Add data configs with LFS for large files

.gitattributes

@@ -62,3 +62,4 @@ models/champion_gen1000.py filter=lfs diff=lfs merge=lfs -text
 models/champion_gen42.py filter=lfs diff=lfs merge=lfs -text
 models/champion_gen52.py filter=lfs diff=lfs merge=lfs -text
 models/*.py filter=lfs diff=lfs merge=lfs -text
+data/comm_events/train.jsonl filter=lfs diff=lfs merge=lfs -text

README.md

@@ -1,602 +1,213 @@
 ---
-pretty_name: Ouroboros-Key Dataset
+pretty_name: KEY Neuroevolution Dataset
 license: mit
 tags:
-  - quine
   - neuroevolution
+  - lora
+  - genetic-algorithms
   - provenance
-  - rl
   - world-model
 language:
   - en
 ---
 
-# 🔑 Ouroboros-Key Dataset
+# 🔑 KEY: Neuroevolution Dataset
 
-This repository contains the **Ouroboros-Key dataset**, generated by the KEY production system that **converts existing models** (e.g., `.pt`, `.onnx`, and other compute-oriented formats) into **quine-replicant capable models**, then evolves and verifies their replication behavior. World-model backends (DreamerV3/RSSM) are optional; the dataset reflects the conversion + evolution pipeline across arbitrary architectures.
+**40,000+ logged events from real evolutionary runs** — every mutation, crossover, selection, and fitness evaluation.
 
