← All incidents
✓ ResolvedConfidence: Low
Responses to fixed prompts no longer match this model’s own history (1344 recent vs 8071 baseline responses). Biggest shifts: output length on “Name one fruit. Reply with one word.”; reasoning tokens on “Name one fruit. Reply with one word.”; visible output length on “Name one fruit. Reply with one word.”
Oct 12change began ≈ 7:10 PM UTC
Oct 14incident opened 9:00 AM UTC
Oct 15resolved 1:12 AM UTC
7.3e-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 14, 2026, 9:00 AM UTC UTC. Resolved Oct 15, 2026, 1:12 AM UTC UTC: the series returned to its baseline. Methodology.