phybuilds

Camera traps now record far more wildlife footage than small research teams can review. Crowded scenes, occlusion, camera pans and broken tracks make full automation unreliable, and manual review is slow and fatigue-prone. Reviewers are also prone to change blindness, so animals entering or leaving the frame get missed. WildScore turns wildlife video clips into an evidence-linked, auditable event timeline that researchers review by sight and sound. SAM 3.1 tracks animals and proposes appearance and disappearance events. Muse Spark 1.3 adds bounded visual labels for review. The researcher hears a short audio cue at each event, stereo-panned to where the animal enters or exits (enters left = left speaker). They then verify each proposal with hotkeys (Accept, Reject, Unsure) and can draw a recovery track around any animal the model missed. Machine proposals are stored separately from human decisions, and the output is an auditable CSV. WildScore is a collaborative verification tool, not a black-box counter, designed to be faster than purely manual review and more auditable than purely automated review. In an exploratory pilot case study comparing manual review with WildScore (n=2) on a crowded waterhole clip, manual review took 34.5 minutes and produced 22 accepted events. WildScore-assisted review took 6 minutes and produced 30 accepted events (26 reviewer-accepted proposals plus 4 recovery tracks). With no adjudicated reference count, this compares review logs, not recall or precision.