Collective-intelligence chess
Protocol summary · discussion draft · 9 October 2026. Not frozen, not a registered protocol, and no results reported. Design choices below require agreement before data collection.
Claim and comparison
Compare simple plurality, reputation weighting, and the fuller protocol using the same participant pool. Group performance against the strongest member identified before testing is a separate hypothesis.
Design under discussion
Use controlled positions, balanced order and independent first choices before discussion. Freeze the reputation algorithm before scoring. Participants and AI coordinators must not use chess engines to generate or evaluate moves; a fixed engine may score choices after submission.
Measures and failure condition
Report move-quality loss, serious errors, time, and comparison with the pre-identified strongest member. If adequately precise estimates exclude the preregistered minimum useful gain over simpler baselines, narrow or reject the institutional-gain claim for this setting. A wide interval is inconclusive, not proof of no effect.
Before recruitment
Recover and reconcile the complete current protocol; freeze the primary outcome, strongest-member definition, sample size, ordering, engine settings, exclusions and stopping rule. Version summaries are not a frozen preregistration.
Control, challenge and stopping
Organizers choose tasks and control records, so their choices can bias the result. Preserve disagreements and protocol changes. Unresolved consent problems, data leakage or compromised scoring should pause the pilot. A negative result must remain reportable. Independent review arrangements and a successor operator are still to be agreed.
Challenge this protocol