release: v0.11.2 (at-Tawbah / G11.0.1 Python verifier corrective)

Phase G11.0.1 (at-Tawbah) closes the genuine prospective
audit gaps the v0.11.0 + v0.11.1 ship did not absorb (per
Round 28 empirical state audit), plus closes Round 28 finding
F18 (flagship: backfill_github_releases.py --dry-run silently
no-op'd).

Round 28 state audit found that C-1, H-5, H-6, and M-7 were
already backported to the unified Python verifier in v0.11.0
via the G11.1 commits ee656ad + c6b219a. The genuine v0.11.2
work is:

- T01: extend SAFETY_INVARIANTS.md Invariant 6 step 6 to
  document CASM-V-032 / CASM-V-035 / CASM-V-036 explicitly.
- T03: H-4 nested-generic recursion in Python signature
  canonicalizer (the Python equivalent of the G11.1 amended_4
  T03 rules 6 + 7). The pre-v0.11.2 implementation fell
  through to ast.unparse(node.slice) for multi-arg generic
  parameters, producing tuple-stringification artifacts; this
  release recurses element-wise on Tuple slices and flattens
  inner unions across the Subscript-Union boundary.
- T06: M-5 corrective applied to step5 docstring (references
  Manifest.from_dict as canonical enforcement site); M-6
  corrective applied to SECURITY.md N2 disclosure wording
  (now names --expected-identity and CASM-V-035 explicitly).
- T08 (flagship F18): scripts/backfill_github_releases.py
  rewritten with argparse; --dry-run / --apply mutually-
  exclusive required; default no-flag exits 2 with usage.
  F19 incidental closure: version filter excludes v0.11.1+
  workflow-managed releases via _WORKFLOW_MANAGED_FLOOR.

Local verification (sandbox):

- pytest --collect-only -q -> 589 tests collected
  (target: 589, +13 from v0.11.1's 576: 8 in
  test_python_signature_h4.py, 5 in test_backfill_dry_run.py)
- pytest -q -> 543 passed, 46 skipped
- ruff check + ruff format --check + mypy clean
- furqan-lint version -> 'furqan-lint 0.11.2'
- Em-dash check (extended scope) clean
- release_sweep.py: clean (README pins synchronized to
  v0.11.2)

After v0.11.2 the Python verifier's signature canonicalization
matches the Rust verifier's amended_4 T03 rules 6 + 7. The
remaining amended_4 defenses (C-1 identity policy, H-5
trusted_root threading, H-6 checker_set_hash Form A, M-7
typed identity-extraction errors) were already shared between
the two pipelines from v0.11.0; v0.11.2 acknowledges that and
pins the audit register state.

Co-authored-by: Claude (cowork) <claude+cowork@anthropic.com>

Author: BayyinahEnterprise

CHANGELOG.md

@@ -19,6 +19,121 @@ introduced this convention.
 
