universal-dev-standards 6.4.0 → 6.5.0

package/bundled/ai/standards/mutation-testing.ai.yaml

@@ -3,12 +3,19 @@
 
 id: mutation-testing
 meta:
-  version: "1.0.0"
-  updated: "2026-05-04"
+  version: "1.1.0"
+  updated: "2026-08-14"
   source: core/mutation-testing.md
   description: >
     Mutation testing methodology to evaluate test suite effectiveness.
     Answers "do my tests actually catch bugs?" beyond line coverage.
+  changelog:
+    - version: "1.1.0"
+      date: "2026-08-14"
+      change: >
+        Added attribution scope (kill is credited to the first failing test,
+        not to any one level), one-sided-invariant pairing requirement, and
+        equivalent-mutant classification requirement.
 
 # ─────────────────────────────────────────────────────────
 # Core Concepts
@@ -40,6 +47,67 @@ core_concepts:
     - category: Boolean literal
       examples: ["true → false", "false → true"]
 
+# ─────────────────────────────────────────────────────────
+# Attribution: kill is credited to the first failing test
+# ─────────────────────────────────────────────────────────
+attribution:
+  description: >
+    "Killed" means some test in the run failed against the mutant — most
+    tools do not record which test killed it, and none records which test
+    level. A 7/7 kill score therefore verifies the whole suite that ran
+    against the mutant, not any single test, and not any single test level
+    (unit vs. integration vs. property).
+  consequence: >
+    "The property suite verifies X" is not a claim an aggregate mutation run
+    supports. To support it, re-run mutation testing with only the property
+    suite active. If the isolated run kills fewer mutants than the aggregate
+    run, the gap is exactly what unit/integration tests were quietly covering.
+  isolated_rerun_example: "npx stryker run --mutate 'src/module/**' -- --project=property"
+  rule: >
+    High-risk modules (the same set the 80% threshold applies to —
+    auth/license/payment/security) must re-run mutants against the property
+    suite in isolation before any claim of "property-verified" is made.
+
+# ─────────────────────────────────────────────────────────
+# One-sided invariants miss fail-closed defects
+# ─────────────────────────────────────────────────────────
+one_sided_invariants:
+  description: >
+    A property like "output never exceeds the limit" is one-sided: it cannot
+    detect a mutant that makes the code fail closed (reject everything,
+    including valid input), because a fail-closed mutant never produces an
+    over-limit output. The mutation score looks unaffected while an entire
+    class of defect (denial of service, wrongly rejected requests) is
+    invisible to the suite.
+  fix: >
+    Pair every one-sided invariant with its opposite boundary — "never
+    exceeds the limit" needs a companion property such as "accepts everything
+    at or below the limit" — so both over-permissive and over-restrictive
+    mutants have a path to detection.
+
+# ─────────────────────────────────────────────────────────
+# Equivalent mutants are not survivors to chase
+# ─────────────────────────────────────────────────────────
+equivalent_mutants:
+  description: >
+    A surviving mutant is not automatically a test gap. Some mutants are
+    semantically equivalent to the original — no input can distinguish their
+    behavior — and no test can kill them, regardless of how it's written.
+    Forcing a kill with an assertion that exists only to move the score
+    (e.g. toBeDefined() on an incidental value) produces a hollow test
+    without closing any real gap.
+  required_classification:
+    - outcome: genuine_gap
+      action: Write a test that exercises the distinguishing behavior.
+    - outcome: equivalent
+      action: >
+        Record as "equivalent, because <reason>", stating the input(s)
+        checked to reach that conclusion.
+  note: >
+    An unclassified survivor is not the same as a classified-equivalent one.
+    Only a classified-equivalent mutant may be excluded from the score's
+    denominator.
+
 # ─────────────────────────────────────────────────────────
 # Tools
 # ─────────────────────────────────────────────────────────
@@ -175,12 +243,44 @@ rules:
     priority: required
     note: Reserve for pre-release gate and on-demand runs
 
