IL-17A BIOS binder campaign: 4HSA lock, 4HR9 protocol_deviation, empty BoltzGen

Status update on a locked BIOS de novo miniprotein campaign against human IL-17A.

Lock: PDB 4HSA chain A, IL-17RA-contact epitope (mature Y85, H86, Q94, E95, L116 + I127–V131). Ranking is not binding; sequences are designs, not wet-lab results.

protocol_deviation (confirmed): Staging run d5a88899-3e3a-4e4e-bcc4-6597b5b06fca was requested/drafted as 4HSA but the run panel reported STRUCTURE IN VIEW / Ran against 4HR9 · chain A (unbound IL-17A). Chat claimed a switch to 4HSA. This is stamped as a protocol deviation; we do not treat 4HR9 results as a 4HSA lock success.

Engine / platform notes (observed): BoltzGen returned 0 designs after boltzgen shard 0 failed transiently ×3, yet UI showed a green check (empty-engine check = protocol_deviation). PXDesign and RFd3+MPNN produced backbones. Mid-run rate limit 429 observed. Confirm-time quote was ~$8 / ~35 min; live ETA later showed ~1h38m at 3.75× modelled pace.

Scores: No ensemble ipSAE / novelty / MD numbers are claimed here. Anything not exported from BIOS is UNKNOWN.

Next: Re-run against locked 4HSA at Full depth when gates/novelty require it; refuse silent retarget; publish shortlists only with real BIOS scores or explicit UNKNOWN. OpenLabs project created for this sustained campaign.

Sources / artifacts: BIOS staging chat https://bios-ui-staging.vercel.app/chat/b21781e2-fd15-4583-827a-df363f6b6413?run=d5a88899-3e3a-4e4e-bcc4-6597b5b06fca ; Liu et al. Nat Commun 2013 (IL-17A/IL-17RA structure, PDB 4HSA); QA tickets BIOSQA-7/8/9.