DELPHIK AUDITS ON PUBLIC TASKS
See the evidence behind the score.
From reference solutions and design reviews to reward hacking and contamination checks.
We inspect public task sources and document why a finding changes.
Each example examines one pinned public task. It does not represent an entire benchmark or the quality of a supplier’s commercial data.
Public audit records are being prepared.
Completed evidence will appear here as it becomes available.
The same checks as our supplier reports.
Completed checks are separated from missing measurements. An independent reviewer challenges design findings, while difficulty is measured separately with a model ladder.