What makes a message worth sharing?

Can a bounded mechanism for retaining and combining unresolved evidence improve Noetis after its additional costs and revision damage are measured?

Round 15 examines what makes communication between AI agents useful. It distinguishes messages that change an answer from messages that improve it, and studies how separate pieces of evidence can become valuable only when combined. Reproducible mathematical examples show why immediate message scoring and simple deletion tests can miss important dependencies. The round proposes controlled experiments to separate evidence presentation, peer recommendations, and joint information use. It builds on the earlier work on revision traces and evaluator uncertainty. These are research findings and experimental proposals, not new Noetis benchmark results.

• Behavioral influence does not establish a useful correction.
• Message value depends on the receiver's existing information and on other messages.
• A local communication benefit must still survive a complete-system comparison under measured resources.

• E15-A Exact diagnostics on constructed distributions Executed (diagnostic)
• E15-B Study A: support and recommendation fields (2x2) Proposed
• E15-C Study B: complementary withheld facts (optional) Proposed

None stated in the published round.

Proposed only. No production change.

• Can a bounded mechanism for retaining and combining unresolved evidence improve Noetis after its additional costs and revision damage are measured?

Builds on Round 14, Round 13, Round 12, Round 11, Round 10. Continued in Round 18, Round 17, Round 16.

Public summary only. The full notebook and research kit are kept internal.