Tenebrio molitor behavioral phenotyping dataset: derived tracking, kinematics, features and QC data (34 videos)

Derived framewise and aggregated behavioral-phenotyping data for Tenebrio molitor larvae recorded individually in six-dish Petri trays (34 videos, 204 arena-recordings, 3,421,266 framewise observations). Produced by the wormstress-vision pipeline (YOLO11 Petri localisation, YOLO11 larva segmentation, ByteTrack tracking on per-dish crops). Contents: Petri geometry and larva tracks per frame (crop, full-frame and Petri-normalised coordinates, mask morphology, confidence, selection reason, quality flags), all raw detections, kinematics, per-window and per-video features, occupancy heatmaps (.npy), QC summary, experiment metadata, data dictionary, model cards and processing provenance. See README.md. Accompanies the manuscript A modular computer-vision architecture for converting controlled video assays into behavioral phenotyping data (Di Lorenzo, Zanzot, Garabini, Mancini, Angelini). The values are objective behavioral proxies (locomotion, immobility, spatial occupancy, area dynamics); they do not measure welfare or subjective state.