V-JEPA looks for motion
It can rank clips with similar movement, without naming a person or deciding why it happened.
ORCHESTRA / PATTERN LAB SIMULATION · NO LIVE DATA
Find the repeat.
Question the story.
A long archive can surface recurring activity. Explore how motion, scene context, and independent windows can separate a review lead from a tempting false pattern.
Every scene, count, and cue on this page is invented in your browser.
SYNTHETIC ARCHIVE / WORKBENCH 01
Which story does the evidence support?
SIX SYNTHETIC TIME SLICES
Repeated windows
MONTH 01 / SAMPLE WINDOWS
One cue in a wider scene
Open a time slice to see which motion fragments match and what context is still needed.
1 / 4windows share a cue
MOTION SEQUENCEORDER RETAINED
Approach
Pause
Depart
All values are syntheticNO RECORDINGS · NO LIVE MODELS · NO FINDINGS
A LEAD IS NOT A CONCLUSION
Three checks keep a pattern honest.
Repeated motion can be ordinary. The review asks whether independent windows agree, whether sequence and context hold up, and whether coverage is good enough to tell.
Cosmos adds scene context
A second model can describe visible activity; its machine-written label still needs checking.
A person reviews the evidence
Only a human can decide whether the available footage supports a real-world conclusion.
Explore the synthetic pipeline walkthrough Return to Archive Lab