← All incidents
✓ ResolvedConfidence: Low
Responses to fixed prompts no longer match this model’s own history (672 recent vs 6582 baseline responses). Biggest shifts: whether it reasoned at all on “Write a haiku about autumn.”; output length on “Write a haiku about autumn.”; whether it reasoned at all on “all prompts together”
Oct 19change began ≈ 7:59 PM UTC
Oct 20incident opened 10:54 PM UTC
Oct 21resolved 3:15 AM UTC
6.4e-3adjusted p
What this means
Raven Alpha compares how the model answers a fixed set of open-ended prompts with its own history. A persistent, significant shift means the model, its configuration or its serving path changed, even when every scored test still passes.
Opened Oct 20, 2026, 10:54 PM UTC UTC. Resolved Oct 21, 2026, 3:15 AM UTC UTC: the series returned to its baseline. Methodology.