Struct NoveltyComputer 

pub struct NoveltyComputer { /* private fields */ }

Novelty signal computer.

Computes novelty = exp(-λ × days_since_agent_init) where days_since_agent_init = (fact.created_at - session_start) / 86400.

Direction: higher novelty = fact was created close to session start; lower novelty = fact was created late in a long session. This corrects stale-episodic interference — facts that accumulated late in long sessions and pollute recall despite weak relevance.

Facts created before session start receive days = 0.0 → novelty = 1.0.

Implementations

impl NoveltyComputer

pub fn new(session_start: i64, decay_rate: f64) -> Self

Create a new NoveltyComputer.

Parameters

• session_start: Unix timestamp (seconds) when the agent session began.
• decay_rate: λ in exp(-λ × days). Larger values penalize late-session facts more strongly.

Examples

use zeph_memory::five_signal::novelty::NoveltyComputer;

let start = 1_700_000_000_i64;
let computer = NoveltyComputer::new(start, 0.1);

// Fact created at session start → novelty = 1.0
assert!((computer.compute(start) - 1.0).abs() < 1e-9);

// Fact created 10 days into the session → novelty ≈ exp(-1.0)
let ten_days_later = start + 10 * 86400;
let expected = (-1.0_f64).exp();
assert!((computer.compute(ten_days_later) - expected).abs() < 1e-9);

pub fn compute(&self, fact_created_at: i64) -> f64

Compute the novelty score for a fact with the given created_at timestamp.

Returns a value in (0.0, 1.0]. Facts created before or at session start return 1.0.