How I work

Measured, not asserted

Every performance claim on this site comes from a run you can reproduce in your own browser. Where an algorithm loses, the number is published anyway: single-hypothesis tracking genuinely beats multi-hypothesis tracking in some regimes, and knowing which regime you are in is worth more than a favourite algorithm.

Published literature, synthetic data

Everything here builds on algorithms that have been public for decades — Reid's multiple-hypothesis tracking, Munkres and auction assignment, k-best hypotheses, S-D assignment, OSPA evaluation — running on synthetic scenarios generated in the page. No proprietary methods, no real sensor data.

Failure modes first

The demo deliberately exposes what most demos hide: the filter's assumptions as a separate control group from the world's truth. Mis-tuned assumptions, degenerate geometry, and clutter-fed ghost tracks are where tracking systems actually fail, so that is what the material here spends its time on.