Principle
Preserve disagreement that survives error correction
Consistency is not the goal when the differences themselves contain information.
When it fits
- A decision-hygiene process is treating all variation as something to eliminate.
When to avoid it
- Do not label avoidable inconsistency 'diversity' to escape process improvement; first test shared evidence and criteria.
Why it matters
After reviewers have seen the same evidence, corrected clear oversights and clarified criteria, preserve remaining disagreement when it reflects defensible perspectives or objectives. Record the reason for the split instead of forcing a false consensus. Noise reduction should remove accidental variability, not diversity of legitimate models.
An example
Two experts can agree on all facts yet differ on how much reversibility versus speed should matter; show that tradeoff instead of averaging their preference into an unexplained midpoint.
Check your result
Residual disagreement has an articulated substantive reason rather than an unexamined mistake or scale mismatch.
Keep this limit in mind
- Do not label avoidable inconsistency 'diversity' to escape process improvement; first test shared evidence and criteria.
Evidence and sources
The AMP proof of concept found that structured independent analysis, information sharing and private re-evaluation reduced self-identified errors or oversights while preserving meaningful differences in perspective.
The result is context-specific and does not prove the exact procedure improves accuracy in unrelated domains.
Analysis from multiple perspectives (AMP): Applying decision hygiene to analysis of musical structure · Abstract results