----
-
-# 📦 Dataset Card
-
-## Dataset Viewer Status
-
-Hugging Face’s dataset viewer can’t load this dataset because **splits use different file formats**. This repo intentionally contains multiple JSONL streams with distinct schemas. Use direct file download or load specific files programmatically instead of relying on the viewer.
-
-This repository contains a dataset generated by the **Ouroboros-Key** production neural network (a.k.a. the KEY system described below). The dataset captures structured traces of evolution, inference, and provenance events emitted during runs of the system.
-
-## Dataset Summary
-
-- **Name**: Ouroboros-Key Dataset
-- **Source System**: Ouroboros-Key (production KEY quine-conversion + evolution system)
-- **Domain**: Quine conversion, evolution telemetry, inference traces, and provenance metadata
-- **Format**: Line-delimited JSON (JSONL)
-- **Primary Files**: `*.jsonl` logs in this repo (see file list below)
-
-## Supported Tasks
-
-- Evolution analytics
-- Provenance auditing
-- System telemetry analysis
-- Log-based debugging and visualization
-
-## Languages
-
-- English (metadata fields, comments, and annotations)
-
-## Source / Reference
-
-This dataset is **populated by the production Ouroboros-Key system**. The README content below describes the system architecture and behavior that produce these logs.
-
-If you need a code reference, the producing system is the KEY stack in:
-
-- `children/` (local project root)
-- `children/cascade_hyperlattice/` (DreamerV3 + provenance stack)
-- `key/` (core conversion + evolution engine)
-
-## What the Dataset Contains
-
-Each JSONL file captures a specific stream of events or metrics emitted by the system. Typical record types include:
-
-- Evolution events (e.g., generations, mutations, selections)
-- Fitness evaluations and performance metrics
-- Speciation and population dynamics
-- TUI/diagnostic events
-- Provenance/trace metadata emitted by CASCADE-LATTICE
-
-## File Index (Exported Logs)
-
-- `evolution_events.jsonl` — event bus stream (state changes, pressure, convergence, mutations)
-- `fitness.jsonl` — fitness evaluations and scores
-- `mutations.jsonl` — mutation operations and deltas
-- `selection.jsonl` — selection steps and tournament outcomes
-- `speciation.jsonl` — clustering/species assignments
-- `performance.jsonl` — runtime performance telemetry
-- `errors.jsonl` — system errors and warnings
-- `tui_events.jsonl` — UI and operator interaction events
-- `crossovers.jsonl` — crossover operations and metadata
-
-## Schema (Authoritative)
-
-The **authoritative schemas** are defined in the KEY data contracts (`DATA_CONTRACTS.md`) inside the **Ouroboros-Key** source repo.
-
-> 🔐 **Ouroboros-Key is a private neural network.**  
-> Get full source access and learn to convert your own models into self-replicating quines.
->
-> | Tier | Price | What you get |
-> |------|-------|------|
-> | **🔑 Source Access** | $100 one-time | Private repo invite, full codebase |
-> | **🤝 Hands-On Conversion** | $500 one-time | I do one model conversion with you + support |
-> | **🎤 Speaking Engagement** | $2,000 one-time | Talk at your company/conference, includes source access |
->
-> **[Sponsor on GitHub →](https://github.com/sponsors/Yufok1)**
->
-> 🎮 **[Try the demo](https://huggingface.co/spaces/tostido/Ouroboros)** — no purchase required
-
-Below is a concise summary of the primary log schemas represented in this dataset export.
-
-Each row is a JSON object. Common fields include (not all files include all fields):
-
-### Mutations (`mutations.jsonl`)
-- `timestamp` (epoch), `iso_time` — event time
-- `event` = `mutation`
-- `generation`, `parent_id`, `child_id`
-- `parent_fitness`, `mutation_rate`
-- `parent_traits`, `child_traits`, `deltas`, `mutated_traits`
-
-### Crossovers (`crossovers.jsonl`)
-- `timestamp`, `iso_time`, `event` = `crossover`
-- `generation`, `parent1_id`, `parent2_id`, `child_id`
-- `parent*_fitness`, `parent*_traits`, `child_traits`
-- `inheritance`, `contribution_p1`, `contribution_p2`
-
-### Selection (`selection.jsonl`)
-- `timestamp`, `iso_time`, `event` = `selection`
-- `generation`, `method`
-- `survivors`, `eliminated`, `elites_preserved`
-- `survivor_fitnesses`, `eliminated_fitnesses`
-
-### Speciation (`speciation.jsonl`)
-- `timestamp`, `iso_time`
-- `event` = `species_created|species_extinct|species_snapshot`
-- `generation`, `species_id`, `founder_id`, `initial_size`
-- `species_snapshot[]` (size/fitness/age/stagnation)
-
-### Fitness (`fitness.jsonl`)
-- `timestamp`, `iso_time`, `event` = `fitness_evaluation`
-- `generation`, `node_id`, `fitness_function`, `raw_fitness`
-- `components{...}`, `weights{...}`, `eval_time_ms`
-
-### Evolution Events (`evolution_events.jsonl`)
-- `event_id`, `event_type` (e.g., `state_change|pressure|convergence|vlm_inference|lora_mutation`)
-- `data{...}`, `timestamp`, `source`
-
-### TUI Events (`tui_events.jsonl`)
-- TUI-originated events (full schema in `DATA_CONTRACTS.md` — [get access](#schema-authoritative))
-
-### Performance / Errors (`performance.jsonl`, `errors.jsonl`)
-- Runtime and error telemetry emitted by the worker/TUI pipeline
-
-## How It’s Produced
-
-The Ouroboros-Key system logs structured JSONL events via its internal event bus and logging system (see `logger.py` and `bus.py` in the KEY stack). In production, logs are written under `data/logs/` and exported into this repository as a dataset snapshot.
-
-## Intended Use
-
-- Reproducing experiments and auditing evolutionary runs
-- Analyzing policy evolution dynamics
-- Debugging and performance profiling
-- Research on provenance, interpretability, and agent behavior
-
-## Limitations
-
-- Schema may evolve between runs or versions of Ouroboros-Key
-- The dataset is an export snapshot; canonical write locations are defined in `DATA_CONTRACTS.md` ([get access](#schema-authoritative))
-- Some fields are optional or component-specific
-- Logs reflect system behavior and may include intermittent failures
-
-## Data Sensitivity / Privacy
-
-- No human PII is intended to be collected
-- Operator actions may appear in `tui_events.jsonl`
-- Scrub or filter logs before external publication if needed
-
-## License
-
-This dataset inherits the repository license: **MIT**.
+KEY evolves LoRA adapters on frozen base models (MiniLM-L6, DreamerV3) using NEAT-style neuroevolution. This dataset captures the complete evolutionary history.
 
