Define a provider-neutral reviewer-diversity evaluation

[nothingnesses]

This issue records a proposal. Implementation requires a selected work brief and separate execution approval.

Background and problem

D-046 retains evaluation of reviewer diversity before any permanent policy. Different model or harness names do not establish independent evidence or fewer shared mistakes.

The library checkpoint supplies practical observations, not a controlled diversity comparison.

Proposal

Prepare a bounded protocol that asks whether reviewer diversity adds reproducible findings at an acceptable cost.

Distinguish these factors:

• Provider identity.
• Harness behaviour.
• Reviewer instructions.
• Access to earlier findings.

State which factors remain uncontrolled.

Scope and exclusions

The initial output is a proposed protocol, not a trial or provider requirement. Keep personal model preferences outside product policy.

Do not add a diversity rule to adoption or existing delivery contracts.

Acceptance criteria

• The protocol defines independent adjudication and comparable review targets.
• It records:
  • Contribution.
  • Overlap.
  • Disagreement.
  • Cost.
• It states confounders and limits before data collection.
• It separates identity diversity from actual independence.
• Execution requires separate approval.

Alternatives and open questions

Use comparable reviewer instructions to examine identity differences, or compare deliberately different review approaches. The latter mixes identity and task differences.

Prefer a narrow question under Ground decisions in evidence. A combined reviewer-count and diversity pilot requires explicit protocol approval.

References

• Decision report, D-046.
• methods, M6/M7.
• trial limitations.
• D-046.