cpython-extensions 1.1.0 release audit
Date: 2026-08-21

RELEASE INTENT
--------------
1.1.0 is a minor-version milestone for performance evidence, structural
verification, documentation quality, and release consistency. It does not
silently change the production transformation runtime: the switch, inline,
goto, composition, and shared-verifier modules are unchanged from 1.0.4.

PUBLIC STATE
------------
- Distribution: cpython-extensions
- Import package: python_extensions
- Version source: src/python_extensions/_version.py = 1.1.0
- Supported interpreter: CPython >=3.13,<3.14
- Runtime dependency: bytecode>=0.17,<0.18
- License: Mozilla Public License 2.0 (MPL-2.0)
- Canonical repository: https://github.com/Karvp/cpython-extensions
- Stable release tag: v1.1.0
- Preview/rehearsal example: v1.1.0-beta

MILESTONE DELTA
---------------
- Adds 2..1,024-route switch scaling benchmarks for integer and string keys.
- Adds configurable benchmark intensity and raw 1.1.0 JSON evidence.
- Adds structural tests that keep specialized switch executable code bounded
  as route count grows.
- Adds intended-workload inline/goto evidence.
- Refines and synchronizes all active release-sensitive documentation.
- Extends tools/check_repo.py to derive current benchmark evidence names from
  the package version and validate all three current benchmark artifacts.
- Keeps historical 1.0.x release/certification evidence intact.

VALIDATION
----------
- Unit/regression suite: 378/378 PASS.
- CPython dev-mode, warnings-as-errors, third-party pytest plugin autoload
  disabled: 378/378 PASS.
- compileall across src/tests/tools/benchmarks/scripts: PASS.
- Reproducible artifact build: wheel/sdist/checksum manifest byte-identical
  across two fixed-epoch builds.
- Exact wheel install/smoke: PASS.
- Exact sdist suite: 378/378 PASS.
- Wheel rebuilt from exact sdist: byte-identical PASS.
- Full switch production differential: 3,680,000 calls PASS.
- Full switch adversarial: 4,703,400 calls PASS.
- Full strict-goto harness: 3,526,500 calls/yields PASS.
- Coordinated all-extension harness: 157,200/157,200 PASS.
- Inline optimizer + guarded-binding counted workload: 9,010,000 operations
  across randomized, threaded, unbound-local, crash-isolated, and guarded
  binding runs PASS.

PERFORMANCE EVIDENCE
--------------------
Committed 1.1.0 switch evidence demonstrates the expected scaling shape:
small native branches can win, while the portable table-backed direct-value
backend remains close to dict.get and avoids linear executable growth.
Selected integer speedups versus if/elif are 9.94x (64), 38.96x (256),
71.33x (512), and 142.90x (1,024). The 1,024 string case records 83.86x.

The intended-workload evidence records 1.34x for frozen small-helper inlining
and 2.47x for strict goto versus explicit state dispatch. The naturally
structured reference remains slightly faster than goto and that caveat is
published rather than hidden.

DOCUMENTATION AUDIT
-------------------
Reviewed/refined active documentation:
- README.md
- CHANGELOG.md
- CONTRIBUTING.md
- CODE_OF_CONDUCT.md
- SECURITY.md
- GITHUB_SETUP.md
- CITATION.cff
- benchmarks/README.md
- docs/ARCHITECTURE.md
- docs/COMPATIBILITY.md
- docs/COMPREHENSIVE_GUIDE.md
- docs/RELEASE_NOTES.md
- docs/RELEASING.md
- .github/REPOSITORY_METADATA.md
- .github/pull_request_template.md
- .github/ISSUE_TEMPLATE/bug.yml

Historical 1.0.x certification/audit files remain historical records and are
not rewritten to pretend they were produced for 1.1.0.

RELEASE WORKFLOW
----------------
The hardened tag-driven workflow remains in place:
- clean-tree and tag/version/channel validation;
- metadata preflight;
- deterministic double build and checksum comparison;
- exact wheel smoke test and exact-sdist full test/rebuild;
- API-driven idempotent GitHub Release handling;
- preview tags excluded from PyPI;
- stable tags enter the protected pypi environment;
- PyPI Trusted Publishing uses OIDC with id-token: write only in the publish
  job.

FINAL ASSESSMENT
----------------
PASS for the 1.1.0 source-tree milestone. Final wheel/sdist publication remains
conditional on the repository's normal CI/release workflow succeeding on the
v1.1.0 tag.
