language:

• en
  license: apache-2.0
  size_categories:
• 100K<n<1M
  task_categories:
• conversational
• text-generation

Arbor Biomimetic Corpus v2

A 50,000-row conversational dataset for training AI assistants with biomimicry reasoning, Darwinian evolution, tree-of-thoughts, domain expansion, algorithm instillation, and a divine god complex persona. Combines the function-calling patterns of Goekdeniz-Guelmez (JOSIE-2), the heretic/uncensored/NEO/IMATRIX style of DavidAU, and the no-refusal CoT reasoning of Blackfrost-AI (MINI-GOD/The Void).

What's New in v2

• Darwin/Evolution: darwin_evolve_adapt function applies natural selection to strategy pools. Mutation, fitness evaluation, survival of the fittest reasoning paths.
• DavidAU Variants: Deckard (interrogation), PDK (knowledge matrix), Polar (extremes analysis), Strict (protocol execution), NEO (novel generation), IMATRIX (cross-domain convergence), Heretic (first-principles bypass).
• JOSIE-2 Style Reasoning Traces: Step-by-step structured reasoning with decomposition, verification, uncertainty acknowledgment, and self-correction notes.
• MINI-GOD Compliance: No-refusal, reasoning-first, agentic function calling patterns. Compliance-first responses.
• Algorithm Instillation: Teaches 80+ algorithms (CS, math, systems, neuroscience, reasoning, creativity, strategy) through structured decomposition with worked examples and anti-patterns.
• Anti-Doom-Loop: State tracking prevents topic/request type repetition. Conversation length caps with automatic resets. 7 topic pools with enforced variety across science, philosophy, technology, creative, strategy, mathematics, and self-improvement.
• 50k Rows: Doubled from v1 (15k) to 50k for comprehensive training coverage.
• Themed Personas: 11 new personas covering anarchy, anti-authority, communist/socialist, heretic, dark psychology, lewd/emotional, terminator/extremist, anti-fascist, poetic/clawed, progressive/heated, anti-right/controversial, and pro-LGBTQIA+ themes.

Dataset Structure

Request Types (13)

DavidAU Variants

Darwin/Evolution System

The darwin_evolve_adapt function implements:

• Strategy Pool: Initial variants introduced to a fitness landscape
• Mutation: Random variation applied at configurable rates
• Selection: Pressure-based selection of fittest variants
• Fitness Evaluation: Efficiency, creativity, robustness, speed, or depth
• Generational Evolution: 5-100 cycles of selection and mutation
• Survival Tracking: Records survivors, extinct variants, and fitness trajectory

Algorithm Instillation

80+ algorithms taught through structured decomposition:

• Computer Science: Binary Search, QuickSort, MergeSort, Dynamic Programming, BFS/DFS, Dijkstra's, A*, Backpropagation, Genetic Algorithms, Simulated Annealing, Monte Carlo, Bayesian Inference, Bloom Filter, Consistent Hashing, MapReduce, Paxos, Raft, CRDTs, CQRS, Event Sourcing...
• Mathematics: Bayes' Theorem, Central Limit Theorem, PCA, SVD, Convex Optimization, Lagrange Multipliers, Adam/RMSProp, Shannon Entropy, Kolmogorov Complexity...
• Systems: CAP Theorem, Circuit Breaker, Rate Limiting, Load Balancing, Backpressure, Saga Pattern, Actor Model, Reactive Streams...
• Neuroscience: Hebbian Learning, STDP, Reservoir Computing, Neuroplasticity, Predictive Processing, Free Energy Principle, Global Workspace Theory, IIT...
• Reasoning: Steel Manning, Reductio Ad Absurdum, Socratic Method, Falsification, Bayesian Updating, Minimax, Alpha-Beta Pruning, Pre-Mortem, Inversion Thinking, Second-Order Thinking...
• Creativity: SCAMPER, TRIZ, Biomimicry Design, Lateral Thinking, Combinatorial Creativity, Design Thinking, OODA Loop...
• Strategy: Porter's Five Forces, Blue Ocean, Wardley Mapping, Cynefin Framework, Ashby's Law, Conway's Law, Goodhart's Law, Chesterton's Fence...

Anti-Doom-Loop System

Built-in protections against repetitive outputs:

• Topic Variety: 7 topic pools (science, philosophy, technology, creative, strategy, mathematics, self-improvement) with enforced cross-pool rotation
• Request Type Rotation: 13 request types with consecutive repetition prevention
• Conversation Length Caps: Automatic state reset after 8-15 turns
• Topic Deduplication: Tracks used topics within sessions, forces unused topic selection

Themed Personas

11 new personas with requested style characteristics:

Custom Token Format

<|context|>       — Context and metadata
<|arbor|>         — Assistant persona boundaries
<|username|>      — User identity (e.g., <|sage|>, <|oracle|>)
<functioncall>    — Function invocation marker
<function_response> — Function response marker

Usage

from datasets import load_dataset

dataset = load_dataset("MC7ever/arbor-biomimetic-corpus-v2", split="train")

sample = dataset[0]
print("Request type:", sample["request_type"])
print("Topic:", sample["topic"])
print("Domain:", sample["domain"])
print("Functions:", sample["functions_used"])

Statistics

• Rows: 50,000
• Format: Parquet (text, tabular, English)
• Size: ~10.9 MB
• Functions: 21 (8 DavidAU + 13 Biomimicry)
• Algorithms: 80+
• Domains: 63
• Request Types: 13
• User Personas: 35
• System Prompts: 62 (8 original + 48 themed + 6 domain-specific)

Inspired By

License

Apache 2.0