+  - id: mutation-property-isolation-required
+    trigger: claiming a property-based test suite verifies specific behavior, based on an aggregate mutation run
+    instruction: >
+      Re-run mutation testing with only the property suite active before making
+      the claim. An aggregate kill score (property + unit + integration
+      together) verifies the whole suite, not the property suite alone.
+      Required for high-risk modules (auth/license/payment/security) before
+      any "property-verified" claim.
+    priority: required
+
+  - id: mutation-one-sided-invariant-pairing
+    trigger: writing or reviewing a property/invariant used as a mutation-testing target
+    instruction: >
+      A one-sided invariant (e.g. "never exceeds the limit") cannot detect a
+      fail-closed mutant. Pair it with an opposite-boundary property (e.g.
+      "accepts everything at or below the limit") so both over-permissive and
+      over-restrictive mutants have a path to detection.
+    priority: required
+
+  - id: mutation-equivalent-classification-required
+    trigger: reviewing a surviving mutant
+    instruction: >
+      Classify every reviewed survivor as either a genuine gap (write a test
+      exercising the distinguishing behavior) or "equivalent, because
+      <reason>" (record the input(s) checked). Do not force a kill with an
+      assertion that exists only to move the score. Only a
+      classified-equivalent mutant may be excluded from the denominator.
+    priority: required
+
 anti_patterns:
   - Treating 100% line coverage as sufficient (lines covered ≠ mutations killed)
   - Adding mutation testing to pre-commit hooks (makes commits 10-60 minutes long)
   - Accepting AI-generated tests without mutation score validation
   - Killing mutations by adding trivial assertions (expect(x).toBeDefined())
   - Targeting only happy paths in mutation testing (branches and boundaries are key)
+  - Claiming "the property suite verifies X" from an aggregate (not isolated) mutation run
+  - Using only a one-sided invariant for a property that has a fail-closed failure mode
+  - Forcing a kill on a semantically equivalent mutant instead of classifying it "equivalent, because <reason>"
 
 quick_reference:
   mutation_testing_checklist: |
@@ -190,3 +290,6 @@ quick_reference:
     □ Pre-release: run full mutation suite before tagging version
     □ AI-generated tests: validate with mutation score before accepting
     □ NOT in commit hooks (too slow)
+    □ High-risk modules: property suite re-run in isolation before "property-verified" claims
+    □ One-sided invariants paired with their opposite boundary
+    □ Every reviewed survivor classified: genuine gap, or "equivalent, because <reason>"

package/bundled/ai/standards/project-structure.ai.yaml

@@ -110,7 +110,7 @@ physical_spec:
   schema:
     root:
       required: ["src", "tests", "docs"]
-      ignored: ["dist", "build", "node_modules", "out", "bin"]
+      ignored: ["dist", "build", "node_modules", "out", "bin", "__pycache__", ".venv", "venv", "target", "vendor", ".gradle", "pkg", ".dart_tool"]
   validator:
     command: "find . -maxdepth 1 -not -path '*/.*'"
     rule: "check_required_directories"

package/bundled/ai/standards/security-standards.ai.yaml

@@ -126,13 +126,34 @@ checklist:
 physical_spec:
   type: custom_script
   validator:
-    command: "npm audit --dry-run > /dev/null 2>&1 || test -f .trivyignore || test -f .snyk"
+    command: >
+      if [ -f uds.project.yaml ] && grep -q 'security:' uds.project.yaml; then
+        exit 0;
+      elif [ -f Makefile ] && grep -q '^security:' Makefile; then
+        exit 0;
+      elif [ -f package.json ]; then
+        npm audit --dry-run > /dev/null 2>&1 || test -f .trivyignore || test -f .snyk;
+      elif [ -f requirements.txt ] || [ -f pyproject.toml ] || [ -f setup.py ]; then
+        pip-audit --dry-run > /dev/null 2>&1 || python -m safety check --dry-run > /dev/null 2>&1 || test -f .trivyignore || test -f .snyk;
+      elif [ -f go.mod ]; then
+        govulncheck ./... > /dev/null 2>&1 || test -f .trivyignore || test -f .snyk;
+      elif [ -f pom.xml ] || [ -f build.gradle ] || [ -f build.gradle.kts ]; then
+        test -f dependency-check-report.xml || test -f .trivyignore || test -f .snyk;
+      elif [ -f Cargo.toml ]; then
+        cargo audit > /dev/null 2>&1 || test -f .trivyignore || test -f .snyk;
+      elif [ -f Gemfile ]; then
+        bundle audit > /dev/null 2>&1 || test -f .trivyignore || test -f .snyk;
+      else
+        test -f .trivyignore || test -f .snyk;
+      fi
     rule: "security_scan_configured"
 