 ---
 
+## [0.11.2] - 2026-05-09
+
+### Substrate corrective (at-Tawbah / G11.0.1)
+
+Phase G11.0.1 closes the two genuine prospective audit
+gaps the v0.11.0 + v0.11.1 ship did not absorb (per Round 28
+empirical state audit), plus closes Round 28 finding F18
+(flagship: backfill_github_releases.py --dry-run flag silently
+no-op'd). The other amended_4 prospective findings (C-1, H-5,
+H-6, M-7) were already backported to the unified Python
+verifier substrate via the G11.1 commits ee656ad +
+c6b219a; this release acknowledges that and pins the
+remaining two gaps explicitly.
+
+#### Closures
+
+- H-4 HIGH (Python signature canonicalization): the Python
+  `_canonical_type_string` fell through to
+  `ast.unparse(node.slice)` for multi-argument generic
+  parameters, producing tuple-stringification artifacts like
+  `Dict[(str, int)]` instead of `Dict[str, int]`. v0.11.2
+  recurses element-wise on the Tuple slice and flattens
+  inner unions across the Subscript-Union boundary so
+  `Union[Optional[T], List[U]]` canonicalizes consistently.
+  Cross-language symmetry with the Rust verifier's amended_4
+  T03 rules 6 + 7 is now restored. (T03)
+- F18 HIGH (Round 28 flagship): scripts/backfill_github_releases.py
+  now requires an explicit `--dry-run` (preview only) or
+  `--apply` (real `gh release create` calls) flag. The default
+  no-flag invocation exits 2 with argparse usage rather than
+  silently mutating live GitHub state. The structural-honesty
+  thesis applied to the project's own scripts: the documented
+  surface claim ("--dry-run is preview mode") now matches the
+  substrate behavior. (T08)
+
+#### Closures from amended_4 prospective defenses (no-op acknowledgement)
+
+The following corrections were already applied in v0.11.0
+via the G11.1 commits ee656ad + c6b219a; this entry pins the
+audit register state explicitly:
+
+- C-1 CRITICAL (verifier UnsafeNoOp default): closed in v0.11.0
+  by `verification.step6_verify_sigstore`'s refuse-without-
+  policy default raising CASM-V-035; cross-applied to both
+  Python and Rust pipelines via the unified verifier.
+- H-5 HIGH (trusted_root threading): closed in v0.11.0 by
+  `Verifier(_inner=trusted_root)` lower-level constructor.
+- H-6 HIGH (checker_set_hash placeholder): closed in v0.11.0
+  by `compute_checker_set_hash()` Form A substantive hash
+  over pinned source-file tuple.
+- M-7 MEDIUM (string-sentinel identity extraction): closed
+  in v0.11.0 by `_extract_identity` raising CASM-V-036
+  typed errors.
+
+#### New corrective work this release
+
+- M-5 MEDIUM: `step5_canonicalize_manifest` docstring now
+  references `Manifest.from_dict` as the canonical
+  enforcement site; redundant in-step enforcement prose
+  removed. (T06)
+- M-6 MEDIUM: SECURITY.md Newman 2022 N2 disclosure wording
+  now names `--expected-identity` and CASM-V-035 explicitly,
+  closing the surface-substrate gap. (T06)
+- F19 LOW (incidental): backfill version-filter excludes
+  v0.11.1+ workflow-managed releases via the new
+  `_WORKFLOW_MANAGED_FLOOR = (0, 11, 1)` constant in
+  `scripts/backfill_github_releases.py`. (T08 side closure)
+
+#### Substrate canonical-authority extension
+
+- SAFETY_INVARIANTS.md Section "Invariant 6" step 6 now
+  documents CASM-V-032 (identity policy mismatch),
+  CASM-V-035 (no identity policy supplied), and CASM-V-036
+  (identity extraction failure) explicitly per amended_4
+  T05 specification. (T01)
+
+#### Deferrals
+
+- F4 ONNX empty-graph: Phase G11.3 scope.
+- F5 Release-vs-tag QUESTION: QUESTIONS.md.
+- F10 SBOM: Phase G11.4 or v1.5.
+- F12 furqan PyPI dep: declined Shape C.
+- F17 framework section 4 amendment: framework-side.
+- F20 sandbox-CI parity: separate release-checklist amendment.
+- F21 PyPI CDN cache transient: separate runbook amendment.
+
+### Tests
+
+Test count: 576 (v0.11.1 ship state) -> 589 (v0.11.2).
+Net delta: +13 (8 in test_python_signature_h4.py covering
+the H-4 fixtures + inner-difference detection + tuple-value
+preservation; 5 in test_backfill_dry_run.py covering F18
+no-flag exit / both-flags rejection / no-subprocess-call /
+expected-version enumeration / F19 workflow-managed-floor).
+
+### Cross-language symmetry restored
+
+After v0.11.2, the Python verifier's signature canonicalization
+matches the Rust verifier's amended_4 T03 rules 6 + 7
+(nested-generic element-wise recursion). The other amended_4
+defenses (C-1 identity policy, H-5 trusted_root threading,
+H-6 checker_set_hash, M-7 typed identity-extraction errors)
+were already shared between the two pipelines from v0.11.0.
+
+### Round-28 closure ledger
+
+| Disposition | Count | Findings |
+|-------------|-------|----------|
+| Genuine v0.11.2 closure | 4 | H-4, M-5, M-6, F18 |
+| Acknowledgement (closed in v0.11.0) | 4 | C-1, H-5, H-6, M-7 |
+| Side closure | 1 | F19 |
+| Deferral | 7 | F4, F5, F10, F12, F17, F20, F21 |
+
+---
+
 ## [0.11.1] - 2026-05-08
 
