← All incidents
✓ ResolvedConfidence: Medium
Responses to fixed prompts no longer match this model’s own history (1297 recent vs 2968 baseline responses). Biggest shifts: output length on “Name one animal. Reply with one word.”; reasoning tokens on “Name one animal. Reply with one word.”; visible output length on “Name one animal. Reply with one word.”
Oct 7change began ≈ 8:53 PM UTC
Oct 9incident opened 3:32 AM UTC
Oct 10resolved 2:47 AM UTC
2.0e-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 9, 2026, 3:32 AM UTC UTC. Resolved Oct 10, 2026, 2:47 AM UTC UTC: the series returned to its baseline. Methodology.