-
Gensokyo authored
Add focused mechanism/failed-case regression selection, immutable evidence, raw-image execution and matching-release checker adjudication. Verify a filtered source-only delivery from an empty build cache, then run its own artifacts on the original official disk and boot both default kernels. Document candidate limits, provenance, AI usage and publication authorization. Validation: run.ab4c57bf passed 18 Linux contracts, offline filtered dual ELF builds, actual original-image execution and same-artifact RV/LA boot/shutdown. Each RV libc group retains 19/32 passed, 12 failed and one fault; preserve checker exit 1 and all original failures. No LTP/LA official user test result or platform score. Actual final Git objects are compared with the reviewed source snapshot before pushing. Codex authored and verified this workflow. Parameter regression: run.e80629b3 used the same RV ELF on QEMU smp=2, 2GiB with two firmware harts and boot hart 0; original image/checker counts and resource balance stayed unchanged. This is not SMP scheduling support.
0b3a76ab