Discovered while completing graph readmission under #487 and benchmark tracker #416.
The deterministic schema-3 benchmark corpus has 128 checked-JavaScript CommonJS owners plus one consumer. On current main at a572f3acd55fec4b15217d665c9433fe952a5741, a ReleaseFast home-tsc does not complete the unchanged positive project in a practical admission window:
python3 bench/vs_tsgo/run.py corpus
./zig-out/bin/home-tsc --noEmit -p bench/vs_tsgo/corpus/commonjs_graph/tsconfig.jsonObserved on Apple M3 Pro/macOS: still CPU-active after 7 minutes, roughly six cores, bounded memory, no diagnostics; manually interrupted. The all-workload schema-3 preflight therefore aborts at commonjs_graph before any timing/result directory. Host load was extreme, so elapsed time is classification evidence only, not a benchmark claim. The frozen admitted #546 checkpoint completed this same 128-owner workload at 42.1 ± 0.7 ms and passed its negative controls.
Required work:
- profile and identify the general algorithmic regression on the unchanged corpus; do not special-case benchmark names or skip checking
- preserve the real checked
module.exportsunion types and exact temporary negative controls (3× TS2322 + 3× TS2339) - retain the 198/198 CommonJS type-transfer audit and 132/132 static-require discovery audit
- add an operation-count or focused scaling regression for the diagnosed repeated work
- rerun complete schema-3 admission before any timing, then document the result and update the README table only with admitted evidence
No corpus, diagnostic, or sample filtering is allowed.