Eight fields
The same method, different subjects.
Maths
9 formal theorems
New computer-checked proofs, in Lean, towards an open Erdős problem (number 686), with rigorous computer-assisted results beyond the previous record.
A partial result. The problem stays open.
Astronomy
Same inputs, same answer
We re-ran the two rival teams' analysis of the Hubble constant on public data. On identical inputs the two methods agree. The disagreement lives in the inputs, not the method.
Water
16 of 23
hidden leaks found in a town's water network. That matched the best published competition entries.
Chemistry
On a par
with a leading neural network at predicting boiling point and solubility, at about a millisecond per molecule, every number traced.
Drought
7 of 7
major Horn of Africa droughts since 1996 warned of in advance, a median 7 months ahead, from public data only.
A small sample, and about half of all warnings were false alarms. The full result and its caveats →
Imaging
All 26 tests
Photo enlargement beat the standard methods on every one. It never invents detail: any guessed area is marked. About 250 times faster than neural upscaling.
Physics
Matches exact answers
Our physics engine reproduces the exact answers where exact answers exist, on ordinary computers.
Translation
86 · 75 · 74 · 67
chrF scores for Italian, German, French and Polish, on the standard 0 to 100 translation-quality scale. About twice the quality of word-for-word machine translation on the same sentences, with numbers, names and terms unchanged in every test.
What proof means here
Specific, checkable, and shown with its limits.
The same answer every time
Ask again and you get the same result, with every step laid out so it can be checked by someone who was not in the room.
Limits stay attached
A partial proof is called partial. A small sample is called small. A result on simulated data is never presented as a clinical one. The caveat travels with the number wherever we print it.
Want the method pointed at your field? request@kinesthetics.io