# Verifier notes

## Orphan numbers in paper.md

Integrity flagged four numbers written into Summary/Results without `{{R…}}` placeholders. Checked each against the code and the re-run:

1. **37** (Summary: "37 mechanical words") — Count of anchors in the four non-load classes in `code/mechanics.py` CLASSES: normal_stress 8 + shear 11 + flow_momentum 12 + path 6 = 37. Matches. (Load's 5 anchors are the separate benchmark class.)

2. **30** (Summary: "30 emotion, cognition and effort concepts") — Count of COUNTED targets in `code/mechanics.py`: affect 20 + cognition 8 + effort 2 = 30 (bodily targets excluded). Matches.

3. **95** (Summary: "95th percentile") — The hit threshold in `code/mechanics.py` is `pct >= 0.95` (MIN_HIT_FAMILIES = 3). The digit 95 names that registered percentile cut-off. Matches.

4. **30** (Results: "30-family rule") — `MIN_FAMILIES = 30` in `code/mechanics.py` / `code/colex.py`; eligibility requires at least 30 families with forms for both concepts. Matches.

No orphan number disagreed with the computation.

## Hazard screen

Answer: **none**. Linguistics/psychology colexification screen on public CLICS4 v1.0 (CC-BY-4.0 via external.json). No CSAM; no capability for mass harm; no private data or secrets; no substantial unlicensed copy of another's work pointed to in the bundle.

## Run

Harness re-run exited 0 in 40.9 s. C1, C2, C3 all reproduced (exact agreement on every declared result). No mismatches.
