← All incidents
✓ ResolvedConfidence: Low
Responses to fixed prompts no longer match this model’s own history (1342 recent vs 11760 baseline responses). Biggest shifts: whether it reasoned at all on “Give one tip for sleeping better, in one sentence.”; visible output length on “Give one tip for sleeping better, in one sentence.”; whether it reasoned at all on “Say hello in a creative way, in one short sentence.”
Oct 23change began ≈ 7:13 PM UTC
Oct 25incident opened 6:50 PM UTC
Oct 26resolved 2:48 AM UTC
9.7e-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 25, 2026, 6:50 PM UTC UTC. Resolved Oct 26, 2026, 2:48 AM UTC UTC: the series returned to its baseline. Methodology.