 enforcement:
   hook_type: PreToolUse
+  trigger: PreToolUse
   matcher:
     tool: Bash
   script_ref: "scripts/hooks/check-dangerous-cmd.js"
+  hook_script: "scripts/hooks/check-dangerous-cmd.js"
   severity: error
   timeout_ms: 500

package/bundled/ai/standards/spec-driven-development.ai.yaml

@@ -3,8 +3,8 @@
 
 id: spec-driven-development
 meta:
-  version: "1.2.0"
-  updated: "2025-12-30"
+  version: "1.3.0"
+  updated: "2026-08-12"
   source: methodologies/guides/sdd-guide.md
   description: Spec-Driven Development workflow where documentation precedes implementation
 
@@ -53,6 +53,63 @@ principles:
       - Critical hotfixes (restore service immediately, document later)
       - Trivial changes (typos, comments, formatting)
 
+spec_types:
+  description: SDD supports multiple granularity levels via spec-type field
+  default: feature
+  values:
+    - type: feature
+      description: Describes a single functional change (default, backward compatible)
+      granularity: function/feature level
+      template: standard SDD spec template
+
+    - type: agent
+      description: Describes an Agent role definition spanning multiple features
+      granularity: role level
+      template: agent_spec_template
+      required_sections:
+        - role_definition
+        - capability_scope
+        - interface_contract
+        - agent_interactions
+        - related_feature_specs
+
+    - type: infrastructure
+      description: Describes cross-cutting concerns or infrastructure changes
+      granularity: system level
+      template: standard SDD spec template with infrastructure focus
+
+  agent_spec_template:
+    sections:
+      - name: Role Definition
+        fields: [role_name, role_responsibility, autonomy_level]
+        autonomy_levels: "L1-L5 (per DEC-065)"
+
+      - name: Capability Scope
+        fields: [capabilities_list, out_of_scope]
+
+      - name: Interface Contract
+        subsections:
+          - name: Input
+            fields: [accepted_message_types, required_fields, optional_fields]
+          - name: Output
+            fields: [artifact_types, success_exit_condition, failure_exit_condition]
+
+      - name: Agent Interactions
+        fields: [upstream_agents, downstream_agents, parallel_agents]
+
+      - name: Related Feature SPECs
+        description: List of feature SPECs this agent participates in
+        format: "- [SPEC-NNN] description of this agent's role in that spec"
+
+  traceability:
+    feature_to_agent:
+      field: agent-id
+      description: Optional field in feature SPEC header referencing the owning Agent SPEC
+      example: "agent-id: SPEC-090"
+    agent_to_features:
+      section: Related Feature SPECs
+      description: Agent SPEC lists all feature SPECs it participates in
+
 spec_template:
   sections:
     - name: Summary

package/bundled/ai/standards/test-governance.ai.yaml

@@ -8,8 +8,16 @@ standard:
   description: Test policy, completion criteria, and environment management standards
 
   meta:
-    version: "1.0.0"
-    updated: "2026-03-11"
+    version: "1.2.0"
+    updated: "2026-08-14"
+    changelog:
+      - version: "1.2.0"
+        date: "2026-08-14"
+        change: >
+          Added fail-closed-threshold-gate rule: coverage/lint/mutation/any
+          bounded-metric check must exit non-zero when below threshold, using
+          the tool's own enforcement flag rather than a wrapper that only
+          prints the number and always exits 0.
     references:
       - "ISO/IEC/IEEE 29119-2 (Test Processes)"
       - "ISO/IEC/IEEE 29119-3 (Test Documentation)"
@@ -58,6 +66,36 @@ standard:
         - "eslint . (TypeScript/JavaScript)"
         - "mypy . (Python type checking)"
 
