Committee verdict
Weak Accept
Workshop/artifact track, with minor-to-moderate changes.
Reviewer hardening / 2026-05-18
A tracked response to the second external-style critique: weaker claims, clearer evidence boundaries, richer target rationale, and a documented scorer smoke test.
Weak Accept
Workshop/artifact track, with minor-to-moderate changes.
ready with limits
Public artifact packaging is complete; biological review and venue formatting remain.
Public source release, licence, tagged release, reproduction README, and Zenodo DOI are complete; biological review and venue formatting remain.
AF3-class and agentic benchmark language could be read as claims rather than context.| Committee request | Implemented response |
|---|---|
| Lower the title claim. | Retitled to Auditable local SimpleFold-3B inference on Apple Silicon: an 8-target reproducibility benchmark. |
| Open with SimpleFold-3B, not broad model-class language. | Abstract rewritten around the exact model, route, hardware, target set, metrics, and non-claims. |
| Put claim boundaries in the paper. | Main manuscript now includes a Claim / Evidence / Limitation table. |
| Explain the target set. | Methods now include inclusion/exclusion rules plus method, resolution, context, rationale, and limits per target. |
| Document the scorer. | Methods now describe sequence-aligned lDDT pseudocode and metadata fields. |
| Add a minimal scorer test. | pilot/scripts/test_lddt_scorer.py verifies offset residue numbering returns lDDT 1.0 under sequence alignment. |
| Separate repeated and single-run evidence. | Results now state only 13CZ:A, 11BC:C, and 9K9Y:A have n=3 controls. |
| Make artifacts citable. | Published the clean v0.1.2-phase1a source snapshot and archived source/data/PDF artifacts at 10.5281/zenodo.20275055. |