PyCForge 1.3.82 / PCFC11 0.2.30 qualification record
Updated 2026-11-03 (America/Chicago). Publication remains CC's responsibility.

OUTCOME AND SCOPE

Direct statement labels and goto now participate in the bounded shared integer
CFG. Entry facts, dead predecessors, nested entry, loops and irreducible cycles
are covered. Constant simplification retains externally entered labeled regions.
Existing constant/copy/known-bit/value-numbering analyses consume this graph.
Switch/fallthrough, computed jumps, expression-contained transfers, reaching
definitions and scalar SSA are not new capabilities. Liveness remains separate.
No analysis budget or integer-IR bound was increased.

WHOLE_PROGRAM_QUALIFICATION.md is the maintained acceptance procedure. Its
runner identifies exact inputs, converts freshly, checks reviewed diagnostics,
compiles with independent compilers, compares independent expected outcomes and
retains execution evidence. Wrong expectations and failed conversion cannot
produce a passing report. Existing work directories are refused without damage.
Finite checks qualify recorded applications and inputs, not arbitrary Python.

README now opens with the entire-supported-program capability, including supplied
modules. The original PyCForge banner, two screenshots, Crucible banner/subtitle
and one SDK page link are retained: exactly five authored URLs. Heading is 1.3.82;
historical screenshot labels remain 1.3.39. No release/test status was added there.

COMPILER IDENTITY AND RESULTS

Pinned archive: pycforge/pcfc11/PCFC11-0.2.30-source.zip
Bytes: 5,982,020; files: 825.
SHA256: 7f67dd72bef137a7bce472d5bb25a99a7a7c55cd91d9e96533ffa2be57c6bba7
All archive CRC and SHA256SUMS entries verified. Only irbuild.c and irflow.c
changed among compiler translation units; ir.h/bounds remain unchanged.

- Standard Linux feedback: 40/40 phases passed.
- Bootstrap: 33/33 steps, including inherited O0/O1, self-compilation and 30 suites.
- Compiler sanitizers: build plus 30/30 suites; ASan/UBSan/float-cast-overflow,
  with LeakSanitizer disabled on this host.
- Additional references/tooling: 9/9 gates passed.
- Direct labels: 94 functions x 64 inputs x 3 outputs = 18,048 outputs per
  native/Windows-model O0/O1 configuration, with 4,864 independent path results.
  Five optimized-assembly assertions per target confirm useful fact propagation.
- Direct builder/refusal: 4,285 checks per profile, including fixture-construction
  assertions; GCC O0/O2, PCFC11 O0/O1 and ASan/UBSan profiles pass.
- Unchanged PCFC11 0.2.29 passes the semantic label corpus too: this change
  broadens optimization admission, not accepted C syntax.
- Five alternating compilation-cost samples are retained. Timings on a busy
  shared host do not support a broad performance claim.

WHOLE-PROGRAM AND PRODUCT RESULTS

- Exact final pinned source rebuilt and verified before fresh qualification.
- 53 fresh source bundles / 265 native executable profiles pass: PCFC11 O0/O1,
  GCC O0/O2 and a GCC UBSan reference. Independent outcomes accompany all four
  complete applications; eight BuildService executions and twelve failed/stale/
  canceled build-preservation checks pass.
- 106 Windows-target assembly-generation checks pass; no native Windows claim.
- 39 PYC4699 Runtime development-preview diagnostics are retained and reviewed.
  All 265 compiler invocations have empty compiler stderr. No warning was
  suppressed or described as a formal proof of Runtime semantics.
- Qualification fault injection checks wrong/correct outcomes, existing-directory
  preservation and diagnostic retention on rejected source.
- Product/documentation gates: 8/8 pass, including Ting, phase16 promotion,
  documentation, README, SDK landscape, compileall, focused Ruff and pip check.
  Documentation covers 32 Help pages, 82 Python examples, 50 Runtime and 32 Typed
  native comparisons, five API examples, six C callers and three module examples.
- Installed wheel product checks verify all 997 product files byte-for-byte,
  SDK exports, isolated imports, managed compiler build/identity and Typed plus
  direct-label program execution at O0/O1. Source-distribution rebuild/installed
  checks and final artifact hashes are recorded in the accompanying release report.
- Installed Linux offscreen UI smoke confirms title/About 1.3.82 and 32 Help
  pages across five categories. This is not native Windows display acceptance.
- README/public-distribution focused regression: 11 tests pass after the wording
  update. Final artifact validation checks the embedded README and payloads.

