binder-lane-round

Binder Lane Round

Use this public Structure Factory skill with Claude Code, Codex, or another command-capable agent. The Claude binder lane guide provides the repository entry point and public example. Compare binder-design methods against a fixed target and site. The CLI validates plans and handoffs, runs synthetic examples, and executes authorized local adapters with fixed arguments.

Read first

Read references/docs/binder-lane-round.md for the contract, references/docs/binder-study-decision-loop.md for study decisions, and references/docs/binder-controls.md for measured calibration. Check references/docs/tooling-and-licensing.md and references/docs/compute-backends.md for the selected execution route. references/NON_CLAIMS.md defines the scientific result boundaries.

For BindCraft2, read references/tools/bindcraft2.md. Its settings, output layout, and deployment terms require a separate adapter from BindCraft.

Define the comparison

Record the target accession, chains, residue selections, and input hashes. Choose the generator, sequence designer, predictors, scorers, filters, and execution route for each stage. Check code and checkpoint terms, API terms, required downloads, and redistribution conditions.

Set the budget ceiling, runtime cap, round count, primary metric, stopping rule, and controls. Declare output counts, formats, hashes, and cleanup requirements. Comparison arms must share candidate counts, predictor panel, scorers, filters, controls, and failure policy.

Use reference_scope: published_tool_identities to check the published tool identities on selected stages. Use published_workflow_shape to compare replacement tools within the same workflow. Identify replacements in each arm. Record acceleration as a runtime choice attached to the model and checkpoint.

Prepare and run

1. Inspect bsf binder-lane menu, validate the request with plan-request, and materialize the plan, round contract, and execution handoff with plan.
2. Run preflight and target-check against the plan and coordinate input. The run command executes only public_synthetic_demo plans.
3. Inspect bundled commands with adapters. For a missing command, supply a validated adapter registry under .runtime/.
4. Use prepare-execution with the target report and stage settings. Resolve each readiness error, then dry-run the adapter or controller.
5. For a remote route, validate remote-request, execute through the selected transport, and validate the exported artifacts and cleanup with remote-receipt.
6. Before a real start, confirm authorization for the route, data, budget, runtime, and required terms or downloads. Reuse approval that covers those conditions. Start the local adapter or controller with --authorize-local-execution. Add --authorize-network or --authorize-license-gates only when required and authorized.
7. Run closeout to count, parse, and hash the declared outputs.
8. For a measured primary metric, run calibrate-controls. Pass its ready record with --calibration to round-decision.
9. Apply round-decision to the sequential round history. Continue only when the decision is continue and the remaining budget covers another round.