What problem does it solve? Single cell-type annotation methods each carry their own biases and failure modes, and combining their outputs naively produces dead tallies when vocabularies differ or inflated confidence when methods are correlated. This Skill provides a framework for running a panel of independent annotation voters and fusing their per-cell votes into one defensible, flag-heavy consensus label set. ## Core Features & Use Cases - Voter panel composition: Coordinates marker consensus, scANVI/scArches reference transfer, popV ensembles, treeArches open-set novelty detection, and Census KNN votes as independent per-cell voters. - Vocabulary harmonization: Maps every voter's native nomenclature into one canonical label space via crosswalk adapters before tallying, preventing the dead-tally bug where labels never string-match. - Conservative fusion policy: Emits a frozen per-cell contract with annotation_confidence, basis_of_label, novelty_flag, and consensus labels, preserving disagreement as Uncertain or Novel rather than forcing labels. - Use Case: Annotating an aging or treatment-remodeling single-cell cohort where no reference atlas covers all states, so treeArches flags out-of-distribution cells as candidate novel states while independent corroborators raise confidence on known types. ## Quick Start Ask the AI to run a multi-tool consensus annotation over your clustered single-cell dataset, harmonizing votes from reference transfer, popV, and treeArches into one frozen label set with confidence and novelty flags.
