Evidence
The authoritative 16-workload snapshot 20260827T154256Z (30 round-robin interleaved runs after 3 warmups) records a checked-JavaScript mean loss:
- tsc 6.0.3: 219.8 ± 20.4 ms
- native TypeScript/tsgo 7.0.2: 56.5 ± 2.7 ms
- Home: 58.4 ± 33.3 ms
Home has a 52.1 ms median, but one 234.3 ms sample raises the unfiltered mean above tsgo. The run must remain documented rather than discarded. The other 15 workloads, including the new null-safe-access workload, are Home mean wins.
Work
- Profile the existing frozen checkjs_jsdoc corpus and scaled same-shape copies.
- Improve only general checker/parser behavior; do not alter the corpus or discard timing samples.
- Preserve the full checker correctness gate and valid/invalid corpus gates.
- Publish a fresh full 30-run suite and retain the observed losing snapshot in the historical record.