TDS GUARANTEED STORAGE — PHASE 11 STATUS
Version: v3.5.3
Feature: Release Qualification

STATUS: LOCAL QUALIFICATION COMPLETE — RELEASED; REMOTE GATES AND PYPI PUBLICATION COMPLETE

FINAL RELEASE OUTCOME

- Pull request #1 merged to `main` as commit
  `84c253f2a7d68a20ddcab96e94cc107439ccdd32` on 2026-07-16.
- Immutable annotated tag `v3.5.3` resolves to that exact merge commit.
- Pull-request run 29499339585, merged-main run 29499946976, and tag run
  29500270923 completed the source, Python 3.10-3.14, Windows, macOS,
  Linux, native-extension, and distribution gates successfully.
- PyPI trusted publishing completed in tag-run attempt 2 at 2026-07-16
  13:38:22 UTC after the pending publisher was registered.
- Production project: https://pypi.org/project/staqtapp-tds/3.5.3/
- Published wheel: `staqtapp_tds-3.5.3-py3-none-any.whl`, SHA256
  `7c557c26247a6411705cf728e3bfa74793601bfd77a03c592c732d0403f9a9ac`.
- Published source: `staqtapp_tds-3.5.3.tar.gz`, SHA256
  `daf21ced4e88ae47f5f8145d9db01b22f9636645e33074dcea0a23a26611b554`.
- PyPI returned HTTP 200 for both uploads and accepted their trusted-publisher
  attestations.

COMPLETED IMPLEMENTATION

- Phase 10 controlled activation and lossless rollback are explicit and tested.
- Segment GC fails closed when any generation is invalid or unreadable.
- GC recomputes closed-world reachability before every candidate deletion and
  repeats the proof after the final fault-injection boundary.
- Candidate file type, device, inode, size, mode, modification time, and change
  time are revalidated before unlinking.
- Interrupted collection is idempotently resumable and exact byte accounting
  includes only files actually unlinked.
- Browser documentation uses 19 genuine, distinct page captures produced by
  selecting every real navigation control; the false overview was removed.
- The independent token-based PyPI workflow was removed. Publication now
  depends on Python, platform, native-extension, and distribution gates and
  uses PyPI trusted publishing.
- Every raw-descriptor TDS, generation, segment, migration-copy, and
  materialization writer now requests Windows binary mode. CRT newline
  translation can no longer change byte offsets, sizes, hashes, or GC evidence.
- Native Spiral scoring now preserves Python's arithmetic operation boundaries,
  preventing Apple Clang fused arithmetic from producing a one-ULP parity drift.
- Windows readers now validate immutable detached snapshots after releasing the
  source handle, preserving existing-reader stability across atomic replacement.
- The source archive contains the immediate-root Phase 6, 7, 8, 9, 10, and 11
  status records plus the cumulative audit record; Phase 10 is no longer absent
  between the Phase 9 and Phase 11 evidence.
- The Programmer Core API Guide opens with a rendered and visually verified
  three-page v3.5.3 supplement covering controlled activation, GC, the new
  public APIs, and the local/remote release boundary. The v3.1.23 API Surface
  PDF is now labeled historical rather than exhaustive for v3.5.3. All 634
  light-blue signature strips have corrected top clearance; every strip retains
  its signature text and none intersects the heading above.

EXACT LOCAL QUALIFICATION EVIDENCE

- Controlled activation tests: 10 passed / 0 failed.
- Complete `tests/test_v353_*.py` Guaranteed Storage group: 133 passed / 0 failed.
- Overlapping v3.5.3, release-workflow, Browser visual, and CSV monitor group:
  157 passed / 0 failed.
- Residual Windows atomic-reader, stress-matrix, CSV-byte, and persistence
  hardening group: 37 passed / 0 failed.
- Pure monolithic suite: 843 collected; 832 passed; 11 skipped; 0 failed in
  53.93 seconds after the final Windows reader repair.
- Opt-in native extensions built with GCC on local Linux x86_64 / Python 3.12;
  `_native_index` and `_csv_scan_kernel` both loaded from compiled modules.
- Native-active monolithic suite: 843 passed / 0 failed in 57.47 seconds after
  a fresh GCC rebuild; strict native/Python score equality also passed.
- The 129-generation incremental soak completed three torn-pointer recoveries,
  proved one 64-byte physical segment write per 4096-byte logical change,
  collected exactly 65 unreachable segments, and reverified current bytes.
- PEP 517 produced `staqtapp_tds-3.5.3-py3-none-any.whl` and
  `staqtapp_tds-3.5.3.tar.gz`; both passed `twine check`.
- Archive audit confirmed version 3.5.3, `Requires-Python: >=3.10`, corrected
  GitHub project URLs, pure `py3-none-any` wheel tagging, all 19 Browser images,
  all Phase status records, and no cache, bytecode, compiled binary, old false
  screenshot, or retired token-publish workflow.
- An isolated wheel install imported from the isolated target and exercised
  version identity, legacy qualification, explicit activation, current-
  generation verification, lossless rollback, destructive segment GC, and
  current-byte reconstruction successfully.
- The generated result-code documentation, source hygiene checker, workflow
  YAML parse, PDF text extraction, all-page PDF rendering, and PDF visual review
  passed. The structural PDF audit found 406 corrected 12-point strips and 228
  corrected 15-point strips, with 634/634 containing text and zero heading
  intersections.

REMOTE REVIEW QUALIFICATION

- The workflow is configured for Linux Python 3.10 through 3.14, Windows and
  macOS pure suites, Linux and macOS native-active suites, source hygiene,
  package validation, and tag-only trusted publication.
- Draft PR #1 exposed and fixed a source-gate bootstrap defect: generated result
  documentation now loads `result.py` without importing the uninstalled package.
- The next review run passed source hygiene, Linux Python 3.10 through 3.14,
  Linux native, and macOS pure qualification. It located Windows raw-descriptor
  text-mode corruption and the macOS native one-ULP score drift.
- Candidate `1aa12ed` made every non-Windows job green and reduced Windows from
  50 failures to 5. The residual evidence isolated an open-mmap replacement
  conflict plus a test comparison that normalized physical CSV line endings;
  both are corrected and locally requalified in this candidate.
- Final candidate `7042ed5` completed the entire pull-request matrix. The
  resulting merge commit completed the entire `main` matrix, and the exact tag
  completed the matrix again before trusted publication.

RELEASE OUTCOME

The original rule prohibited a tag or publication until every cross-platform
GitHub Actions gate was green and the resulting diff was reviewed. That rule
was satisfied: PR #1 was reviewed through its complete green matrix, merged,
requalified on `main`, tagged exactly once, requalified on the tag, and
published through the tag-only trusted-publisher job. Version 3.5.3 is now an
immutable released artifact; subsequent repository documentation changes do
not alter its tag or published files.