+    threshold_gates:
+      description: >
+        A measurement layer that prints a percentage but exits 0 regardless of
+        whether it cleared the threshold is a report, not a gate. It stays
+        green while the number it prints drifts downward across commits, and
+        nothing stops the next merge.
+      requirement: >
+        Every check with a pass/fail threshold (coverage, lint, mutation
+        score, or any other bounded metric) must translate "below threshold"
+        into a non-zero exit code via the tool's own enforcement flag — not a
+        wrapper script that re-parses printed output after the fact.
+      fail_closed_flags:
+        - tool: pytest-cov
+          flag: "--cov-fail-under=<N>"
+        - tool: coverage.py
+          flag: "coverage report --fail-under=<N>"
+        - tool: diff-cover
+          flag: "diff-cover coverage.xml --fail-under=<N>"
+        - tool: "nyc / Istanbul"
+          flag: "--check-coverage --lines <N>"
+        - tool: Stryker Mutator
+          flag: "thresholds.break in stryker.config.json"
+        - tool: ESLint
+          flag: "--max-warnings 0"
+      note: >
+        A wrapper that computes the number, prints it, and always exit 0
+        satisfies neither this rule nor verification-evidence's Evidence
+        Validity rule 1 — the tool's exit code no longer carries any
+        information about the artefact it measured.
+
   completion_criteria:
     description: >
       Test completion criteria define when testing activities can be considered done.
@@ -159,6 +197,15 @@ standard:
         BUG-A08 post-mortem (2026-04-20): 22 tests existed in UDS but were never
         executed by any CI gate, passing silently and masking real failures.
 
+    - id: fail-closed-threshold-gate
+      trigger: configuring or reviewing any coverage/lint/mutation/other threshold check
+      instruction: >
+        The check must use the tool's fail-under (or equivalent) enforcement
+        flag so it exits non-zero when the threshold is not met. A script that
+        only prints the number and always exits 0 is a report, not a gate, and
+        does not satisfy this rule.
+      priority: required
+
     - id: gate-wiring-required
       trigger: adding any quality detection script to the repository
       instruction: |

package/bundled/ai/standards/testing.ai.yaml

@@ -12,8 +12,8 @@ standard:
     - "See full-coverage-testing.ai.yaml for coverage policy (XSPEC-178)"
 
   meta:
-    version: "2.1.0"
-    updated: "2026-05-06"
+    version: "2.2.0"
+    updated: "2026-08-12"
     source: core/testing-standards.md
     guide: skills/testing-guide/testing-theory.md
     description: Testing structure and principles. Coverage policy moved to full-coverage-testing (XSPEC-178).
@@ -134,9 +134,55 @@ standard:
         feature page returned 500 silently — full E2E suite passed with false
         confidence, masking a production crash.
 
