← All incidents
✓ ResolvedConfidence: Low
Responses to fixed prompts no longer match this model’s own history (1344 recent vs 1008 baseline responses). Biggest shifts: whether it reasoned at all on “Name one animal. Reply with one word.”; visible output length on “Name one animal. Reply with one word.”; reasoning tokens on “Suggest a name for a pet cat. Reply with only the name.”
Oct 5change began ≈ 11:31 PM UTC
Oct 7incident opened 5:22 AM UTC
Oct 7resolved 7:43 PM UTC
9.3e-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 7, 2026, 5:22 AM UTC UTC. Resolved Oct 7, 2026, 7:43 PM UTC UTC: the series returned to its baseline. Methodology.