← All incidents

Drift incident · Raven Alpha (behaviour fingerprint)

gpt-4.1

Model page
✓ ResolvedConfidence: Medium
Responses to fixed prompts no longer match this model’s own history (84 recent vs 5964 baseline responses). Biggest shifts: token probabilities on “Pick a day of the week. Reply with one word.”; first token on “Pick a day of the week. Reply with one word.”; token probabilities on “Suggest a name for a pet cat. Reply with only the name.”
Oct 22change began ≈ 9:39 PM UTC
Oct 23incident opened 1:00 AM UTC
Nov 5resolved 1:41 AM UTC
6.7e-4adjusted 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 23, 2026, 1:00 AM UTC UTC. Resolved Nov 5, 2026, 1:41 AM UTC UTC: the series returned to its baseline. Methodology.

Share

Permanent link: https://martingale-bench.vercel.app/incidents/inc-gpt-4.1-fingerprint-2026-10-23T01

Incident id inc-gpt-4.1-fingerprint-2026-10-23T01. Every public incident is also in the Atom feed and the JSON feed, and in the raw data. Methodology v0.2.