3.2 KiB
feab202 narrow native preparation revision
New helper .local/rename-scroll-native-feab202.py and config .local/rename-scroll-native-feab202.config.json replace no prior files. Tested-source admission is fixed to feab2020d8e5b6bfdb59c86ab4aee161617b8b6f; only the reviewed test file differs from bebf production. All seven stems now end in feab202 and GUI is gui-feab202. Config pins actual r3 expected/admit/run/build helper bytes. The pending source flag is cleared, but every actual clean gate/admission/build record remains mandatory; none was executed or inferred by this preparation.
Preserved old bebf helper SHA d4647d4bf9643f0a806d22a6f039fffc4b87953720b21f79239c29aab83fbb26 and config SHA484a07b39348449e32f54c59a0b55e703e8adce7ffac614f69e4703308d14b96. New config retains original bebf preview failure inputs/log/result, old440 normal closure and original six-file fixture provenance. Fresh trial/stage use rename-scroll-feab202-20260913-01 / rename-scroll-stage-feab202-20260913-01; neither exists.
The checkpoint now records complete before/snapshot/after file maps. It reads the actual copied files, rereads work after copying, retains copy/read errors, and requires all three maps equal plus exact expected baseline/stage. Mismatch receipts and partial snapshots are written before refusal. Git cat-file parsing now explicitly requires the newline byte immediately following each declared blob payload, in addition to exact bytes and total consumption.
Actual old440 GUI and server Cargo records both have opt_level0, debuginfo0, debug_assertions true, overflow_checks true, test false. The helper's full profile equality remains unchanged; required opt0/debugtrue/nontest was not weakened. The future feab GUI must actually supply that schema/profile.
Validation executed only AST parsing, assert-node exclusion and isolated selected-function simulation. No helper main/admission/check/prepare/launch, compiler, tool query or GUI input ran. The synthetic directory .local/rename-scroll-checkpoint-simulation-feab202 contains five clearly separated synthetic trials: stable copy accepted; work mutation before copy rejected; work mutation after copy rejected; snapshot corruption rejected; source disappearance during copy rejected. Every scenario retained a result receipt with actual snapshot and after maps; all four negative cases raised only after recording failure. These synthetic launch/preparation records do not represent any native process. The exact AST-selected Git terminator predicate accepted LF and rejected wrong/missing terminators. Result .local/rename-scroll-checkpoint-proof-feab202.json binds the helper and five receipt hashes. This does not prove full gate admission or native behavior; parent must review and execute against actual closed future gates.
Exact source diff is .local/rename-scroll-native-feab202.patch; config diff is .local/rename-scroll-native-feab202-config.patch. Parent review should concentrate on the new capture function/result ordering and current r3 pins. Normal owned GUI launch/closure, private profile/cwd, Cancel-only disk equality and original trial scope are otherwise retained. Awaiting the parent's next task; no automatic launch is scheduled.