Submit results

Submit your own KST run

Submitting is not publishing. Every submitted run is replayed from its audit envelope before it appears on the board: we do not take the numbers on face. No trained-rater set exists yet, so recomputed scores are placeholders.

Status and corrections (October 2026). This site describes the Kari-Sheldon Test as designed, and much of that design is not implemented: there are no trained or certified raters, no reliability statistic, no DIF or fairness analysis, no IRT scaling or calibration sample, and no expert review (the “expert panel” was not a panel of human experts); every item is sent as a single prompt. Scores come from automated keyword, regular-expression and word-overlap rules. They are placeholders, not validated measurements of sapience or of any other construct. See the correction notice in the KST repository.

Validated submissions open soon

The self-serve submission flow is being finalized so that every run can be replayed end to end from its audit envelope before it is published. While we finish it, send us your run and we will validate it by hand.

Email your run → Request a test instead

What a complete submission includes

When the form opens, and when you email a run today, a complete submission carries everything needed to reproduce the result:

  • The per-item JSON envelope the harness emits, plus the composite report.
  • The KST version and the battery config used.
  • The exact model version-pin: the snapshot you tested.
  • Whether certified raters were used, or whether the run is submitted for KST-side rater validation.

We then replay every item and recompute the automated scores (no trained-rater set exists yet). If our recomputed numbers diverge from yours, you will see exactly which items and why, before anything is public. Self-submitted runs are recomputed, not trusted on face. That is a feature: it is what lets a reader trust the board.