Module aggregate 

v0.35: Consensus aggregation — the inference layer.

Given a target finding, the kernel can find other findings making similar claims, weight them by evidence quality (replication count, review state, time decay), and return a consensus confidence with a credible interval.

This is what turns Vela from “a database of claims” into “a reasoning surface over claims.” Other parts of the substrate describe what’s believed (findings) and what’s expected (predictions). This module describes what the field collectively holds — derived deterministically from canonical state, never stored.

Doctrine: aggregation is a derived view, not a kernel object. Same input frontier → same consensus result, byte-for-byte.

Structs

AggregateFilter

v0.38.2: filter constraints for consensus aggregation. Consensus computed without a filter blends all claim-similar findings — fine when “what does the field hold?” is the question, but wrong when the question is specifically “what does the field hold as causation?” or “what’s the consensus among RCT-grade evidence?”

ConsensusConstituent

One finding’s contribution to a consensus result.

ConsensusResult

Derived consensus over claim-similar findings.

Enums

WeightingScheme

How candidate findings are weighted when computing consensus.

Functions

consensus_for

Compute consensus over findings similar to target_id.

consensus_for_with_filter

v0.38.2: same as consensus_for, with a structured AggregateFilter applied to candidate findings before similarity is checked. Lets callers ask sharper questions: “what’s the consensus as intervention?” or “consensus among RCT-grade evidence only?”