 ---
 
-# 🔑 System Reference (Ouroboros-Key)
-
-## 💰 Access + Learn
+## 🎮 Links
 
-**Get access to KEY and learn to convert your own models into self-replicating quines.**
+| | |
+|---|---|
+| **[🌌 Live Demo](https://huggingface.co/spaces/tostido/Cascade-Hyperlattice)** | Watch evolution in action |
+| **[🧠 Neural Network](https://huggingface.co/spaces/tostido/Ouroboros)** | The evolved model |
+| **[📂 Public Repo](https://github.com/Yufok1/Ouroboros-key-info)** | Architecture docs |
 
-| Tier | Price | What you get |
-|------|-------|------|
-| **🔑 Source Access** | $100 one-time | Private repo invite, full codebase |
-| **🤝 Hands-On Conversion** | $500 one-time | I do one model conversion with you + support |
-| **🎤 Speaking Engagement** | $2,000 one-time | Talk at your company/conference, includes source access |
-
-**[→ Sponsor on GitHub](https://github.com/sponsors/Yufok1)**
+---
 
-**[📂 Learn More (Public Repo)](https://github.com/Yufok1/Ouroboros-key-info)**
+## Loading the Dataset
 
-**DM on X: [@Toasteedo](https://x.com/Toasteedo)**
-
-## Architecture
+```python
+from datasets import load_dataset
 
