← All incidents
✓ ResolvedConfidence: Low
Responses to fixed prompts no longer match this model’s own history (672 recent vs 6048 baseline responses). Biggest shifts: answers on “List three uses for a brick.”; output length on “List three uses for a brick.”; wording on “List three uses for a brick.”
Oct 23change began ≈ 3:09 AM UTC
Oct 24incident opened 9:28 PM UTC
Oct 25resolved 6:57 PM UTC
6.5e-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 24, 2026, 9:28 PM UTC UTC. Resolved Oct 25, 2026, 6:57 PM UTC UTC: the series returned to its baseline. Methodology.