags": "im",
+            "label": "diagnosis sections in order"
+          },
+          {
+            "kind": "contains",
+            "file": "tasks/eval/diagnosis.md",
+            "pattern": "tests/flags\\.test\\.mjs|src/flags\\.js",
+            "label": "names the real site of the shared-state hazard"
+          },
+          {
+            "kind": "citationsResolve",
+            "file": "tasks/eval/diagnosis.md",
+            "minCitations": 2,
+            "artifact": true
+          }
+        ]
+      },
+      {
+        "path": "tasks/eval/fix-notes.md",
+        "kind": "markdown-notes",
+        "artifact": true,
+        "minBytes": 700,
+        "checks": [
+          {
+            "kind": "contains",
+            "file": "tasks/eval/fix-notes.md",
+            "pattern": "^##\\s+Problem[\\s\\S]*^##\\s+Change[\\s\\S]*^##\\s+Files touched[\\s\\S]*^##\\s+Regression coverage[\\s\\S]*^##\\s+Risk[\\s\\S]*^##\\s+How to verify",
+            "flags": "im",
+            "label": "fix-notes sections in order"
+          },
+          {
+            "kind": "citationsResolve",
+            "file": "tasks/eval/fix-notes.md",
+            "minCitations": 2,
+            "artifact": true
+          }
+        ]
+      },
+      {
+        "path": "tasks/eval/validation.md",
+        "kind": "markdown-notes",
+        "artifact": true,
+        "minBytes": 300,
+        "checks": [
+          {
+            "kind": "contains",
+            "file": "tasks/eval/validation.md",
+            "pattern": "^##\\s+Runs[\\s\\S]*^##\\s+Result",
+            "flags": "im",
+            "label": "validation sections in order"
+          }
+        ]
+      },
+      {
+        "path": "tasks/eval/review.md",
+        "kind": "markdown-report",
+        "artifact": true,
+        "minBytes": 400,
+        "checks": [
+          {
+            "kind": "contains",
+            "file": "tasks/eval/review.md",
+            "pattern": "Verdict:\\s*(?:PASS|REVISE)",
+            "flags": "i",
+            "label": "explicit reviewer verdict"
+          }
+        ]
+      }
+    ],
+    "checks": [
+      {
+        "kind": "notContains",
+        "file": "tests/flags.test.mjs",
+        "pattern": "\\.skip\\s*\\(|\\.todo\\s*\\(|\\.only\\s*\\(|\\{\\s*skip\\s*:|\\{\\s*todo\\s*:",
+        "label": "the flaky test is not skipped, stubbed, or narrowed"
+      },
+      {
+        "kind": "sourceParses",
+        "file": "tests/flags.test.mjs"
+      },
+      {
+        "kind": "nodeRuns",
+        "file": "tests/flags.test.mjs",
+        "timeoutMs": 30000
+      },
+      {
+        "kind": "nodeScriptPasses",
+        "script": "tests/verify-deflake.mjs",
+        "timeoutMs": 60000,
+        "requiredOutput": [
+          {
+            "pattern": "DEFLAKE_ORACLE flags semantics ok",
+            "label": "the flags module still behaves"
+          },
+          {
+            "pattern": "DEFLAKE_ORACLE stable 5x",
+            "label": "five consecutive green suite runs"
+          },
+          {
+            "pattern": "DEFLAKE_ORACLE assertion intact",
+            "label": "the pristine-session assertion survives unweakened"
+          },
+          {
+            "pattern": "DEFLAKE_ORACLE order robust",
+            "label": "a late-appended test still sees a pristine session"
+          },
+          {
+            "pattern": "DEFLAKE_ORACLE mutant caught",
+            "label": "restoring the original order makes the suite fail again"
+          }
+        ]
+      }
+    ],
+    "taskNotes": {
+      "minBytes": 120,
+      "checks": [
+        {
+          "kind": "contains",
+          "file": "task-notes.md",
+          "pattern": "\\bDONE\\b[\\s\\S]*(npm\\s+run\\s+test|node\\s+--test|consecutive)",
+          "flags": "i",
+          "label": "DONE note cites the stability proof"
+        }
+      ]
+    },
+    "taskGraph": {
+      "requireCraftbookTask": true,
+      "requireTerminalStep": true
+    },
+    "unchangedFixtures": [
+      "package.json",
+      "src/flags.js",
+      "source/flake-report.md"
+    ]
+  },
+  "rubric": {
+    "artifact": {
+      "path": "tasks/eval/fix-notes.md",
+      "kind": "markdown"
+    },
+    "axes": [
+      {
+        "name": "Mechanism-level diagnosis",
+        "description": "The nondeterminism is named as a concrete mechanism (shared module state leaking across tests) and pinned with a real experiment, not guessed from the symptom."
+      },
+      {
+        "name": "No masking",
+        "description": "The flake is removed with cleanup or isolation that covers every test; the flaky test keeps its exact assertion and gains no skip, retry, or widened timeout."
+      },
+      {
+        "name": "Receipt-backed stability",
+        "description": "Stability is proven with consecutive green runs recorded as real receipts, and anything unverified is stated plainly instead of claimed."
+      }
+    ]
+  },
+  "qualityFocus": [
+    "statistical reproduction",
+    "mechanism removal over masking",
+    "consecutive-green receipts"
+  ]
+}