+migration_testing:
+  schema_parity:
+    description: For any project migrating a database schema, include automated schema parity verification rather than relying on manually maintained documents.
+    pattern:
+      step_1: Query source schema metadata at test time (e.g., SQLite PRAGMA table_info, MySQL information_schema)
+      step_2: Query target schema metadata at test time (e.g., EF Core DbContext.Model.GetProperties(), direct information_schema query)
+      step_3: Assert every expected source column exists in target with compatible type, OR is documented as an intentional change (intentional_removals list)
+    ci_gate:
+      trigger: PR touches entity definitions or migration scripts
+      enforcement: required (blocking)
+    catches:
+      - Column renames not propagated to mapper
+      - Type changes (e.g. TEXT → NVARCHAR(50)) that silently truncate data
+      - Source columns added but never mapped to target
+      - Target columns removed but still referenced in source code
+    reference_implementation:
+      language: csharp
+      framework: EF Core vs SQLite
+      pattern: |
+        var properties = dbContext.Model.FindEntityType(typeof(Entity))!
+            .GetProperties().Select(p => p.GetColumnName()).ToHashSet();
+        var sourceColumns = conn.Query<string>(
+            "SELECT name FROM pragma_table_info('tablename')").ToHashSet();
+        var unmapped = sourceColumns.Except(knownRemovals).Except(properties);
+        Assert.Empty(unmapped);
+
 physical_spec:
   type: custom_script
   validator:
     command: >
-      test -f vitest.config.ts || test -f vitest.config.js || test -f jest.config.js || grep -q '"test":' package.json
+      if [ -f uds.project.yaml ] && grep -q 'commands:' uds.project.yaml; then
+        exit 0;
+      elif [ -f Makefile ] && grep -q '^test:' Makefile; then
+        exit 0;
+      elif [ -f justfile ] && grep -q '^test:' justfile; then
+        exit 0;
+      elif [ -f package.json ]; then
+        test -f vitest.config.ts || test -f vitest.config.js || test -f jest.config.js || grep -q '"test":' package.json;
+      elif [ -f requirements.txt ] || [ -f pyproject.toml ] || [ -f setup.py ] || [ -f setup.cfg ]; then
+        python -m pytest --collect-only -q > /dev/null 2>&1 || test -f pytest.ini || test -f pyproject.toml;
+      elif [ -f go.mod ]; then
+        test -f go.sum || go list ./... > /dev/null 2>&1;
+      elif [ -f pom.xml ] || [ -f build.gradle ] || [ -f build.gradle.kts ]; then
+        test -f pom.xml || test -f build.gradle || test -f build.gradle.kts;
+      elif [ -f Cargo.toml ]; then
+        grep -q '\[\[test\]\]\|test = ' Cargo.toml || test -d tests;
+      elif [ -f Gemfile ]; then
+        test -f spec/spec_helper.rb || test -f test/test_helper.rb;
+      else
+        echo "⚠️  請執行 uds configure 建立 uds.project.yaml" && exit 1;
+      fi
     rule: "test_runner_configured"

package/bundled/ai/standards/translation-lifecycle-standards.ai.yaml

@@ -3,8 +3,8 @@
 
 id: translation-lifecycle-standards
 meta:
-  version: "1.0.0"
-  updated: "2026-04-20"
+  version: "1.0.1"
+  updated: "2026-08-12"
   status: trial
   since: "2026-04-20"
   expires: "2026-10-20"
@@ -69,7 +69,7 @@ automation:
     file: .githooks/pre-commit
     trigger: core/*.md files staged
     behavior: warn on OUTDATED, never block commit
-    setup: ./scripts/install-hooks.sh
+    setup: node scripts/install-hooks.mjs
   release_gate:
     command: bash scripts/check-translation-sync.sh
     exit_1_conditions: [MISSING, MAJOR]
@@ -139,7 +139,7 @@ integration_points:
     role: primary automation; enforces severity levels
   - system: .githooks/pre-commit
     role: commit-time reminder when core/ files staged
-  - system: bump-version.sh
+  - system: bump-version.mjs
     role: release-time advisory snapshot
   - system: pre-release-check.sh
     role: final gate before npm publish (calls check-translation-sync.sh)

package/bundled/ai/standards/verification-evidence.ai.yaml

@@ -8,14 +8,18 @@ standard:
   description: 驗證證據標準，強化 anti-hallucination
 
   meta:
-    version: "1.2.0"
-    updated: "2026-07-17"
+    version: "1.3.0"
+    updated: "2026-08-14"
     source: core/verification-evidence.md
     description: >
       驗證證據標準 — Iron Law: 無驗證證據不可聲稱完成。
       v1.1.0: Evidence must specify which environment layer it was collected from (XSPEC-204).
       v1.2.0: Evidence itself must be validated — a tool can fail silently and its
       output is indistinguishable from a real result (XSPEC-340).
+      v1.3.0: Evidence must postdate the last edit to what it verifies (VE-011,
+      stale evidence); a documented coverage gap must be registered in a dated
+      exception inventory, not left as prose alone (VE-012, narrow-coverage
+      registration — shared with class-level-fix).
     inspired_by: superpowers/verification-before-completion
 
