attuned-core

Core types and traits for Attuned - human state representation for LLM systems.

Overview

This crate provides the foundational types for representing human state as interpretable vectors across 23 canonical axes. It includes:

• StateSnapshot - Point-in-time human state representation
• RuleTranslator - Converts state to LLM-consumable PromptContext
• Canonical Axes - 23 validated axes across 6 categories with governance metadata

Quick Start

use attuned_core::{StateSnapshot, RuleTranslator, Source};

// Create a state snapshot
let state = StateSnapshot::builder()
    .user_id("user_123")
    .source(Source::SelfReport)
    .axis("cognitive_load", 0.8)
    .axis("stress_level", 0.6)
    .axis("warmth", 0.7)
    .build()?;

// Translate to prompt context
let translator = RuleTranslator::default();
let context = translator.to_prompt_context(&state);

// Use context.guidelines in your LLM system prompt
println!("Tone: {}", context.tone);
println!("Verbosity: {:?}", context.verbosity);
println!("Guidelines: {}", context.guidelines);

Axis Categories

Design Principles

• No Actions: Produces context, never executes actions
• No Persuasion: No engagement optimization or nudging
• Transparent: All axes have declared governance metadata
• Self-Report Priority: User-declared state always overrides inference

License

Apache-2.0