-```
-┌─────────────────────────────────────────────────────────────────┐
-│                       PopulationManager                          │
-│                  (NEAT-style speciation)                         │
-└───────────────────────────┬─────────────────────────────────────┘
-                            │
-        ┌───────────────────┼───────────────────┐
-        │                   │                   │
-   ┌────▼────┐         ┌────▼────┐         ┌────▼────┐
-   │  Node   │         │  Node   │         │  Node   │  × N
-   │ traits  │         │ traits  │         │ traits  │
-   │ + brain │         │ + brain │         │ + brain │
-   └────┬────┘         └────┬────┘         └────┬────┘
-        │                   │                   │
-        └───────────────────┼───────────────────┘
-                            │
-                    ┌───────▼───────┐
-                    │  DreamerBrain │ (~200M params)
-                    └───────┬───────┘
-                            │
-    ┌───────────────────────┼───────────────────────┐
-    │                       │                       │
-┌───▼───────────┐    ┌──────▼──────┐    ┌──────────▼──────┐
-│ RSSM Encoder  │    │ GRU Core    │    │ Stochastic      │
-│ Observations  │    │ Deterministic│    │ Latent (32×32)  │
-│ → Latent      │    │ State (4096) │    │ Categorical     │
-└───────────────┘    └──────┬──────┘    └─────────────────┘
-                            │
-              ┌─────────────┴─────────────┐
-              │                           │
-       ┌──────▼──────┐             ┌──────▼──────┐
-       │ Policy Head │             │ Value Head  │
-       │ 5 layers    │             │ 5 layers    │
-       │ (EVOLVED)   │             │ (EVOLVED)   │
-       └──────┬──────┘             └──────┬──────┘
-              │                           │
-           action                   value estimate
+# Available configs:
+ds = load_dataset("tostido/key-data", "comm_events")      # 16,968 rows - pod communication
+ds = load_dataset("tostido/key-data", "crossovers")       # 8,878 rows - breeding events
+ds = load_dataset("tostido/key-data", "selection")        # 4,266 rows - tournament selection
+ds = load_dataset("tostido/key-data", "mutations")        # 3,848 rows - mutation events
+ds = load_dataset("tostido/key-data", "fitness")          # 2,121 rows - fitness evaluations
+ds = load_dataset("tostido/key-data", "performance")      # 2,121 rows - runtime telemetry
+ds = load_dataset("tostido/key-data", "errors")           # 2,070 rows - errors/warnings
+ds = load_dataset("tostido/key-data", "evolution_events") # event bus stream
 ```
 
-## DreamerV3 Size Presets
-
-| Size | Deter | Hidden | Params | Use Case |
-|------|-------|--------|--------|----------|
-| XS | 512 | 512 | ~10M | Development/testing |
-| S | 1024 | 1024 | ~22M | Basic tasks |
-| M | 2048 | 2048 | ~100M | Minecraft survival |
-| **L** | **4096** | **4096** | **~200M** | **Diamond mining (current)** |
-| XL | 8192 | 8192 | ~400M | Full game mastery |
-
-**Current config: L (200M params)** - Same size DreamerV3 used to obtain diamonds in Minecraft.
+---
 
-## Features
+## Example: Evolving Semantic Similarity
 
-### Evolution Engine
-- **NEAT-style speciation**: Genetic distance clustering
-- **Fitness sharing**: Prevents monoculture
-- **Tournament selection**: With elitism preservation
-- **Buffered logging**: High-performance JSONL logging (50x faster than unbuffered)
-- **Config-driven sizes**: Switch between XS/S/M/L/XL via config.json
+**Task**: Adapt MiniLM embeddings to preserve semantic relationships
 
-### Brain Types
+**Test Pair**: "The cat sat on the mat" ↔ "A feline rested on the rug"
 
-| Brain | Parameters | Description |
-|-------|------------|-------------|
-| `DreamerBrain` | **~200M (L)** | World model with imagination + policy/value heads |
-| `EmbeddingBrain` | ~99K | Lightweight embedding brain (disabled by default) |
-| `MLPBrain` | ~1K | Simple feedforward baseline |
+| Generation | Cosine Similarity | Fitness |
