PyCForge 1.3.85 / PyCForge SDK 0.6.4 / PCFC11 0.2.32
November 4, 2026 — unpublished development handoff

The SDK 0.6.3 reproducer with PyCForge 1.3.84 reported ready-to-package and built the same package bytes after the external GCC selection was replaced by an unavailable path. Fresh verification failed. The new optional Crucible native profile capability and public PCFC11 backend query close that specific gap in SDK 0.6.4. Direct GCC remains usable on earlier compatible engines. Use SDK 0.6.4 for backend-aware evidence.

The recorded PCFC11 backend includes the target, selected command, invocation path, resolved file and SHA-256. The selected absolute invocation path is supplied to each native compilation without changing the parent process environment. Twelve public-workflow observations cover the native default, relative selection, path change with identical bytes, same-path replacement, removal, recovery, changes during verification and packaging, old engine and driver refusal, and direct GCC compatibility. An earlier SDK report on the very same engine was rejected, the previous package preserved, and fresh verification restored readiness. The desktop independently showed Package enabled, disabled on removal and enabled after verification. Real native compiler failure, wrong fixture expectation and invalid JSON-mode command-line syntax have separately recorded diagnostics.

Application and compiler qualification

The final unfiltered application run accounts for all 3,947 collected cases. 3,908 passed, 39 skipped and zero failed. All 51 cases that failed on 1.3.83 passed in this complete run. I interrupted the earlier high-load attempt after it stopped making progress. It had terminal events for 3,925 cases and 4 failed reports, including a 30-second GCC compile timeout, a 3-second wait for a child process to appear and two lost test workers. Its incomplete event stream is preserved alongside the separate clean full rerun. The final run used four workers and a 180-second outer test watchdog. Product and fixture assertions were preserved. The skipped set comprises 20 unavailable Clang cases, 18 sealed historical artifacts and 1 test for an environment without PySide6. That last test passed separately without Qt. The full-accounting file lists each original outcome, skipped case and both run records.

The rebuilt pinned compiler passed 41 standard Linux phases, including 350 language cases. Its backend query passed explicit, empty and default selector tests across both supported target identities on this host. The 53-bundle whole-program qualification passed all 265 native execution configurations and 8 additional BuildService profiles. Windows-target assembly passed 53 cases at two optimization levels on Linux. Native Windows execution still requires platform evidence. The regenerated 40-case GC kit passed 80/80 checks with PCFC11 and 80/80 with GCC on Linux. The installed application verified all shipped product bytes, rebuilt the managed PCFC11 slot and ran the converted example at O0 and O1. All eight source gates passed, followed by rechecks of the final README and exported catalogue.

SDK client qualification

The final isolated SDK suite passed 141 headless tests and 28 Linux offscreen desktop tests. A preliminary reused test environment had conflicting installed engine metadata. The discovery assertion reproduced that mismatch and passed for the final clean install. The historically version-pinned inventory assertions now check actual engine and GUI contracts. Their recorded older results remain preserved. All seven current starter projects completed on PCFC11 and GCC, with 119 native and 82 expected rejection cases per compiler. All workflows preserved a successful artifact on a stale attempt. An independently authored namespace worked with records and vectors in a complete two-module program under five native configurations.

The associated operator review describes the active trust boundary and the remaining unaided first-project and native Windows work. The SDK does not attest a wrapper’s transitive dependencies, system headers or linked libraries. See the source evidence and the external delivery report for hashes and package parity. CC controls publication.
