Fetching from the wire…
Skills2026-06-08 · source-backed
A single passing eval hides nondeterminism. Measure all-runs consistency with pass^k (the probability all k runs pass) so flaky agents get caught before release. Pair it with a judge calibrated against a human gold set, and grow that golden set from real production traces so the metric tracks how your agent actually fails.
Each link below shares sources, entities, or timing with this story.
Shared entity: Gate / Shared topic / What happened next / Tension
Both cover Gate; overlapping topics (against, agent, gate); picks up the Gate thread on 2026-06-24.
Same source / Shared topic
Cite the same source (all-runs consistency with pass^k); overlapping topics (actually, agent, consistency, pass).
Shared entity: Gate / Shared topic / What happened next / Tension
Both cover Gate; overlapping topics (agent, gate); picks up the Gate thread on 2026-07-30.
Same source domain / Shared topic / Tension / Downstream implication
Reported by the same outlet (digitalapplied.com); overlapping topics (against, agent); pushes against this story (against).
Shared entity: Measure / Shared topic / Earlier coverage
Both cover Measure; overlapping topics (agent, calibrated, consistency); earlier Measure coverage from 2026-03-29.
Shared entity: Gate / Shared topic / What happened next
Both cover Gate; overlapping topics (agent, gate); picks up the Gate thread on 2026-08-07.
Both cover Gate; overlapping topics (agent, gate); picks up the Gate thread on 2026-06-30.
Shared entity: Measure / Shared topic / What happened next
Both cover Measure; overlapping topics (agent, gate); picks up the Measure thread on 2026-06-27.