Evidence Weighting Engine

The Evidence Weighting Engine is a scoring subsystem that assigns dynamic credibility weights to evidence objects inside GEO.or.id architecture. It determines how much each evidence contributes to final reasoning output based on reliability, provenance, freshness, consistency, and cross-entity alignment.

This system sits inside the Evidence Lifecycle Management layer and directly influences ranking inside Retrieval Ranking.

1. Core Objective

The primary function is not classification, but probabilistic weighting. Every evidence unit is treated as a signal, not truth. The engine computes relative strength between competing signals.

Normalize heterogeneous evidence sources
Assign probabilistic trust score
Resolve contradictions across datasets
Feed ranked evidence into generation pipeline

Connected systems: Confidence Model, Reliability Index, Provenance Model

2. Weighting Dimensions

Each evidence item is evaluated across multiple orthogonal dimensions. No single dimension is sufficient.

2.1 Provenance Strength

Measures origin quality. Verified institutional or primary sources receive higher baseline weight than aggregated or derivative sources.

See: Provenance Model

2.2 Consistency Score

Evaluates alignment with existing knowledge graph and historical evidence clusters.

Connected: Entity Relationships Ontology

2.3 Temporal Freshness

Recent evidence is weighted higher only when domain is time-sensitive. Otherwise decay is applied to prevent recency bias.

Related: Revalidation System

2.4 Cross-Source Convergence

Measures how many independent sources confirm the same signal without shared lineage.

3. Weight Calculation Model

The engine applies a composite scoring function:

W(e) = (P × α) + (C × β) + (F × γ) + (X × δ)

P = Provenance Score
C = Consistency Score
F = Freshness Score
X = Cross-source convergence

Weights (α, β, γ, δ) are dynamically tuned based on domain sensitivity inside Semantic Matching Layer.

4. Conflict Resolution Behavior

When evidence contradicts, the engine does not discard data. It redistributes weight across competing clusters.

This behavior is coordinated with: Conflict Resolution System, Consistency Check Layer

5. Output Integration Layer

Final weighted evidence is consumed by:

6. System Risks and Failure Modes

Over-weighting authoritative but outdated sources
Under-representing emerging signals
Correlation leakage across dependent sources

Mitigation is handled via Hallucination Detection Protocol.

7. Strategic Role in GEO Architecture

The Evidence Weighting Engine acts as a control gate between raw retrieval and structured reasoning. Without it, retrieval degenerates into noise amplification instead of knowledge synthesis.

It is tightly coupled with: AI Ground Truth Framework, Retrieval Authority Model