 ### Tooling (al-Mubin / G10.5)

README.md

@@ -41,20 +41,20 @@ pip install "furqan-lint[rust,go,onnx-runtime,onnx-profile,gate11]"  # all adapt
 ### Install from a specific commit or tag
 
 ```bash
-pip install "git+https://github.com/BayyinahEnterprise/furqan-lint.git@v0.11.1"
+pip install "git+https://github.com/BayyinahEnterprise/furqan-lint.git@v0.11.2"
 ```
 
-Replace `v0.11.1` with any tag from the [release history](https://github.com/BayyinahEnterprise/furqan-lint/releases) or `main` for the development tip.
+Replace `v0.11.2` with any tag from the [release history](https://github.com/BayyinahEnterprise/furqan-lint/releases) or `main` for the development tip.
 
 ### Furqan dependency
 
-furqan-lint requires `furqan>=0.11.0`, the Furqan programming-language tooling. As of 2026-05-03 the PyPI release of `furqan` is at v0.10.1; install v0.11.1 directly from GitHub:
+furqan-lint requires `furqan>=0.11.0`, the Furqan programming-language tooling. As of 2026-05-03 the PyPI release of `furqan` is at v0.10.1; install v0.11.2 directly from GitHub:
 
 ```bash
-pip install "git+https://github.com/BayyinahEnterprise/furqan-programming-language.git@v0.11.1"
+pip install "git+https://github.com/BayyinahEnterprise/furqan-programming-language.git@v0.11.2"
 ```
 
-This GitHub-pin step will not be necessary once `furqan` v0.11.1 is published to PyPI.
+This GitHub-pin step will not be necessary once `furqan` v0.11.2 is published to PyPI.
 
 ### Rust support (opt-in)
 
@@ -485,7 +485,7 @@ jobs:
     runs-on: ubuntu-latest
     steps:
       - uses: actions/checkout@v4
-      - uses: BayyinahEnterprise/furqan-lint@v0.11.1
+      - uses: BayyinahEnterprise/furqan-lint@v0.11.2
         with:
           path: src/
 ```
@@ -505,7 +505,7 @@ Python files:
 # .pre-commit-config.yaml
 repos:
   - repo: https://github.com/BayyinahEnterprise/furqan-lint
-    rev: v0.11.1
+    rev: v0.11.2
     hooks:
       - id: furqan-lint
 ```
@@ -540,7 +540,7 @@ repos:
       - id: mypy
 
