Fetching from the wire…
Skills2026-03-16 · source-backed
At 75% per-trial success and k=3: pass@k = 98%, pass^k = 42%. Use pass@k during development ("did it ever work?"). Use pass^k for production ("does it always work?"). At k=10 they diverge catastrophically. Anthropic Engineering
Each link below shares sources, entities, or timing with this story.
Shared entity: Anthropic Engineering / Same source / Tension
Both cover Anthropic Engineering; cite the same source (Anthropic Engineering); pushes against this story (vs).
Shared entity: Anthropic Engineering / Same source domain / What happened next / Tension
Both cover Anthropic Engineering; reported by the same outlet (anthropic.com); picks up the Anthropic Engineering thread on 2026-03-19.
Shared entity: Anthropic Engineering / Same source domain / Earlier coverage / Tension
Both cover Anthropic Engineering; reported by the same outlet (anthropic.com); earlier Anthropic Engineering coverage from 2026-02-27.