3.7 KiB
REPAIR50 source-ready handoff
Scope: control classification/memo, tree() read method only, stream equivalence control and documentation. No Git write, engine, Runner, integration or formal test was performed. Three changed/new Python sources passed AST parsing.
Root-only execution and estimation
From this package directory, run the following pure controls once and retain their separate JSON output and external elapsed/RSS measurements:
python3 s5/s5_order_adapter_controls.py --pure-only
python3 repair50_stream_controls.py
The first emits separate negative counts/lists, identity positive counts/lists, no-genuine-order lists, memo hit/miss, parser-rejection list and elapsed per guard. The second compares the complete whole-read and streamed SHA maps and key insertion order for empty, Unicode name/content, binary, sparse, chunk boundary -1/exact/+1, multiple chunks and nested files. It uses temporary synthetic files only and creates no native acceptance evidence.
Before heavy typed controls, parse each actual retained guard record with:
python3 s5/s5_order_adapter_controls.py --estimate-record /absolute/path/to/guard.var
This performs no derivation or verify: output includes record byte size, AFTER node count, kind histogram, replacement count and replacement count x2 (including identities, never reported as negatives). Guard mutation disk estimate is zero; stream control logical fixture size is about 10 MiB, with the sparse part filesystem-dependent. Memory scales with retained parsed representatives (at most one per mutated kind, plus first/last), record bytes, parser overhead and output labels. Memo holds one pair of expected/result raw bytes for unchanged inputs; it never caches AFTER or recorded EXPECTED. Time scales with mutation count x record bytes: parse/verify still examine whole records. Use the synthetic run's external peak RSS and elapsed, together with its emitted counts/bytes, as an initial calibration; real record shape and source-derivation costs make this an estimate, not an ETA or resource cap. Actual guard bytes/counts and calibrated measurements are Root-pending.
complete_B scope reuse remains conditional
The supplied commander handoff reports R49 session56645 exit 0, 49-file map, S5 facts/payload equality, no B changes, log SHA prefix bdef6715 and closure tree prefix b54a35ed. This worker has not independently verified that run. Root must mechanically compare its saved full raw map with the new tree and record only the authorized control/tree-read/new-control/document changes. Inspect import/exec and input closure for builder, collector, native-series, reader, oracle, wrapper, probe and helper; preserve S47 raw, 21pin, legacy5, full184 originals, oldQ source/engine SHA and public run_all_tests.py SHA.
Important: repair48_identity_controls.CLOSURE explicitly lists the changed order control, identity control and S5_ROOT_RUN.md as snapshot material. The roster is unchanged. Direct searches in builder/native-series/order adapter did not reference these names, but this is not a complete graph proof and does not establish reuse. Root must distinguish the actual two-directory analysis/payload execution/input closure from this snapshot roster and apply all commander conditions. If any condition fails, rerun complete_B once. No R49 closure bytes or raw evidence were rewritten to obtain reuse.
Then preserve the recorded Root order: pure controls → clean exact-Q check-only → capacity/PID confirmation → fresh native6 → reader/collector → all eight full exact-source reuse conditions (failure means one full run) → index → one registered p04-acceptance validation → independent whole-unit review → normal gates/RootDB join/ledger/main push. Only whole-unit acceptance awards credit.