Crate thermogram 

Thermogram

A plastic memory capsule with 4-temperature tensor states (hot/warm/cool/cold), rule-governed deltas, and hash-chained auditability.

Core Concept

Traditional storage is either:

• Mutable (databases, files) - fast but no audit trail
• Immutable (Engram, Git) - auditable but can’t evolve

Thermogram combines both with 4 thermal states that mimic biological memory:

4-Temperature Model

Temperature	Analog	Decay Rate	Behavior
Hot	Working memory	Fast (0.1/tick)	Volatile, immediate task
Warm	Short-term	Medium (0.01/tick)	Session learning, persists
Cool	Procedural/skill	Slow (0.001/tick)	Expertise, long-term
Cold	Core identity	Glacial (0.0001/tick)	Personality backbone

Bidirectional Flow

HOT ←→ WARM ←→ COOL ←→ COLD
 ↑        ↑        ↑        ↑
fast    medium   slow    glacial
decay   decay    decay   decay

• Cement forward: Reinforcement strengthens, promotes to colder layer
• Degrade backward: Lack of use weakens, demotes to hotter layer

Features

• Delta chain (append-only) - fast writes with audit trail
• Plasticity rules (STDP-like) - when to update vs create new
• Hash chain - cryptographic audit trail
• Thermal transitions - automatic promotion/demotion based on strength
• Colonies - multiple thermograms per mesh that grow, split, merge
• Distillation - share semantic deltas across instances
• Engram export - archive without deletion

Use Cases

1. LLM Activation Mining - Cluster centroids evolve as new patterns discovered
2. Agent Memory - Episodic memory with replay and consolidation
3. Knowledge Graphs - Concepts strengthen/weaken over time
4. Neural Weights - Save checkpoints with full training history
5. Personality Substrate - Synaptic weights that define identity

Example

use thermogram::{Thermogram, PlasticityRule, Delta, PackedSignal, Signal};

// Create new thermogram for neural weight storage
let mut thermo = Thermogram::new("neural_weights", PlasticityRule::stdp_like());

// Apply delta (weight vector update using PackedSignal values)
let weights: Vec<PackedSignal> = (0..64).map(|i| PackedSignal::pack(1, i as u8, 1)).collect();
let delta = Delta::update(
    "layer_0",
    weights,
    "training",
    Signal::positive(204), // ~0.8 strength
    thermo.dirty_chain.head_hash.clone(),
);
thermo.apply_delta(delta)?;

// Read current state (hot tensors take priority)
let stored = thermo.read("layer_0")?;

// Consolidate (crystallize hot → cold, prune weak)
thermo.consolidate()?;

// Export to JSON
thermo.export_to_json("neural_knowledge_v1.json")?;

Re-exports

pub use crate::colony::ColonyConfig;

pub use crate::colony::ColonyConsolidationResult;

pub use crate::colony::ColonyStats;

pub use crate::colony::ThermogramColony;

pub use crate::consolidation::ConsolidatedEntry;

pub use crate::consolidation::ConsolidationPolicy;

pub use crate::consolidation::ConsolidationResult;

pub use crate::consolidation::ConsolidationTrigger;

pub use crate::core::CrystallizationResult;

pub use crate::core::ThermalConfig;

pub use crate::core::ThermalState;

pub use crate::core::Thermogram;

pub use crate::core::ThermogramStats;

pub use crate::delta::Delta;

pub use crate::delta::DeltaType;

pub use crate::distillation::apply_delta_to_embedding;

pub use crate::distillation::cosine_similarity;

pub use crate::distillation::distill_learning;

pub use crate::distillation::DeltaBatch;

pub use crate::distillation::DeltaSource;

pub use crate::distillation::DistillationConfig;

pub use crate::distillation::SemanticDelta;

pub use crate::distillation::ThermogramSnapshot;

pub use crate::embedded_snn::EmbeddedSNN;

pub use crate::embedded_snn::EmbeddedSNNConfig;

pub use crate::error::Error;

pub use crate::error::Result;

pub use crate::hash_chain::HashChain;

pub use crate::plasticity::PlasticityRule;

pub use crate::plasticity::UpdatePolicy;

pub use crate::plasticity_engine::NeuromodState;

pub use crate::plasticity_engine::NeuromodSyncConfig;

pub use crate::plasticity_engine::PlasticityEngine;

pub use crate::ternary::TernaryWeight;

Modules

codec

Binary Codec — .thermo v2 format

colony

Thermogram Colony

consolidation

Consolidation - Dirty → Clean state transitions

core

Core Thermogram implementation

delta

Delta - Represents a change to the Thermogram state

distillation

Distillation - Semantic Delta Sharing

embedded_snn

Embedded SNN - Small spiking neural network for Thermogram plasticity

error

Error types for Thermogram

export

Export - Archive Thermogram to Engram

hash_chain

Hash Chain - Cryptographic audit trail for deltas

plasticity

Plasticity Rules - STDP-like policies for memory updates

plasticity_engine

Plasticity Engine - SNN-based delta generation

ternary

Ternary Weight type for embedded SNN plasticity

Structs

PackedSignal

Compact 1-byte signal: [pol:2|mag:3|mul:3]

Signal

Signal: polarity × magnitude × multiplier (s = p × m × k)

Enums

Polarity

Polarity of a neural signal — strictly {-1, 0, +1}