Crate attuned_infer 

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);
}

Structs

AxisEstimate

A single axis estimate with full metadata.

Baseline

Baseline statistics for a user.

BayesianUpdater

Bayesian state updater for a single axis.

DeltaAnalyzer

Delta analyzer for detecting deviations from baseline.

DeltaSignals

Signals derived from deviation analysis.

InferenceConfig

Configuration for the inference engine.

InferenceEngine

Main inference engine.

InferredState

Complete inferred state across multiple axes.

LinguisticExtractor

Fast linguistic feature extractor.

LinguisticFeatures

Linguistic features extracted from text.

Prior

Prior distribution for an axis.

Enums

InferenceSource

Source of an axis inference with full provenance.

Functions

infer

Quick inference function for simple use cases.

max_confidence_for_axis

Get maximum confidence for a specific axis based on research evidence strength.

word_count_confidence_factor

Scale confidence based on text length (word count).