Research round ten: evidence exchange and selective revision
From learning procedures to trusting corrections: what communication and revision protocol makes a correction trustworthy enough to reuse?
The question this round asks
Round nine asked whether a system can learn reusable procedures from verified experience. Round ten concentrates on the step before that: the interaction that should produce a trustworthy correction. Can selective, evidence-conditioned revision outperform both a well-equipped single agent and a simpler shared-context system at the same total cost?
The candidate hypothesis
H10 proposes that a modular system with independent initial hypotheses, recoverable evidence, explicit claim dependencies, and selective checking will outperform equally resourced systems that rely on unrestricted discussion or repeated full-answer rewriting. It is deliberately narrow, and it comes with five falsifiable predictions and the rival explanations, including extra compute and stronger verification, that the experiment must rule out.
The central experiment
Six comparison arms, from a strong single agent to multiple agents with a shared evidence workspace, run over independently authored task families with controlled reasons to update. The metrics separate correction from damage: a high gross correction count can coexist with worse final performance, so correction rate and harmful revision rate are reported separately.
The honest limits
This round is a critical review and proposed experimental program. No new Encevra benchmark, model training, or performance result was produced in it, and the central hypothesis is provisional. The debate literature is genuinely mixed, close prior art already exists for verified shared context and belief revision, and a credible research program must be able to conclude that a simpler mechanism wins.