Release validation — PCFC11 0.2.29

This checkpoint adds shared verified integer CFG topology, reachability,
reverse postorder and immediate dominators. Copy, constant and known-bit
analyses use the common snapshot. The AST builder's admitted control, backward
liveness, parser and code generators are unchanged. See CFG_TOPOLOGY.md.
CC alone publishes; this prepares an unpublished bundled source release.

| Gate | Completed result |
| --- | --- |
| Standard Linux feedback | 39/39 phases pass, retaining compiler, ABI, optimizer, application, Csmith and benchmark gates. |
| Topology and dominance | 1,114,172 direct checks per configuration; 2,197 exhaustive three-block graphs, 384 larger seeded graphs, three maximum-size fixtures. Independent Python/C node-deletion oracles. |
| Direct configurations | GCC O0/O2, PCFC11 O0/O1, Linux ASan/UBSan and Windows instruction-model O0/O1 all pass. Edge identities, unreachable regions, idoms, every dominance pair, exact budgets, allocations and immutable input are checked. |
| Existing analysis suites | Condition paths, copy, executable constants, known bits, arithmetic bits, value numbering and liveness remain passing. |
| Bootstrap | 32/32 steps: inherited O0/O1, O1 self-compilation/inherited tests and 29 suites through the self-compiled compiler. |
| Compiler sanitizers | 29/29 suites through ASan/UBSan/float-cast-overflow compiler; leak detection disabled. |
| Additional references | 9/9 gates: sanitized references, Windows constraints, toolchain paths, harness fault injection and Windows assembly preflight. |
| Fresh PyCForge 1.3.81 | 53 conversions pass 265 execution configurations; four applications pass eight BuildService builds/executions and twelve failure/stale/cancel output-preservation checks. |
| Windows target | The same 53 fresh conversions pass assembly generation at O0/O1 (106 checks); no native Windows claim. |
| Numerical applications | Current 104-array sum/mean matrix, complete application, supplied modules and independent C probes pass GCC O0/UBSan and PCFC11 O0/O1; all 208 scalar results satisfy existing rational bounds. |

Counts identify distinct kinds of checks; they are not one aggregate total.
The inherited Csmith reference-timeout skip remains explicit. Optional external
OpenBLAS/LAPACKE was not newly qualified. Older native Windows results retain
their exact versions. The new default runner has 39 Linux / 38 Windows phases.

#Compilation cost

Seven alternating samples per compiler/profile compare 0.2.28 and 0.2.29 on
parse.c, irconst.c, the existing control-flow corpus and a generated workload
of 32 functions with 48 compound conditions inside each loop. The same sources,
include paths, O0/O1 flags, raw elapsed times and Linux peak RSS are retained.
O1 current/baseline median ratios: parse.c 0.977, irconst.c 0.869, control-flow.c 0.967, topology-cost.c 0.951.
These shared-host observations showed no sampled slowdown but do not establish
a general performance gain. Some reference checking overlapped early samples.
The extra topology/dominance work shares existing budgets and may cause earlier
conservative fallback; no budget was increased.

#Limits and retained attempts

Native Windows 0.2.29 execution is pending. Windows instruction models retain
the existing stack-probe stub and do not qualify PE/UCRT, SEH, virtual unwinding
or guard pages. The initial new direct sanitizer run reached LeakSanitizer's
unsupported process-inspection path in this environment; the corrected run
used the documented detect_leaks=0 with ASan/UBSan active. Both logs remain.
No LeakSanitizer pass is claimed. A first make invocation ran from the scratch
root and had no makefile; the subsequent compiler-directory build passed.
No test assertion was relaxed and no corpus was removed.

This is structural analysis, not new goto/switch admission, def-use chains,
SSA, code motion, broader memory analysis or full register allocation. Backward
liveness remains a separate graph. See the updated objective ledger and roadmap.

Run bash test-pcfc11.sh, or tools/test_windows.py with native Windows Python.
RELEASE_TESTING.txt lists the extra bootstrap/sanitizer gates. Concise records
are under evidence/0.2.29; full logs and fresh sources are retained in PyCForge's
source archive under evidence/pcfc11_topology_1_3_81. Previous validation remains
in VALIDATION_0.2.28.md and the preserved original compiler source archive.
PyCForge distribution/installed-artifact qualification has a separate report.
