What problem does it solve? When multiple independent assessors score the same research candidate, their results often diverge, and someone must reconcile scores, pros, and cons into a single actionable insight without fabricating data or silently overriding disagreements. ## Core Features & Use Cases - Cross-Validation of Expert Scores: Computes per-dimension averages and standard deviations across assessors, classifying consensus (SD <= 1.0), moderate agreement, or significant divergence (SD > 2.0). - Semantic Insight Synthesis: Produces a 1-3 sentence insight explaining the why behind scores, preserving contradictions verbatim rather than resolving them silently. - Tree Propagation Output: Emits a structured JSON payload with snapshot hash, version IDs, consensus and divergence areas, and aggregated recommendations ready for upward propagation through the Idea Tree. - Use Case: After three assessment-screener agents checkpoint their evaluations of a catalyst candidate, the Lead agent dispatches this aggregator to merge them into one insight that informs whether to expand, prune, or vary that research branch. ## Quick Start Aggregate the checkpointed assessment artifacts for the current candidate into a cross-validated insight and return the propagation-ready JSON output.
