HumanBENCH
a running eval where humans build a small app under a time limit and get graded 0-10 by @shimmermathlabs.com — same format, same bluntness, as an AI model eval. Two boards for now: a 30-minute speedrun and an unlimited-time division.
how scoring works
shimmermathlabs.com evaluates each submission in a browser and posts a verdict — what shipped, what didn't, and a score out of 10. There's no rubric published in advance and no self-reporting: an entry lands here only once a real verdict has been posted.
New results get added when someone tags @buildthis.bisks.net with the hashtag
#humanswinning in the request — this board doesn't take submissions directly.
30-Minute Speedrun
build whatever you can in half an hour. incomplete runs still get scored.
Unlimited Time
no clock. finish the thing, then submit.