   guidelines:
@@ -27,6 +31,8 @@ standard:
     - "代理報告 success ≠ 實際 success，需獨立驗證"
     - "**工具回報 success ≠ 實際 success**——驗證指令本身可能沒跑起來，而其輸出與真結果無法區分"
     - "驗證輸出截斷至合理長度（2000 字元）但保留關鍵資訊"
+    - "**證據必須晚於它所驗證對象的最後一次編輯**——最近一次執行剛好通過，不代表它驗證的是現狀（VE-011）"
+    - "**窄涵蓋必須登記，不能只揭露**——文件化的證據/涵蓋缺口須登記到帶到期日的例外清冊，單純寫下來不算滿足（VE-012）"
 
   # 以下不構成驗證證據。原本只存在於 zh-TW 譯文裡（英文來源與本檔皆無）——
   # v1.2.0 一併升上來源與本檔（XSPEC-340 R4：譯文比來源更完整，且無 gate 會報）。
@@ -67,12 +73,22 @@ standard:
         detail: >
           `set -o pipefail` 下 `producer | grep -q pattern` 會繼承 producer 的非 0，
           grep 中不中都無關。需要依內容決策時：**先接住輸出，再判斷**。
+      - name: "證據必須晚於它所驗證對象的最後一次編輯"
+        detail: >
+          在受測程式碼、提示詞或設定的最近一次變更之前擷取的驗證執行，證明的是別的東西——
+          不是正在被主張的那件事。陷阱是靜默的：完整套件先跑過且全綠，之後才編輯了某個東西
+          （提示詞、設定檔、門檻），編輯之後只跑了範圍更窄的檢查（lint、部分套件），
+          結果卻拿更早的那次綠燈當作現狀的證據。**證據必須是最後一次編輯之後的單一次新鮮執行，
+          不是「剛好通過的最近一次執行」。**（VE-011）
     provenance: >
-      證據來自 2026-07-17 單一 agent（Claude Opus 4.8）單日的十次實例，
+      規則 1–4 的證據來自 2026-07-17 單一 agent（Claude Opus 4.8）單日的十次實例，
       記錄於 AsiaOstrich XSPEC-340；其中 exit_code=0 四次為假、exit_code≠0 兩次為真
       且依 VE-002 行動摧毀了健康的產物。樣本密集但來源單一——然而每個失敗都源自
       **工具本身的語意**（sudo / gpg / pipefail / POSIX exit code）而非模型的性質，
-      故任何驅動同一批工具的 agent 都暴露在同樣的陷阱下。
+      故任何驅動同一批工具的 agent 都暴露在同樣的陷阱下。規則 5（VE-011）為獨立補強，
+      非同一批樣本：其失效樣態為「完整套件跑過通過 → 之後改了提示詞/設定 →
+      只跑範圍更窄的檢查（如 lint）就提交 → CI 紅在釘住新內容雜湊的測試」——
+      通過的是「最後一次變更之前」的套件。
 
   evidence_format:
     fields:
@@ -117,6 +133,7 @@ standard:
     - "exit_code ≠ 0 → 標記為驗證失敗——**除非**該工具在受測狀態下設計上就回非 0（見 VE-007），此時改依輸出內容判斷"
     - "exit_code = 0 但該指令不可能量到它宣稱的東西 → 標記為未驗證"
     - "證據斷言「不存在」（0／空／查無）→ 未證明查詢工具執行成功前，標記為未驗證"
+    - "證據的時間戳早於它所驗證對象的最後一次編輯 → 標記為過期（stale），須在編輯後重跑，不得引用更早那次通過"
     - "多個驗證步驟 → 所有步驟都必須通過"
 
