PyCForge SDK 0.7.0 — Conversion coverage discovery
Release verification — 2026-11-04 (America/Chicago)

SDK client: 0.7.0. Selected engine: unchanged PyCForge 1.3.88.
PCFC11: unchanged pinned 0.2.34. Nothing published; only CC publishes.

IMPLEMENTED
SDK.coverage(profile=None), the coverage CLI command and the desktop's
Engine reference / Conversion coverage tab share one public report. Developers
and AI tools can discover engine-reported families, exact authoring profiles,
operation choices/signatures, limits and starters. Current PyCForge 1.3.88
reports four families, six compatibility profiles and eight starters.
Scalar profile versions are not counted as extra conversion families. Profiles
using authored fields are not misreported as having zero operations.

The optional profile argument selects an exact advertised declaration schema.
Unknown profiles return SDK_COVERAGE_PROFILE with alternatives and a remedy;
bad arguments return SDK_ARGUMENT. Malformed/unsupported coverage metadata
returns SDK_COVERAGE_REFERENCE with a JSON pointer and remedy, and cannot
replace the last compatible cache. Future optional metadata stays inert.
Older engines retain their available forms/starters with family coverage
explicitly not reported. Offline browsing never approves execution or packaging.
No template or native fixture executes during browsing.

methods() now includes text_operations and authoring_coverage;
describe("TextFunction") includes text operation signatures and explanations.
No conversion kernels, engine dispatch or native compiler changes are included.
The client remains independent from Qt and engine imports; GUI is optional.

The README receives only current feature/usage and version changes. Its single
approved banner URL, placement and width remain intact. The four-family table
renders correctly, and the public image returned HTTP 200 with valid PNG bytes.
Local first-run Help and API documentation explain the new browser and results.
Existing visual-provenance assets retain their identities. Reverify projects
after the SDK upgrade; old reports are not silently promoted.

VERIFICATION
- 158 installed headless tests passed; no skips, failures or errors. Imports
  were checked against the isolated SDK installation, which has no Qt or engine.
- 30 installed Linux offscreen desktop tests passed; no skips or failures.
  Search, exact text signatures, inert text, offline labels and existing worker,
  readiness, cancellation, save, startup and file-dialog behavior are covered.
- All eight starters passed through a separately installed PyCForge with both
  PCFC11 and GCC: 125 native fixtures and 87 expected conversion rejections
  per compiler. Scan, verification, readiness, repeat byte-identical packaging,
  explicit registration and preservation on stale edits passed.
- SDK-upgrade exercise: 0.6.4 evidence refused, old package bytes preserved,
  then explicit verification and packaging with 0.7.0 succeeded.
- Actual installed PyCForge 1.3.60 and 1.3.61 compatibility passed. Catalogue-only
  and legacy authoring behavior remain distinct. Synthetic future metadata and
  malformed refresh checks supplement, rather than replace, those real versions.
- The completed PyCForge 1.3.88 application baseline preceded this SDK review:
  3,930 passed and 39 skipped, followed by 85 Help closeout checks. Every one
  of the 999 installed engine product files still matches the final 1.3.88
  wheel. No new full application run is claimed for this client-only release.
- New coverage-focused development checks: 17 headless and two GUI checks
  passed before the final full suites. Visual review prompted visible scalar
  version labels and a capability-first overview. Initial test-environment
  setup lacked pytest; dependencies were installed before the final runs.

PRESERVATION AND DISTRIBUTIONS
All 1017 predecessor source-distribution paths are retained.
297 historical evidence/report files are unchanged byte-for-byte.
Only new rebuildable caches/native objects are excluded, with recorded hashes
and reasons in evidence/coverage_0_7_0/rebuildable-omissions.json.
The separate release report records final artifact sizes/hashes, exact installed
payload checks, strict Twine and archive checks, and source-to-wheel parity.

BOUNDARIES
These are Linux native/offscreen results. Native Windows execution and an
unaided human or independent AI author trial remain pending. This client release
does not expand the engine's conversion semantics, add mixed-family packages,
or establish general plugin stability. Browsing success is not native approval.

REPRODUCTION
Use separate installed SDK and engine environments. The SDK test source contains
tools/run_installed_tests.py, tools/qualify_sdk_starter.py and
tools/qualify_sdk_upgrade.py. Set PYCFORGE_TEST_ENGINE to the 1.3.88 interpreter,
PYCFORGE_TEST_OLD_ENGINE to installed 1.3.60 and PYCFORGE_LEGACY_ENGINE to 1.3.61.
Run headless tests without Qt and tests_gui separately with QT_QPA_PLATFORM=offscreen
on Linux. tools/prepare_coverage_release.py accounts for evidence; the final
pair audit uses tools/verify_sdk_distribution.py with the 0.6.4 tar.gz baseline.
Full raw SDK qualification is under evidence/coverage_0_7_0. The selected engine
remains authoritative for all conversion contracts and capabilities.