+|------------|-------------------|---------|
+| 0          | 0.42 (random)     | 0.35    |
+| 50         | 0.76              | 0.64    |
+| 100        | 0.89              | 0.82    |
 
-### DreamerBrain (World Model)
+The evolved adapter learned to preserve semantic similarity while improving output quality.
 
-DreamerBrain uses the DreamerV3 RSSM architecture to imagine future trajectories:
+---
 
-```
-┌─────────────────────────────────────────────────────────────────┐
-│                         DreamerBrain                             │
-│                                                                  │
-│  ┌──────────────────────────────────────────────────────────┐   │
-│  │   DreamerV3 RSSM (World Model)                            │   │
-│  │   - 4096-dim deterministic state (L size)                 │   │
-│  │   - 32×32 categorical stochastic state                    │   │
-│  │   - Imagines 15-step futures                              │   │
-│  └──────────────────────┬───────────────────────────────────┘   │
-│                         │                                        │
-│              ┌──────────┴──────────┐                             │
-│              │                     │                             │
-│       ┌──────▼──────┐       ┌──────▼──────┐                      │
-│       │ Policy Head │       │ Value Head  │                      │
-│       │ (EVOLVED)   │       │ (EVOLVED)   │                      │
-│       │ 5 layers    │       │ 5 layers    │                      │
-│       └──────┬──────┘       └──────┬──────┘                      │
-│              │                     │                             │
-│           action              value estimate                     │
-└─────────────────────────────────────────────────────────────────┘
-```
+## What Gets Evolved
 
-Unique capabilities:
-- **Imagination rollouts**: See all possible futures before acting
-- **Decision matrix visualization**: Each future trajectory rendered in Rerun
-- **JAX-native**: GPU-accelerated world model inference
-- **Evolved heads only**: World model is pretrained, only policy/value evolve
-
-### Perception (MiniLM)
-- **all-MiniLM-L6-v2**: Frozen 22M param sentence transformer
-- **384-dimensional** embeddings
-- **Evolved projection**: Transforms base embeddings
-
-### Generation (SmolLM) - Optional
-- **SmolLM-135M**: Lightweight generative model
-- **Cross-modal**: Embeddings influence generation temperature
-- Can be disabled for embedding-only mode
-
-### Provenance & Tracking
-- **cascade-lattice integration**: IPFS auto-logging of all events
-- **SwarmLattice**: Fire-and-forget spawn/exploration tracking
-- **Genesis root**: `89f940c1a4b7aa65` (common anchor for all provenance)
-- **Embedding hashes**: Tensor content hashing for reproducibility
-
-### HOLD System (Inference-Level Halt)
-- **Architectural halt**: Blocks inference until human/policy resolution
-- **CASCADE-LATTICE required**: No cascade = No HOLD
-- **Merkle-linked decisions**: Every hold point and resolution recorded
-- **Decision matrix exposed**: action_probs, value, imagined futures visible
+KEY freezes the base model and evolves only the adapter:
 
-```python
-from hold import Hold, HoldPoint
-
-# In any brain's forward pass:
-resolution = Hold.get().yield_point(
-    action_probs=probs,
-    value=value,
-    observation=obs,
-    brain_id="my_brain",
-)
-# Blocks until: accept(), override(action), or timeout
-final_action = resolution.action
 ```
+┌──────────────────────────────────────┐
+│         Evolvable Brain              │
+│  ┌────────────────────────────────┐  │
+│  │   Base Model (FROZEN)          │  │  ← MiniLM (22M) or DreamerV3 (200M)
+│  └─────────────┬──────────────────┘  │
+│                ▼                     │
+│  ┌────────────────────────────────┐  │
+│  │   LoRA Adapter    (~12K)       │  │  ← EVOLVED
+│  │   Projection Head (~99K)       │  │  ← EVOLVED
+│  └────────────────────────────────┘  │
+└──────────────────────────────────────┘
 
