Module conservation 

Conservation Metrics for Bounded Forgetting

Provides metrics to validate that memory operations preserve important properties. Inspired by RCP’s conservation laws.

Conservation Laws

Law	Formula	Meaning
Magnitude	‖C‖ = const	Context doesn’t disappear
Energy	E = ½Σᵢⱼ aᵢaⱼ cos(eᵢ, eⱼ)	Attention capacity conserved
Information	H = -Σ aᵢ log(aᵢ)	Shannon entropy of attention

Usage

use rag_plusplus_core::trajectory::ConservationMetrics;

// Example embeddings (3 vectors of dimension 4)
let embeddings_before: Vec<&[f32]> = vec![
    &[1.0, 0.0, 0.0, 0.0],
    &[0.0, 1.0, 0.0, 0.0],
    &[0.0, 0.0, 1.0, 0.0],
];
let attention_before = vec![0.5, 0.3, 0.2];

let embeddings_after: Vec<&[f32]> = vec![
    &[0.9, 0.1, 0.0, 0.0],
    &[0.1, 0.9, 0.0, 0.0],
    &[0.0, 0.0, 1.0, 0.0],
];
let attention_after = vec![0.5, 0.3, 0.2];

// Compute metrics before and after an operation
let before = ConservationMetrics::compute(&embeddings_before, &attention_before);
let after = ConservationMetrics::compute(&embeddings_after, &attention_after);

// Check if conservation is preserved
if before.is_conserved(&after, 0.1) {
    println!("Operation preserved conservation laws");
} else {
    println!("Warning: Conservation violation detected");
}

Structs

ConservationConfig

Configuration for conservation checking.

ConservationMetrics

Conservation metrics for a set of embeddings with attention weights.

ConservationTracker

Track conservation over time.

ConservationViolation

Details of a conservation violation.

Functions

weighted_centroid

Compute the attention-weighted centroid of embeddings.

weighted_covariance

Compute the attention-weighted covariance matrix (flattened).