   - repo: https://github.com/BayyinahEnterprise/furqan-lint
-    rev: v0.11.1
+    rev: v0.11.2
     hooks:
       - id: furqan-lint
 ```
@@ -614,7 +614,7 @@ MARAD  example.py
 Per-version closure ledgers are in
 [CHANGELOG.md](CHANGELOG.md). This README previously
 mirrored closures from v0.2.0 through v0.11.0; that mirror
-was retired in v0.11.1 (Phase G10.5 al-Mubin) because
+was retired in v0.11.2 (Phase G10.5 al-Mubin) because
 CHANGELOG.md is the canonical closure ledger and the
 README mirror was repeatedly drifting out of sync. The
 framework section 10.2 retirement procedure was followed:

SAFETY_INVARIANTS.md

@@ -282,7 +282,13 @@ The nine steps:
    no cache exists)
 5. Re-canonicalize the manifest (RFC 8785; Invariant 3)
 6. Verify Sigstore signature against the canonical bytes
-   (`CASM-V-030..034` by failure mode)
+   (`CASM-V-030..034` by failure mode); enforce the
+   configured Identity policy (`CASM-V-032` on identity
+   mismatch; `CASM-V-035` if no `--expected-identity` and
+   no explicit `--allow-any-identity` -- the default refuse-
+   without-policy state); on failed identity extraction from
+   the signing certificate raise `CASM-V-036` rather than
+   returning a string sentinel
 7. Compare `module_root_hash` to the on-disk module
    (`CASM-V-040` on mismatch)
 8. Compare `public_surface.names` to the live extraction;

SECURITY.md

@@ -76,8 +76,14 @@ residual disclosures apply to any consumer running
    not in the lint.
 2. **Typosquatting at the publish boundary.** Sigstore proves
    "identity X signed bytes Y at time T", not "identity X is
-   the legitimate maintainer". Relying Parties must pin both
-   the package name and the expected signing identity.
+   the legitimate maintainer". Relying Parties MUST configure
+   `--expected-identity <pattern>` for both the Python
+   verifier (v0.11.2+) and the Rust verifier (v0.11.0+); the
+   default `CASM-V-035` refuse-without-policy ensures
+   verification fails closed when no policy is configured.
+   Use of `--allow-any-identity` is the explicit opt-in to
+   `UnsafeNoOp` policy; presence in CI logs is itself a
+   Sulayman-Naml ADVISORY signal per Phase G11.A Strategy 7.
 3. **Rekor entry queryability and privacy.** The public Rekor
    log publishes manifest hashes and public-surface name lists
    unencrypted. Confidential codebases should sign to the

pyproject.toml

@@ -4,7 +4,7 @@ build-backend = "setuptools.build_meta"
 
 [project]
 name = "furqan-lint"
-version = "0.11.1"
+version = "0.11.2"
 description = "Structural-honesty checks for Python, powered by Furqan"
 readme = "README.md"
 requires-python = ">=3.10"

scripts/backfill_github_releases.py

@@ -1,12 +1,26 @@
 #!/usr/bin/env python3
-"""Phase G10.5 (al-Mubin) T03: one-time backfill of GitHub Release
-objects for v0.8.2 through v0.11.0.
+"""Phase G10.5 (al-Mubin) T03 + Phase G11.0.1 (at-Tawbah) T08:
+one-time backfill of GitHub Release objects for v0.8.2 through v0.11.0.
 
 Run manually after Phase G10.5 T01 + T02 land. After this script
 runs against the upstream repository, finding F1 (12 historical
 versions tagged on origin but absent from the GitHub Releases UI)
 closes.
 
+Phase G11.0.1 (at-Tawbah) T08 / F18 closure: this script REQUIRES
+an explicit ``--dry-run`` (preview only, no API mutation) or
+``--apply`` (real ``gh release create`` calls) flag. The default
+no-flag invocation now exits 1 with usage text rather than silently
+mutating live GitHub state. The Round 28 audit observed the
+runbook's ``--dry-run`` invocation creating 12 Release objects
+because the pre-v0.11.2 script silently no-op'd ``--dry-run``;
+v0.11.2 closes that structural-honesty gap loudly.
+
+Phase G11.0.1 T08 / F19 incidental closure: the version filter
+now excludes v0.11.1 and later because release.yml's T02 step
+(added in Phase G10.5 al-Mubin) creates Release objects on tag
+push. Backfilling them would race the workflow.
+
 The script:
 
   1. Reads CHANGELOG.md and enumerates every ``## [X.Y.Z]`` header.
@@ -37,16 +51,14 @@
 
 from __future__ import annotations
 
+import argparse
 import re
 import subprocess
 import sys
 from pathlib import Path
 
-
 _HISTORICAL_UNTAGGED_VERSIONS = frozenset({"0.2.0", "0.7.0"})
-_PRE_CHANGELOG_FORMAT_VERSIONS = frozenset(
-    {"0.7.3", "0.8.0", "0.8.1"}
-)
+_PRE_CHANGELOG_FORMAT_VERSIONS = frozenset({"0.7.3", "0.8.0", "0.8.1"})
