Start from a disease and let Halffield fuse every signal into a ranked, defensible candidate ladder — then promote only what a human signs off.
Halffield’s discovery surface is indication-first: pick a disease (a first-class, EFO-keyed workspace) and it pulls associated targets from Open Targets, your own screens, your run outputs, and literature watchers — fusing them into one evidence spine. Every target gets an auto-built dossier enriched from public biology: disease association, tractability, safety/constraint, clinical precedent, structures.
Candidates are ranked by a weighted harmonic-sum ladder with a genetic-evidence tier, orthogonality-aware breadth, and direction-of-effect flags — and scored by a full multi-parameter optimization that yields a diverse Pareto portfolio, not one brittle winner.
Nothing auto-promotes. A candidate becomes a campaign target only when an operator signs off, with an auto-assembled validation package. And the loop closes: results from the resulting campaign flow back as evidence for the next round.
Predicted target↔ligand affinity (Boltz-2) plus experimental structures enriched from RCSB PDB, ranked as the gold standard.
Private upload of internally-solved structures, viewed in an inline 3D viewer — never exposed publicly.