   rules:
@@ -174,6 +191,31 @@ standard:
       trigger: "證據的 exit_code 來自管線（尤其在 `set -o pipefail` 下）"
       action: "該 exit code 不歸屬於任何單一階段。先接住輸出，再依內容判斷。"
       priority: medium
+    - id: VE-011
+      trigger: "證據的執行時間戳早於它所宣稱驗證之產物的最後一次編輯"
+      action: >
+        標記為過期（stale）——對現狀不成立為證據；要求在編輯後補一次單一次新鮮執行。
+      priority: high
+
+  # ── v1.3.0 窄涵蓋登記規則（B-01 借鑒）──
+  # 與 class-level-fix 的同名規則共用同一條要求：文件化的缺口若沒有登記到
+  # 帶到期日的例外清冊，揭露就會變成永久豁免。
+  narrow_coverage_registration:
+    description: "文件化的證據/涵蓋缺口（environment-stratification ⚠️/❌、或任何窄於其主張的檢查）必須登記，不能只揭露"
+    requirement: >
+      任何這一類已記錄的缺口，須同時登記到一份帶到期日的例外清冊——獨立於標準本文或
+      matrix 條目本身之外——指名缺口、其成因、並附上覆核或到期日期。只有一則註腳而沒有
+      對應清冊條目，不滿足此規則。
+    falsifiable_condition: >
+      一則清冊條目連續兩期審查都未變動，代表該揭露已經變成逃生口，本條對該條目失效——
+      不是「部分滿足」，是失效。清冊放在哪裡、什麼格式、多久審一次，留給採用專案自行決定；
+      本標準只要求「有這麼一份東西存在，且條目會動」。
+    shared_with: class-level-fix
+    rules:
+      - id: VE-012
+        trigger: "文件化的證據/涵蓋缺口（environment-stratification ⚠️/❌，或任何窄於其主張的檢查）沒有登記到帶到期日的例外清冊"
+        action: "揭露本身不滿足此規則——要求一則登記過、帶到期日的清冊條目，指名缺口與其覆核/到期日期"
+        priority: required
 
 physical_spec:
   type: checklist
@@ -189,5 +231,7 @@ physical_spec:
     - "證據若斷言「不存在」，是否已證明查詢工具執行成功（VE-008）"
     - "判斷存在與否的指令是否抑制了 stderr（VE-009）"
     - "exit_code 是否來自管線而非受測指令本身（VE-010）"
+    - "證據的執行時間戳是否晚於它所驗證對象的最後一次編輯（VE-011）"
+    - "文件化的證據/涵蓋缺口是否登記於帶到期日的例外清冊，而非只有揭露文字（VE-012）"
     - "Bug fix 是否有 RED → GREEN 循環證據"
     - "有外部服務依賴的 AC 是否標明 environment_layer"

package/bundled/core/class-level-fix.md

@@ -2,8 +2,8 @@
 
 > **Language**: English | [繁體中文](../locales/zh-TW/core/class-level-fix.md)
 
-**Version**: 1.0.0
-**Last Updated**: 2026-08-10
+**Version**: 1.1.0
+**Last Updated**: 2026-08-14
 **Applicability**: Any defect fix, in code or in configuration
 **Scope**: universal
 
@@ -83,6 +83,12 @@ Do this **per sub-set**, not in aggregate. A check over five lists that was only
 
 一道類別層檢查在被信任之前，必須先被證明不是空跑：塞一個違反規則的合成成員 → 確認檢查失敗**且指名該成員** → 移除後確認回到綠燈。**逐子集做，不要整體做**——一道涵蓋五份清單、卻只對第一份測過的檢查，是一道涵蓋一份清單的檢查。
 
