Research focus / Predict the Self
Regression to the mean is not yet a reliable edge.
A cross-fitted length forecast improves descriptively, but its paired interval spans zero.
In 150 leave-one-out training cases, a regression-to-the-mean word-count distribution reaches CRPS 31.572068 versus 32.579884 for a raw no-source distribution. Its +1.007816 reduction has interval −1.235220 to +3.013792. The model corrects the much worse unit-slope persistence assumption, but it does not establish incremental response-length skill from the focal person's earlier answer. Private test results for the frozen text submission are still pending.
Read the research
From the lab notebook
Ipseity Daily Pulse / September 26, 2026
A trend leaderboard is not a discovery list
Among 704 eligible identity-signifier trends, dramatic point estimates appear in both tails—but none passes either the 5% Benjamini–Hochberg or Bonferroni screen.
Read the newest Pulse entry →Infrastructure / October 4, 2026
The Version 1 checks now preserve responsive zoom
All 39 public pages declare a responsive viewport without disabling scaling or setting a nonnegative maximum below 200%. Rendered reflow, keyboard, and assistive-technology review remains open.
Open the v1 summary →Paused project / NFL Team Fandom Identities
Fandom, happiness, and the right comparison
In a verified archive, happy endorsement is 83.23% among Browns-fan respondent-days versus 76.55% among explicit non-fans. The PI paused further work on September 15; the published evidence remains available. Read the results and provenance audit →