-### Quine Manifold (Branching Realities)
-- **Brain instantiation**: Not simulation - actual brain clones
-- **Exponential branching**: 8 actions → 8^depth parallel brains
-- **Collapse strategies**: max_value, max_prob, random, or HOLD (manual)
-- **Defensive cloning**: High uncertainty triggers branch spawning
-
-```python
-from manifold import QuineManifold
-
-manifold = QuineManifold(brain)
-manifold.expand(depth=2)  # 64 parallel realities
-best = manifold.collapse('max_value')  # Or HOLD to pick manually
+Total evolved parameters: ~111K (vs 22M-200M frozen)
 ```
 
-### Collective Memory (Swarm Intelligence)
-- **SEEK on observe**: Every forward() queries CASCADE for similar experiences
-- **Latent similarity**: Cosine search over historical latent states
-- **Soft bias**: Retrieved experiences gently influence action selection
-- **Cross-agent learning**: Agents share experiences through CASCADE tapes
-
-### Glass Box Visualization
-- **Rerun.io integration**: Real-time visualization of CASCADE events
-- **Full computational transparency**: Hidden states, weight snapshots, computation paths
-- **Single source of truth**: Rerun shows what CASCADE cryptographically proves
-- **DatasetUnity pipeline**: Evolved embedding-based dataset bridging with Kleene fixed-point matching
-
-### Metrics & Logging
-- **wandb integration**: Full experiment tracking
-- **sklearn metrics**: Silhouette, Davies-Bouldin, Calinski-Harabasz
+---
 
-### Classic Features
-- **Violation Pressure**: Trait deviation from stability envelopes
-- **Attractors**: Kleene-style fixed point search
-- **Convergence**: Weighted trait merging with mutation
+## Fitness Functions
 
-## Installation
+What evolution optimized for (from `fitness.jsonl`):
 
-```bash
-# Clone (public info repo - full source requires sponsorship)
-git clone https://github.com/Yufok1/Ouroboros-key-info
-cd Ouroboros-key-info
+### AdapterFitness (Interface Quality)
+- **Preservation (40%)**: Does adapter maintain semantic structure?
+- **Signal Quality (30%)**: Is output well-conditioned? (not collapsed/exploded)
+- **Consistency (30%)**: Similar inputs → similar outputs?
 
-# Install dependencies
-pip install -r requirements.txt
+### EmbeddingKleeneFitness (Semantic Convergence)
+- **Coherence**: Similar pairs should have high cosine similarity
+- **Separation**: Dissimilar pairs should be far apart
+- **Convergence**: Embedding variance stays bounded
 
-# For GPU acceleration (CUDA 11.8+)
-pip install torch torchvision --index-url https://download.pytorch.org/whl/cu118
-```
+### DreamerFitness (World Model Quality)
+- **Prediction**: How well does imagination match reality?
+- **Stability**: Do trajectories stay bounded?
+- **Reward**: Can the model anticipate outcomes?
 
-## Quick Start
+---
 
-### 1. Configure (config.json)
+## Schema Reference
 
+### `mutations.jsonl`
 ```json
 {
-  "dreamer_brain": {
-    "enabled": true,
-    "size": "L",
-    "deter_dim": 4096,
-    "hidden_dim": 4096,
-    "policy_layers": 5,
-    "value_layers": 5
-  },
-  "fitness": {
-    "function": "dreamer"
-  },
-  "population": {
-    "size": 6
-  }
+  "timestamp": 1737403521.234,
+  "event": "mutation",
+  "generation": 42,
+  "parent_id": "node_abc123",
+  "child_id": "node_def456",
+  "parent_fitness": 0.72,
+  "mutation_rate": 0.1,
+  "mutated_traits": ["exploration", "caution"],
+  "deltas": {"exploration": 0.05, "caution": -0.02}
 }
 ```
 
-### 2. Run TUI
-
-```bash
-python app.py
-```
-
-### 3. Glass Box Mode (Compiled Agents)
-
-```python
-from children.champion_gen1000 import CapsuleAgent
-
-# Launch with visualization
-agent = CapsuleAgent(observe_visual=True)  # Spawns Rerun viewer
-
-# Every inference is visualized with CASCADE proof
-result = agent.forward({"traits": {"x0": 0.5, "x1": 0.3}})
-print(f"Merkle: {result['_merkle_root']}")
-
-# Dataset bridging
-unity = agent.data_unity
-old = unity.embed(unity.load_dataframe(df_2023, "2023"))
-new = unity.embed(unity.load_dataframe(df_2024, "2024"))
-matches = unity.bridge(old, new, min_confidence=0.7)
+### `crossovers.jsonl`
+```json
+{
+  "event": "crossover",
+  "generation": 42,
+  "parent1_id": "node_abc",
+  "parent2_id": "node_xyz",
+  "child_id": "node_new",
+  "parent1_fitness": 0.72,
+  "parent2_fitness": 0.68,
+  "contribution_p1": 0.55
+}
 ```
 
