ReferenceAI skillsReproducibility and citation
Reproducibility and citation
Use to reproduce or cite earlier results, or rerun the same inputs.
Skill version: 1.0.0. Resource: skill://nadoobio-reproducibility/SKILL.md.
What is fixed
- An execution plan bundles the tool release (
manifestHash), definition, settings, and input references and fixes them withplanHash. Its body does not change. - If
statefromget_planisstale, the release, authorization, or policy changed after plan creation. Do not modify the plan. - To rerun the same inputs, pass the original
planIdand current release torerun_plan. The new plan links back to the original asrerun-source-planin its lineage and also requires human approval.
When citing results
- Before citing result files, verify the input → plan → submission → job → attempt → output chain with
get_provenance(view:'lineage'). - Include the plan identifier,
planHash, and release version.
Citing saved comparisons
- Find saved structure comparisons using actual IDs and cursors from
list_structure_comparisons({projectId}). Recheck onlycurrentrecords withget_structure_comparison. Do not bypass unavailable records or permission denials by computing a new comparison. - In citations, distinguish comparison ID,
comparison.sha256, source references/hashes, model/chain/region selections, and calculation version.comparison.sha256hashes the saved comparison payload; do not equate it with source-file hashes or the image hash returned byrender_result_view. - Use the
render_result_view(kind="structure-comparison", projectId, comparisonId)response and link for the same comparison ID when presenting a figure. Rereading saved results, recalculating from sources, and new model inference are distinct tasks. Listing records or receiving images is not independent reproduction or experimental validation.
Limits
The same plan means the same declared computation. Random seeds and GPU operation ordering may produce small numerical differences. Do not promise identical numbers from identical plans.
Generation versus validation
Define inputs for antibody/protein sequence or structure design and distinguish design from validation.
Choosing a research task
Map research intent to structure prediction, docking, design, complex analysis, trajectories, or metric interpretation and find currently executable tools.