 # v0.1.0 already has a Release object on the upstream repo and
 # is excluded from this backfill so the script does not attempt
 # to recreate it.
@@ -58,6 +70,11 @@
 # release.yml + Trusted-Publishing era and are not eligible
 # for retroactive Release-object backfill via this tool.
 _BACKFILL_FLOOR = (0, 8, 2)
+# Phase G11.0.1 T08 / F19 incidental closure: workflow-managed
+# release boundary. v0.11.1 was the first release whose Release
+# object was created by release.yml T02 (Phase G10.5 al-Mubin
+# ship); backfilling at v0.11.1+ would race the workflow.
+_WORKFLOW_MANAGED_FLOOR = (0, 11, 1)
 
 
 def _version_tuple(v: str) -> tuple[int, ...]:
@@ -82,6 +99,14 @@ def _extract_versions_from_changelog(
             continue
         if _version_tuple(v)[:3] < _BACKFILL_FLOOR:
             continue
+        # Phase G11.0.1 T08 / F19 incidental closure:
+        # workflow-managed releases (v0.11.1+) are intentionally
+        # excluded from backfill scope. The release.yml T02 step
+        # (added in Phase G10.5 al-Mubin) creates GitHub Release
+        # objects on tag push for these versions; backfilling
+        # them here would race with the workflow.
+        if _version_tuple(v)[:3] >= _WORKFLOW_MANAGED_FLOOR:
+            continue
         versions.append(v)
     return versions
 
@@ -103,16 +128,11 @@ def _extract_notes(version: str) -> str:
     notes = proc.stdout
     if not notes.strip():
         # Per S2 calibration: empty output (NOT skeletal) -> fallback.
-        notes = (
-            f"Patch release. See `git log v<previous>..v{version}` "
-            f"for changes."
-        )
+        notes = f"Patch release. See `git log v<previous>..v{version}` " f"for changes."
     return notes
 
 
-def _create_release(
-    tag: str, notes: str
-) -> tuple[bool, str, str | None]:
+def _create_release(tag: str, notes: str) -> tuple[bool, str, str | None]:
     """Run ``gh release create``; return (created_ok, status, blocked_reason).
 
     ``status`` is one of ``created`` / ``already_exists`` /
@@ -141,9 +161,52 @@ def _create_release(
     return False, "blocked", result.stderr.strip()
 
 
-def main() -> int:
+def _build_arg_parser() -> argparse.ArgumentParser:
+    """Phase G11.0.1 T08 / F18 closure: argparse with required
+    mutually-exclusive --dry-run / --apply group.
+
+    No-flag invocation exits 1 with usage rather than silently
+    no-op'ing the documented --dry-run flag and mutating live
+    GitHub state.
+    """
+    p = argparse.ArgumentParser(
+        prog="backfill_github_releases.py",
+        description=(
+            "One-time backfill of GitHub Release objects for "
+            "historical tags. REQUIRES explicit --dry-run or --apply."
+        ),
+        epilog=(
+            "REQUIRED: --dry-run for preview OR --apply to execute. "
+            "No silent-execute; the substrate's surface claim must "
+            "match its behavior."
+        ),
+    )
+    mode = p.add_mutually_exclusive_group(required=True)
+    mode.add_argument(
+        "--dry-run",
+        action="store_true",
+        help="Preview the backfill scope without API mutation.",
+    )
+    mode.add_argument(
+        "--apply",
+        action="store_true",
+        help="Execute the backfill (creates real GitHub Release objects).",
+    )
+    return p
+
+
+def main(argv: list[str] | None = None) -> int:
+    parser = _build_arg_parser()
+    args = parser.parse_args(argv)
     versions = _extract_versions_from_changelog()
-    print(f"Backfilling {len(versions)} versions: {versions}")
+    if args.dry_run:
+        print(f"DRY-RUN: would backfill {len(versions)} versions:")
+        for v in versions:
+            print(f"  v{v}")
+        print("\nNo API calls executed. Pass --apply to execute.")
+        return 0
+    # args.apply is True (mutually exclusive with --dry-run guarantees this).
+    print(f"APPLYING: backfilling {len(versions)} versions: {versions}")
     blocked: list[tuple[str, str]] = []
     created: list[str] = []
     skipped: list[str] = []
@@ -170,8 +233,7 @@ def main() -> int:
     )
     if blocked:
         print(
-            "\nBLOCKED versions (record in "
-            "docs/release-checklist.md Appendix A):",
+            "\nBLOCKED versions (record in " "docs/release-checklist.md Appendix A):",
             file=sys.stderr,
         )
         for tag, reason in blocked: