The statistics are plain code
Every number comes from code, not from a language model: Newcombe confidence intervals, with HIGH needing at least 300 matches in each group.
12 / Game patch analysis · solo · paused
After each League of Legends update, a game operations team spends days cross-checking what changed. Aftershock reads the patch notes, measures real matches on the Vietnam server and says what the patch actually did. Each finding carries its match count, confidence interval and match IDs, and when the data falls short it declines to conclude. The project is paused and its public build has been taken down.

No number appears bare. Each one carries how sure the system is about it.
Every number comes from code, not from a language model: Newcombe confidence intervals, with HIGH needing at least 300 matches in each group.
Benjamini-Hochberg filters the findings when many are tested at once. Naafiri dropped 5.3 points over 808 matches and the claim was still withheld, because the interval touched zero.
Locke won 54.0% in mid and 42.7% in the jungle, an 11.3-point gap rated MEDIUM. A 9.1-point rise sounded better, but its interval crossed zero, so it was marked LOW and left unclaimed.
By default it replays a frozen run so a demo does not depend on the network, and the design says so openly. The LIVE switch recomputes on the server from the same data.


Build a product together, distribute one that exists, commission software, or join in — write a few lines and NYX will get back to you.
Send an invitation →