# experiments/bare_metal/CLAUDE.md -- loaded when working under experiments/bare_metal/
# (moved from .claude/CLAUDE.md's word-level ACL section, 2026-11-06)

**ACL-RWT DoE campaign (June 15–16 2026) — superseded, kept for historical record:**
- 3×3 Latin square: 3 seeds × amd64/aarch64/riscv64, 30 reps each
- Measured overhead: +0.0054%–+0.0088% across all 9 cells; CV = 0.000%
- Report confirmed to exist: `experiments/bare_metal/analysis/report/bare_metal_doe_report.pdf`
  (~2MB, with LaTeX source and figures) — patent support material
- **Correction (found 2026-08-20/21, `FABRIC-1.md` §§M–T):** the "ACL-RWT" name itself was
  wrong — the Rolling-Window-of-Truth mechanism it was named after was dead code, removed
  2026-07-08 (`ACL-RECHECK-RW` was never reachable — `acl_recheck()` only ever looks up the
  11-char `ACL-RECHECK`, never the 14-char `-RW` variant). What this campaign actually
  measured was the live `ACL-TTL` mechanism under a misleading name, and at `-O0`, before
  real compiler optimization landed. The corrected, current figure — same live mechanism,
  correct name, re-run on the current substrate — is **+0.0603% ACL-TTL enforcement overhead,
  architecture-independent, CV=0.000%** (`FABRIC-1.md` §T, memory
  `project_acl_ttl_overhead_final.md`). Use that number, not the one above.