Confidence does not reliably track accuracy for subjective judgments

Claim: Confidence does not reliably track accuracy for subjective judgments. The confidence-accuracy relationship is task-dependent and often weak. An owner who is "very confident" they understand their market is no more likely to be correct than one who is "somewhat confident."

Confidence: Industry-consensus across the calibration literature.

Why this matters for Candid: Forbids any widget question that uses confidence as a proxy ("how confident are you that..." / "rate your knowledge of..."). Confidence ratings carry no incremental information beyond the underlying observation. See R1 — Convert every judgment into an observation or counting task.