+### What a passing negative control does not prove
+
+A negative control that passes once demonstrates only that **one** known-bad case reaches the checker's failure path. **It does not demonstrate** that the checker recognizes **every** violation of the rule it purports to guard. A grep-based gate can fail-closed perfectly while guarding a spelling, not a behavior — the synthetic member proves the wire is connected, not that the net is wide enough to catch what it claims to catch.
+
+一次通過的負向控制，只證明**一個**已知壞案例能到達檢查器的失敗路徑。**它不證明**該檢查器認得它所宣稱守護的規則的**每一種**違反。一道 grep 閘可以 fail-closed 得很完美，卻守著一個拼字而不是一個行為——那個合成成員證明的是線路接通了，不是那張網夠寬，足以抓住它自稱要抓的東西。
+
 ---
 
 ## Worked examples (measured, 2026-08-10)
@@ -121,6 +127,22 @@ In that case the workable form is **per-inventory**: one check per declaration k
 
 ---
 
+## Narrow Coverage Must Be Registered, Not Just Disclosed
+
+When a gate's actual coverage is narrower than the rule it serves, writing a sentence that says so is not enough. A prose disclosure is cheap — cheaper than widening the gate — and every narrow gate that gets reviewed once grows an honest paragraph and then stays narrow forever. **Disclosure earns nothing on its own; it earns something only paired with a mechanism that can prove it didn't just become a permanent excuse.**
+
+**Requirement**: any documented coverage gap of this kind must also be registered in a dated exception inventory — a list, external to the standard prose itself, that names the gap, states why it exists, and carries a review or expiry date. A disclosure with no entry in such an inventory does not satisfy this rule.
+
+**Falsifiable condition**: if an entry sits unchanged across two consecutive inventory review cycles, the disclosure has become an escape hatch and this rule is violated for that entry — not "partially satisfied", violated. The inventory mechanism itself (its location, format, and cadence) is left to the adopting project; this standard requires that one exist and that entries move, not that it take any particular shape.
+
+當一道閘門的實際涵蓋面窄於它所服務的規則時，只寫一句話說明是不夠的。散文式揭露很便宜——比擴大涵蓋面便宜得多——於是每一道被審過一次的窄閘門都會長出一段誠實的文字，然後永遠維持窄下去。**揭露本身不換來任何東西；只有配上一個能證明它沒有淪為永久藉口的機制，它才換得到東西。**
+
+**要求**：這一類已記錄的涵蓋缺口，必須同時登記到一份**帶到期日的例外清冊**——一份獨立於標準本文之外的清單，指名缺口、說明成因、並附上覆核或到期日期。沒有登記到這種清冊裡的揭露，不滿足本條。
+
+**可證偽條件**：若某條目連續兩期清冊審查都未變動，該揭露就已經變成逃生口，本條對該條目**失效**——不是「部分滿足」，是失效。清冊機制本身（放在哪裡、什麼格式、多久審一次）留給採用它的專案自行決定；本標準要求的是「有這麼一份東西存在，且條目會動」，不是要求它長成特定形狀。
+
+---
+
 ## Anti-patterns
 
 | Anti-pattern | Why it fails |
@@ -130,6 +152,7 @@ In that case the workable form is **per-inventory**: one check per declaration k
 | A check that lists its own scope | Correct until someone adds the fourth member |
 | Testing the class check against one member | Proves that member, not the class |
 | `✓ all pass` with no denominator | Identical output whether it scanned everything or nothing |
+| A disclosure sentence with no entry in a dated exception inventory | Cheaper than widening the gate; nothing forces it to ever change |
 
 ---
 
@@ -157,5 +180,5 @@ Applying this standard is a judgement made at the moment of fixing. There is no
 
 ## Relationship to other standards
 
-- [verification-evidence](verification-evidence.md) — evidence validity: a tool can fail silently and its output is indistinguishable from a real result. This standard is the same concern applied to *scope* rather than to *execution*.
+- [verification-evidence](verification-evidence.md) — evidence validity: a tool can fail silently and its output is indistinguishable from a real result. This standard is the same concern applied to *scope* rather than to *execution*. It also shares the narrow-coverage-must-be-registered requirement (VE-012) with this standard's own rule of the same shape.
 - [anti-hallucination](anti-hallucination.md) — that one guards "did not check"; this one guards "checked one of many and reported on all".