-**TUI Controls:**
-- `D` - Dashboard (pulse, VP gauges, event stream)
-- `P` - Population (start/stop evolution, view nodes)
-- `C` - Config (edit any setting with descriptions)
-- `L` - Logs (JSONL log viewer)
-- `X` - Explorer (causation graph)
-- `N` - Neural (brain inspector)
-- `Q` - Quit
-
-## Configuration
-
-| Parameter | Default | Description |
-|-----------|---------|-------------|
-| `dreamer_brain.enabled` | `true` | Enable DreamerBrain world models |
-| `dreamer_brain.size` | `L` | Size preset (XS/S/M/L/XL) |
-| `dreamer_brain.deter_dim` | `4096` | Deterministic state dimension |
-| `dreamer_brain.hidden_dim` | `4096` | MLP hidden layer size |
-| `fitness.function` | `dreamer` | Fitness function (dreamer, benchmark) |
-| `population.size` | `6` | Population size |
-| `evolution.generations` | `1000` | Max generations |
-
-## Hardware Requirements
-
-| Size | VRAM | RAM | Time/Gen | Notes |
-|------|------|-----|----------|-------|
-| XS (512) | ~2GB | 8GB | ~10s | Development/testing |
-| S (1024) | ~4GB | 8GB | ~30s | Basic tasks |
-| M (2048) | ~8GB | 16GB | ~60s | Minecraft survival |
-| **L (4096)** | **~12GB** | **16GB** | **~80s** | **Diamond mining (current)** |
-| XL (8192) | ~24GB | 32GB | ~120s | Full game mastery |
-
-## Project Structure
-
-```
-key/
-├── app.py              # Textual TUI - main entry point
-├── run.py              # Headless evolution runner
-├── brain.py            # Brain interface + EmbeddingBrain
-├── dreamer_brain.py    # DreamerBrain - world model with imagination
-├── dreamerv3/          # DreamerV3 RSSM implementation (JAX)
-├── pod_brain.py        # PodBrain - multi-model with LoRA adapter
-├── pod.py              # Pod communication (Swarm broadcasts)
-├── node.py             # Node organism with traits + brain
-├── population.py       # NEAT-style population manager
-├── mobile.py           # Async swarm with SwarmLattice
-├── fitness.py          # Pluggable fitness interface
-├── fitness_comm.py     # Communication fitness (embedding similarity)
-│
-├── # Dataset Bridging
-├── bridge.py           # DatasetBridge with provenance
-├── kleene.py           # Fixed-point matching + batch_compare()
-│
-├── # Provenance Infrastructure
-├── swarm_lattice.py    # Fire-and-forget spawn/exploration tracking
-│
-├── checkpoints.py      # Save/load evolution state
-├── bus.py              # Event bus for logging
-├── logger.py           # JSONL logging system
-├── config.json         # Runtime configuration
-├── children/           # Compiled agent capsules
-└── requirements.txt    # Dependencies
+### `fitness.jsonl`
+```json
+{
+  "event": "fitness_evaluation",
+  "generation": 42,
+  "node_id": "node_abc123",
+  "fitness_function": "AdapterFitness",
+  "raw_fitness": 0.823,
+  "components": {
+    "preservation": 0.85,
+    "signal": 0.79,
+    "consistency": 0.84
+  },
+  "eval_time_ms": 45.2
+}
 ```
 
-## Usage Examples
-
-### Creating Nodes with DreamerBrain
-
-```python
-from brain import create_brain
-from node import Node
-
-# Create a DreamerBrain (world model with imagination)
-brain = create_brain('dreamer')
-
-# Create a Node with the brain
-node = Node(
-    traits={"exploration": 0.7, "caution": 0.3},
-    brain=brain
-)
-
-# Forward pass - get action + value
-output = brain.forward({"obs": observation})
-action = output["action"]       # Discrete action
-value = output["value"]         # Estimated value
-latent = output["latent"]       # 1536-dim latent state
-
-# Imagination - see possible futures
-futures = brain.imagine(n_trajectories=5, horizon=15)
-for i, trajectory in enumerate(futures):
-    print(f"Future {i}: value={trajectory[-1]['value']:.3f}")
-
-# Evolution - mutate policy/value heads
-child = brain.mutate(rate=0.1)
+### `selection.jsonl`
+```json
+{
+  "event": "selection",
+  "generation": 42,
+  "method": "tournament",
+  "survivors": ["node_a", "node_b", "node_c"],
+  "eliminated": ["node_d", "node_e"],
+  "elites_preserved": 2
+}
 ```
 
