Do your two region sets really overlap more than chance?
Paste two BED files. Your browser computes their overlap exactly as gtars does - Jaccard, coverage both ways, intervals with a hit, nearest gaps - and tests it against shuffles and a universe, free, nothing uploaded. A paid run then interprets the overlap or writes the gtars script that reproduces it.
Each example has a saved model run, so you can see the whole page for free.
Your recent runs
What this does, and what it does not
The overlap follows gtars 0.9.2's RegionSet: BED read as 0-based half-open
(browser, track and # lines skipped, a first line with a non-numeric start read as a header),
intervals that share at least one base overlap, and reduce() merges overlapping
and book-ended intervals before jaccard(), coverage() and
overlap_coefficient(). It was checked against gtars 0.9.2, bedtools 2.26.0 and
scipy 1.15.3 on 300 random pairs of sets - duplicates, book-ended intervals, contigs only one
set uses, universes - with no disagreement in 38,679 comparisons.
The shuffle test moves every B interval to a uniformly random place on its own chromosome, keeps its width, and counts the bases it then shares with A; with a fixed seed it replays exactly. A whole-chromosome background ignores that peaks sit in mappable, accessible DNA, so it favours enrichment - a universe of regions either set could have hit is fairer, and the Fisher test on it follows the LOLA design. The paid run reads only what the browser computed and your notes; it is told never to compute a new number, and the page checks every number it writes. Derived from the agent skill @k-dense-ai/gtars (k-dense-ai/scientific-agent-skills, K-Dense Inc.). The example intervals are illustrative, not measured.