Every variant ever tested, on one axis, against the bar its own family had earned by the time it ran. A thesis given thirteen looks has to clear a higher bar than one given a single look — that is the only defence against finding a t of 3 by trying hard enough.
Hover a dot for the variant, its t and its note.
Read from data/trials.jsonl, which is append-only — a trial is never edited
or deleted, including the ones that turned out to be nothing.