The project site lists 36 manuscripts and groups them into nine broad areas. Schwartz's own announcement counts 18 fields and 19 coauthors over roughly three months.

"22 fields" is the magazine's count, not the author's. More important, these are preprints and drafts. Several are listed as in preparation, and the site warns that many have not been fully checked by humans. Peer review has not happened yet. The work is real. The tally is generous.

This is the strongest claim and the best supported. Closed forms must match independent numerical evaluations to thirty decimal places at held out points. Evidence integrals run in exact rational arithmetic, cross checked with two disjoint sets of primes. The origin story holds up. Thirty elliptic Feynman integrals computed end to end, fifteen of them never computed before. In phylogenetics, two candidate trees for humans, chimpanzees, gorillas and orangutans come out exactly tied in probability, something estimation could never settle.

Schwartz is careful here, more careful than the coverage. His own paper notes that agreement to thirty digits is numerical evidence, not a symbolic proof. "Precisely" is fair. "Proven" would be a different word.

This is the paper a healthcare reader should open. The Medicare Advantage regulation says two things. Section 422.162(a) says star levels go to the clusters that "minimize the within-cluster sum of squares." Section 422.166(a)(2)(i) says to find them with hierarchical clustering and mean resampling. CMS uses Ward's method, which is a heuristic and can stop short of that minimum. For scores on a single line, the exact answer has been computable since Walter Fisher's dynamic program in 1958 (the same math cartographers use to shade maps). Of 123 cut point computations for 2024 through 2026, 104 miss the minimum. Rescoring changes 9,809 of 47,752 contract measure stars, about one in five.

The billion is gross, not net. It is $0.9 to $1.4 billion over payment years 2025 to 2027, priced on a $400 to $600 per enrollee reference band, from 34 contract years that cross the 4.0 bonus line in both directions (16 up, 18 down). One contract year accounts for 59 percent of the enrollment in the estimate, and its rating sits 0.03 above a rounding edge. CMS does not publish its inputs, so the score lists are reconstructions. The paper attributes no error to CMS, takes no side on which sentence of the regulation controls, and lists no health policy coauthor. That last point matters with active litigation over star ratings (SCAN and Alignment both filed new suits in July 2026).

Hubbell's neutral theory of biodiversity is about species, not genes. It treats every tree species as ecologically equivalent, so a forest's composition drifts by birth, death and migration. The BootLoops result was that on Barro Colorado Island in Panama the species mix changes about 4.5 times faster than neutral theory predicts. The ecologist James O'Dwyer called the results solid and interesting, and unlikely to move the field much.

Science described Kimura's neutral theory of molecular evolution, a cousin with the same adjective. In a story about AI crossing disciplines, the human reporters crossed two fields that share a word.

This is the honest center of the story, and Schwartz says it more bluntly than the magazine. In his announcement he describes a model that overclaims, declares itself done too early and misjudges timelines. Automated checks were not trustworthy, so he looked at the plots himself. He calls his role "Claude handler." On colleagues' advice he withheld results that were insufficiently validated. On a Voynich manuscript forum, readers called his decipherment paper slop. Science adds that the approach favors older, highly cited problems over new ones, which could narrow what science explores.

His own line is the summary. "Claude can find Claude-shaped problems, but there are thousands, perhaps millions, of them." The bottleneck moved from computing the answer to choosing the question.