A second mathematician publicly questioned on September 10 whether researchers can safely share unpublished proofs with OpenAI at all, two days after NYU's Tristan Buckmaster disputed the lab's uncredited use of a Navier-Stokes result he and an Anthropic researcher had already posted.
how work was measured today
Counted layer · the radar, 135 of 208 sources reporting
Work's day ran 12.2 per 100 sources; usual is 11.5
counted 17:17 UTC · per 100 reporting · against 12 stamped days
1 day 12.2 +0.14 sd 3 days 13.5 −0.21 sd 7 days 13.9 −0.66 sd 0 10 20 per 100 reporting · tick = a stamped day 1 day 12.2 +0.14 sd 3 days 13.5 −0.21 sd 7 days 13.9 −0.66 sd 0 10 20 per 100 reporting tick = a stamped day The idea Each window counts the sources a day publishing on work, per 100 sources reporting, so a day when more sources report does not read as busier. The signed number is today’s distance from its usual, in standard deviations.
Look for The dot is today; each tick is one of the 12 stamped days before it. The band is the usual range, one standard deviation each side: 7 of 12 days fall in it. 1 day sits above its usual; 3 days and 7 days sit below. The numbers +0.14 1 day, in standard deviations
12.2 sources a day per 100, today
11.5 its usual, the mean of 12 days
12 stamped days behind it
Limits It says the volume moved; it does not say why. It counts sources publishing, not a share of all talk. Under 14 stamped days a window reads early, a hollow dot; under 7 it does not print. How it is measured Per 100: sources publishing on the domain in the window, a day, over sources reporting that day, times 100. Usual: the mean of up to 56 stamped days before this sheet, one per scan, this sheet’s own scan left out; the spread is their standard deviation. Deviation: today less the usual, over the spread. src/data/cover-2026-09-11.json counted.domains, scripts/build-cover.py rep_scores.
This is the counted layer: how much the feeds we read published about work, per source reporting, against what they usually publish. The story above is the heard layer, written from sources. They are kept apart on purpose.
See every domain on the board .