COMPLETE APPLICATION SUITE: NOT CLEAN

The unfiltered completed run collected 3,941 cases in 2,365.73 seconds. Pytest
reported 57 failed, 3,845 passed, 39 skipped, two warnings and one setup error.
JUnit contains 56 failed cases (one with two failure reports) plus one error case.
The raw failure result is retained and is not replaced by focused repeats.

Of these 57 affected cases:
- 43 also fail in the selected unchanged 1.3.81 baseline run. These include real
  Qt completion/help/async/save-close/lifetime issues and stale inventory checks.
- 11 memory/JSON/NumPy cases pass on both versions with ASan/UBSan active and
  detect_leaks=0. LeakSanitizer cannot inspect the process task directory on this
  host; raw stderr is retained. Leak detection is not qualified here.
- 2 SDK cases reproduce missing lxml/jsonschema on both versions and pass on
  both after installing those review dependencies. Product dependencies did
  not change. The source release procedure now lists the required review tools.
- 1 completion-popup setup error passes focused repeats on both versions and
  also appears in the historical baseline full run. Order/lifetime cause is open.

The 39 skips are 20 unavailable Clang cases, 18 unavailable optional sealed
predecessor-artifact checks and one requiring PySide6 absence. No current test
was removed, skipped or weakened to hide a failing result. Full-suite product
hashes remained unchanged while it ran. No new compiler regression was observed
within the checked scope; there is no full application or cross-platform all-clear.

ENVIRONMENT AND RETAINED ATTEMPTS

Linux x86-64, Python 3.12.14, GCC 13.3.0 and PySide6 6.12.0. Review environment
uses NumPy 2.3.5, Numba 0.63.1 and llvmlite 0.46.0 after an initially downloaded
truncated LLVM shared library caused collection failure. The replacement's
recorded size/hash were verified. This did not change application dependency
constraints. Full dependency inventories and the original failure are retained.
Two earlier incomplete full-suite attempts are labeled incomplete, not passes.
The initial compiler bootstrap overlap failure and direct-fixture corrections,
provisional compiler pin, rejected qualification attempts and initial offscreen
Help-count mistake remain available under their original attempted identities.

Native Windows 0.2.30 execution, PE/UCRT/unwind behavior, display acceptance,
external BLAS and general third-party Python compatibility are not established.
The inherited Csmith reference timeout remains explicit in compiler evidence.

SOURCE, PACKAGING AND NEXT OBJECTIVES

Historical evidence and superseded compiler sources stay in the full development
ZIP. The installable gz carries the current compiler, maintained tools/tests and
this procedure; the wheel carries the runtime product/current compiler. Neither
upload accumulates top-level history. Final sizes/hashes and parity results are
in PyCForge-1.3.82-Release-Report.txt beside the artifacts, avoiding self-hashes.

Evidence root: evidence/pcfc11_labels_1_3_82. Consult full-review-summary.json,
final-pinned-whole-program/qualification.json, diagnostic-review.json, compiler
standard-feedback, bootstrap/sanitizer/reference summaries and installed reports.
The pinned compiler's docs/VALIDATION.md and docs/CFG_LABELS.md explain its scope.

Next: resolve recorded app qualification gaps and obtain native Windows evidence;
then admit switch/fallthrough conservatively and add bounded reaching-definitions/
use chains before considering SSA. Every extension must retain independent
whole-program outcomes, memory checks and conservative fallback.
