Actuary
← All incidents

Code reasoning

Moonshot Kimi K2.5

moonshotai/kimi-k2.5

Open Opened 2026-08-06 01:05:28 UTC

Fixed public statement

Behavioral change evidence

Estimated effect size (frozen null floor minus observed panel mean): 0.233, with a 95% anytime-valid uncertainty interval [-0.300, 0.700]. statistically significant behavioral change detected. Methodology.

Observed via aggregator: through an aggregator serving layer, not a direct adapter.

Anytime-valid evidence

E-value trajectory

E-value at open
21.408
log₁₀ E at open
1.331
Onset estimate
12

The onset is a diagnostic estimate of the first affected post-calibration observation. Statistical validity attaches to the e-process alarm, not to this estimate.

E-value trajectoryBase-ten logarithm of the e-value after calibration. The red line marks the configured alarm threshold. log10 E Completed panel 0.0 1.0 2.0 3.0 0 5 10 15 20 25 30 35
The trajectory plots log₁₀ E through the incident epoch. The red line is the configured 1/α alarm threshold.

Audit trail

Epoch lineage

Incident epoch
fbac9ea0-9c5d-48d1-8454-df1049493db7
Pool version
code_reasoning-v4
Checker version
code_reasoning-v2
Previous epoch
2582d56c-7eab-4b5e-b1d1-1314be891eb3
Epoch transition
checker_refresh
View model status →

Citing this incident: link this page and name the UTC day. The register is append-only — corrections are published beside the entry, never over it — so a dated citation stays true. The evidence bundle above replays offline.