Method catalogue

All eight methods share one scientific module/database. Each has a separate standalone import and documentation page; importing several methods does not create several engine copies. Create one client for sequential operations, or several explicitly scheduled clients for independent parallel models.

Import suffix	Function / instance method	ABI ID	Changes sequence	Fixed work budget
/score	score	score	No	Scoring only
/relax	relax	relax	No	One torsion FastRelax repeat
/minimize	minimize	minimize	No	≤200 L-BFGS iterations
/repack	repack	repack	No	One side-chain packing round
/fixed-backbone-design	fixedBackboneDesign	fixed_backbone_design	Yes	One fixed-backbone packing/design round
/fastdesign	fastDesign	fastdesign	Yes	One FastDesign repeat
/cartesian-minimize	cartesianMinimize	cartesian_minimize	No	≤200 Cartesian iterations
/cartesian-relax	cartesianRelax	cartesian_relax	No	One repeat, ≤200 iterations per minimizer invocation

Common options are seed, per-worker duty target, AbortSignal and observation callbacks. Repeats, convergence settings, score functions, custom constraints and residue-specific design policies are fixed rather than user-selectable in this API. Input/output contracts define canonical single-chain PDB admission with no residue-count cap; finite resource limits still apply.

All eight methods passed the recorded engine's native/WASM and Chrome/Brave protocol checks. Coverage is fixture-specific: most tests use 20-residue 1L2Y, while browser FastDesign uses eight residues. Separate 1UBQ scoring extends the input-size evidence. The benchmarks describe these limits; input admission does not establish general scientific coverage.

Broader upstream application modes live in separate research modules. They are not silently routed through client.run, and unknown identifiers reject instead of falling back to scoring.