PyCForge 1.3.63 — Source housekeeping and conversion-import completion
Verified 2026-10-28

SCOPE
Built from the documentation-cleaned 1.3.62 maintenance source. This is a new
1.3.63 application release. No package has been published; CC controls release.

SOURCE PRESERVATION
Excluded 623 regenerable cache/packaging files (13,590,467 bytes) from the
maintained working copy. All 3,677 non-generated baseline files remain;
18 received intentional edits for versioning, completion, current guidance
or release verification. No historical evidence was rewritten or discarded.
Tests, fixtures, specifications, SDK framework requirements, training material,
transition records, licenses, Runtime headers, application icons and all Ting
authoring inputs are retained. Source-only Ting specifications stay in source.
Current HTML Help remains authoritative; obsolete PDFs/generator stay excluded.
Added dry-run/apply housekeeping and release rejection of generated caches.
Setuptools-generated PKG-INFO/egg-info inside the sdist are required metadata.

EDITOR BEHAVIOR
Enabled Ready conversion packages supply direct namespace/alias member names,
constants, positional arity, return types and escaped hover descriptions.
Math supplies its actual 35 functions and 5 constants, not the larger Python
math catalogue. Unary log is shown accurately. Unknown, disabled, unusable,
duplicate or rebound conversion aliases never fall back to Python libraries.
Nested call argument positions, CRLF, Unicode suffix replacement, both source
panes, keyboard/mouse insertion, undo, stale requests and Escape were tested.
Ordinary Python assistance retains Jedi behavior. Conversion assistance uses
the existing background process, 350 ms automatic debounce, 500,000-character
analysis bound and four-second request deadline. Registry state is read afresh;
dialog refresh cancels pending/displayed suggestions. No package/source runs.

ENGINE / SDK BOUNDARY
No conversion semantics or public authoring API changes. SDK 0.2.0 is unchanged.
The engine-owned API/reference inventory identifies 1.3.63. Installed SDK 0.2.0
refreshed its prior 1.3.62 cache, browsed the reference and rejected stale math
verification evidence without overwriting an existing artifact. It authored,
natively verified (2 fixtures), built and registered a custom wave_ops import.
The installed engine, without the SDK installed, completed its wave_root call.
A separate installed Qt window exercised custom completion insertion and undo.

VERIFICATION
Full collection: 3,415 tests; 3,343 passed, 72 skipped, 0 failures.
Also 4,886 passing subtests. Four disjoint groups were reconciled against the
complete collection: no omissions or duplicates. All group exit codes were 0.
Follow-up: 16 housekeeping/distribution tests and 14 subtests passed after
normalizing generated setup.cfg whitespace and recognizing its exact inert
setuptools-generated content in the archive. Missing or modified implicit
configuration is rejected. No application files changed after suite launch.
Skips cover unavailable sealed predecessor archives, unconfigured PCFC11_PATH,
absent clang and one test requiring PySide6 to be absent. Exact reasons and
per-test outcomes are retained in evidence/completion_1_3_63/reconciliation.json.
Current documentation: 30 pages, 81 Python examples, 49 Runtime and 32 Typed
native comparisons, 5 API examples, 4 C callers, 3 module and 3 README examples.
README renderer: four approved image links/order/widths preserved; only release
heading and the relevant completion feature sentence changed.

PERFORMANCE — isolated worker round trips, Linux / Python 3.12.14
Warm medians exclude process startup and the 350 ms automatic debounce.
Timings vary with host, load, buffer structure and package metadata size.
Case | cold ms | warm median ms | worst ms | maximum Qt timer gap ms
ordinary Python | 692.25 | 60.81 | 692.25 | 6.04
math member | 443.7 | 1.89 | 443.7 | 5.8
13k source | 410.12 | 14.74 | 410.12 | 5.79
130k source | 551.44 | 203.97 | 551.44 | 5.28
480k source | 1834.95 | 1280.99 | 1834.95 | 5.81
480k incomplete call hints | 1753.99 | 1316.6 | 1753.99 | 8.89
64 registered packages, 128 functions + 128 constants each | 798.68 | 397.02 | 798.68 | 8.2
480k source + 64 registered packages | 2504.91 | 1817.55 | 2504.91 | 8.36

All benchmark requests completed within the deadline; no worker errors.
The Qt timer ran every 5 ms; largest observed gap was 8.89 ms. This measures
event-loop availability, not end-to-end rendering latency. Large buffers remain
slower; structurally incomplete buffers may need the recovery parser.

LIMITS
Completion covers direct conversion namespace members/calls outside strings
and f-strings. Generic value/arg1 labels mean positions, not keyword names.
Completion is assistance, not proof that a complete program converts. Scalar
Runtime bounds and all existing unsupported math/NumPy/general-API gates remain.
Native/Qt checks ran on Linux with Qt offscreen. This is not a new Windows,
macOS, desktop-compositor or PCFC11 execution qualification.

WHEEL QUALIFIED
pycforge-1.3.63-py3-none-any.whl
Bytes: 2450649
SHA256: de3babf2bb4f150d9ca34b7af7680a4fe38b1ef56adbc49a09b57b372223b636

The separate final handoff report records the exact source-archive hash and
final distribution/twine checks; an archive cannot contain its own final hash.
Raw test, timing, preservation, installed SDK and UI records accompany source.
Historical records keep their original version and verification scope.
