attuned-infer

Fast, transparent inference of human state axes from natural language.

This crate provides declared, bounded, and subordinate inference - all estimates are auditable, confidence-bounded, and always overridable by self-report.

Design Principles

1. Declared: Every inference includes its source and reasoning
2. Bounded: Inferred values have capped confidence (default 0.7)
3. Subordinate: Self-report always overrides inference
4. Fast: Sub-millisecond inference for real-time use

Architecture

[Message] ──→ [LinguisticFeatures] ──→ [AxisEstimate]
                    ~100μs                    │
                                              ▼
[History] ──→ [DeltaAnalysis] ──────→ [BayesianUpdater] ──→ [StateEstimate]
                  ~500μs                      │
                                              ▼
[Self-Report] ─────────────────────→ [Override (σ² → 0)]

Example

use attuned_infer::{InferenceEngine, InferredState};

let engine = InferenceEngine::default();

// Infer state from a message
let state = engine.infer("I need help ASAP, this is urgent!!!");

// All estimates include source and confidence
if let Some(urgency) = state.get("urgency_sensitivity") {
    println!("urgency: {:.2} (confidence: {:.2}, source: {:?})",
        urgency.value, urgency.confidence, urgency.source);
}