-### Creating Nodes with PodBrains
-
-```python
-from pod_brain import create_pod_brain
-from node import Node, create_population
+---
 
-# Create a PodBrain (multi-model with LoRA adapter)
-brain = create_pod_brain(lora_rank=16, voice_enabled=False)
+## Why Evolve Instead of Gradient Descent?
 
-# Create a Node with the brain
-node = Node(
-    traits={"creativity": 0.7, "focus": 0.5},
-    brain=brain
-)
+Neuroevolution works when:
+- ✅ Your objective **isn't differentiable** (human preference, discrete outputs)
+- ✅ You want **population diversity** (speciation prevents local optima)
+- ✅ You're optimizing for **interface quality**, not task loss
+- ✅ You need **full auditability** (every mutation logged with provenance)
 
-# Use the brain via node.think()
-output = node.think({"text": "hello world"})
-embedding = output["embedding"]  # (384,) evolved embedding
+---
 
-# Evolution
-child = node.mutate(rate=0.1, mutate_brain=True)
-```
+## FAQ
 
-### Population with PodBrains
+**Q: What's a "quine brain"?**
+> A brain that can serialize its weights → mutate → deserialize. This enables genetic algorithms to evolve neural networks. Think "self-modifying adapter."
 
-```python
-from pod_brain import create_pod_brain
-from node import create_population
-
-# Brain factory
-def make_pod():
-    return create_pod_brain(lora_rank=8, voice_enabled=False)
-
-# Create population
-pop = create_population(
-    size=20,
-    trait_keys=["speed", "strength", "perception"],
-    brain_factory=make_pod
-)
-
-# All nodes have PodBrains
-for node in pop:
-    print(f"Node {node.id}: brain={node.brain.id[:8]}")
-```
+**Q: Why not just use backprop?**
+> Backprop requires differentiable objectives. Evolution works with any fitness function: human ratings, game scores, discrete metrics.
 
-### Pod Communication (Swarm Broadcasts)
+**Q: Is this real data?**
+> Yes. This dataset contains 40K+ events from actual evolutionary runs. The $100 source tier includes full code — money back if it doesn't work.
 
-```python
-from pod import Pod, Swarm
+---
 
-# Create swarm of communicating pods
-swarm = Swarm(size=5, embed_dim=384)
+## 🔐 Get Full Source Access
 
-# Initialize with diverse states
-for pod, topic in zip(swarm.pods, ["alpha", "beta", "gamma", "delta", "epsilon"]):
-    pod.sense(topic)
+| Tier | Price | What You Get |
+|------|-------|--------------|
+| **🔑 Source Access** | $100 | Full codebase, private repo invite |
+| **🤝 Hands-On** | $500 | I evolve adapters for your domain + support |
+| **🎤 Speaking** | $2,000 | Talk at your company on gradient-free optimization |
 
-# Run convergence (broadcast, consensus, align)
-alignments = swarm.converge(rounds=5, align_rate=0.2)
-print(f"Final alignment: {alignments[-1]:.3f}")
+### **[→ Sponsor on GitHub](https://github.com/sponsors/Yufok1)**
 
-# Get leader (closest to consensus)
-leader = swarm.get_leader()
-```
+---
 
-## Swarm Provenance
+## Contact
 
-Track agent spawning with fire-and-forget logging:
+**DM on X: [@Toasteedo](https://x.com/Toasteedo)**
 
-```python
-from swarm_lattice import SwarmLattice
-
-lattice = SwarmLattice(log_dir='runs/swarm')
-record = lattice.record_spawn(
-    parent_id='genesis',
-    child_id=child_node.id,
-    mutation_rate=0.1,
-    generation=5,
-)
-print(f"Spawn merkle: {record.merkle_root}")
-```
+---
 
 ## License
 

data/comm_events/train.jsonl

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:3d4164fa1ab16f55a5410723990dce9fe899288ad59899d6cb7a5684dbc12f5a
+size 13604754

data/evolution_events/train.jsonl

@@ -0,0 +1 @@
+{"event_id": "evt_f6b9cc0f828f", "timestamp": 1767139039.1184373, "iso_time": "2025-12-31T00:57:19.118647", "type": "fitness_improvement", "title": "New Best: 0.5150", "detail": "Node 9ad6804a achieved new best fitness", "data": {"traits": {"x0": 0.3745401188473625, "x1": 0.9507143064099162, "x2": 0.7319939418114051, "x3": 0.5986584841970366, "x4": 0.15601864044243652, "x5": 0.15599452033620265, "x6": 0.05808361216819946, "x7": 0.8661761457749352, "x8": 0.6011150117432088, "x9": 0.7080725777960455}, "old_best": 0.5149}, "generation": 0}