Deep Reading Notes — Part 6: Are We Reinventing Wheels?

Date: 2026-03-30
Scope: Survey of 2024-2026 literature at the intersection of autonomous AI, neuroscience, biology, music theory, and markets
Trigger: Discovered that our tree-structured market model maps to Elliott Wave Theory (1930s). If we accidentally rediscovered a 90-year-old idea, what else have we been unknowingly re-treading?

0. The Question

Parts 1-5 built an architecture from first principles: geometric music theory, IDyOM predictive coding, fruit fly sparse coding, Kuramoto synchronization, behavior trees, Schenker/GTTM hierarchical analysis, and financial ecosystems. The growing tree — Kayle's core structure — emerged from cross-domain pattern-matching rather than from reading one field deeply.

Now we ask: has anyone else done this? Are there teams combining these same threads? Have we been building in a gap, or in a crowd?

Verdict after this survey: We are building in a gap. But the walls of that gap are closing in fast.

1. Neuroscience-Inspired AI Architecture (2024-2026)

1.1 The JHU "Architecture-First" Finding

The single most important paper for our project in this entire survey:

Bonner et al., "Convolutional architectures are cortex-aligned de novo," Nature Machine Intelligence, 2025.

Key finding: AI systems designed with biologically-inspired architecture can simulate human brain activity before being trained on any data. The JHU team built dozens of untrained neural networks and compared their responses to human/primate brain activity when exposed to the same images. Convolutional networks with expanded neuron populations rivaled billion-parameter trained models in cortical alignment.

Implications for us: This validates a core Kayle premise — that structure matters more than training data. We have been saying that the architecture of the tree (how the organisms are arranged, how signals flow) is more important than the weights. The JHU team just proved this for vision. The question is whether it extends to temporal/sequential domains like markets and music.

Status: We had the intuition. They have the proof. We are aligned but behind on evidence.

1.2 Neuromorphic Computing Wave

The post-transformer generation is actively pursuing brain-inspired design, but almost entirely in hardware:

- Neuromorphic chips (event-driven spiking networks) for edge IoT and autonomous vehicles
- 20-watt brain vs. megawatt training run — the energy argument is winning converts
- Hybrid neuro-symbolic architectures for intrusion detection (Sami et al., Scientific Reports, 2025)

Most of this work focuses on hardware substrate, not algorithmic architecture. Nobody in the neuromorphic community appears to be building multi-agent tree-structured reasoning systems. They are building better individual neurons, not better organizational patterns.

Status: Different problem. Our work is orthogonal — we are asking about the topology of intelligence, not the substrate.

1.3 Neuro-Symbolic AI

Active area but mostly focused on knowledge representation (ontologies, logic rules) bolted onto neural networks. Not what we are doing. We are not combining symbolic logic with neural nets — we are growing an organism-like structure that develops computational behavior through its developmental process.

Sources:
- Bonner et al., Nature Machine Intelligence, 2025 (JHU architecture study)
- Sami et al., "Neuro-symbolic dual-model architecture," Scientific Reports, 2025, doi:10.1038/s41598-025-27076-9

2. The Drosophila Connectome Revolution (2024-2026)

2.1 FlyWire: The Complete Map

The FlyWire connectome was published in Nature in October 2024: 139,255 proofread neurons, 100K+ cell annotations, 50M+ synapses with neurotransmitter information. This is the first complete wiring diagram of an adult brain for any animal.

What has been done with it:

a) Hierarchical community structure: Researchers discovered conserved circuit archetypes in the connectome's hierarchical community structure. The fly brain is not a random graph — it has nested modular organization at multiple scales. This is directly relevant to our tree architecture.

b) Behavioral circuit tracing: Teams have traced whole-brain circuits for specific behaviors, including a study on feeding suppression during escape — showing how the connectome enables rapid behavioral switching through distributed architecture.

2.2 The ESA Fly Connectome as Computational Reservoir

Costi, Leone, Hadjiivanov et al., "The Drosophila Connectome as a Computational Reservoir for Time-Series Prediction," Biomimetics, May 2025. (European Space Agency Advanced Concepts Team)

This is a bombshell for us. The ESA team took the actual FlyWire connectome topology, used it as the weight matrix for a reservoir computing (echo state) network, and tested it on chaotic time-series prediction.

Key findings:
- The connectome architecture is significantly more resilient to overfitting compared to standard random reservoirs
- Both topology AND synaptic weights contribute to this resilience
- When the full connectome is used, normalized error stays under 2% at lower regularization than any random network
- This contradicts the standard assumption that reservoir topology does not matter

This is someone doing almost exactly what we theorized. In Part 3, we discussed fruit fly sparse coding as an architectural principle. The ESA team actually implemented the connectome as a computation engine. They did not build organisms on top of it (that is our contribution), but they proved the substrate is computationally valuable.

Status: They validated the substrate. We are building the ecology on top. Complementary but we should cite this.

2.3 conn2res: Connectome Reservoir Toolbox

Suarez et al., "Connectome-based reservoir computing with the conn2res toolbox," Nature Communications, 2024, doi:10.1038/s41467-024-44900-4

An open-source Python toolbox for implementing biological neural networks as artificial neural networks. Modular design allowing arbitrary network architecture and dynamics. 41 citations already.

The existence of this toolbox means anyone can now take a connectome and use it as computation. The barrier to entry for bio-inspired architecture is collapsing.

2.4 Sparse Coding / FlyHash Still Active

The Drosophila olfactory circuit's sparse expansion coding continues to inspire AI work:
- BioHash (MIT-IBM Watson AI Lab): data-driven sparse high-dimensional hashing outperforming classical LSH
- DevFly: developmental methods for building locality-preserving sparse codes
- Fruit fly neural motif applied to word embeddings and NLP

Status: The sparse coding insight from Part 3 is well-known in the community. We are not ahead here — but nobody has connected it to market microstructure the way we have.

Sources:
- Dorkenwald et al., "The FlyWire connectome," Nature, 2024
- Costi et al., Biomimetics, 2025 (ESA reservoir computing)
- Suarez et al., Nature Communications, 2024, doi:10.1038/s41467-024-44900-4

3. Music Theory + AI + Neuroscience (2024-2026)

3.1 IDyOMpy: The Python Rewrite

IDyOMpy, published March 2025 in Journal of Neuroscience Methods.

The canonical IDyOM model (Information Dynamics of Music, Pearce 2005) has been reimplemented in Python (from Lisp). New features include:
- Probability estimation of silences (rests as information-bearing events)
- Enculturation modeling (how cultural exposure shapes expectation)
- Outperforms the original Lisp model on two metrics
- Validated by replicating prior EEG and behavioral results

For us: IDyOM was a foundational reference in Parts 1-2. The Python rewrite makes it far more accessible for integration into Kayle. The enculturation modeling is particularly interesting — it maps to how our organisms develop "style" from their training corpus.

3.2 Music-to-Market Mapping: Still a Gap

I searched extensively for formal work connecting music structure to market structure. Almost nobody is doing this.

The closest finding: fractal self-similarity is recognized in both domains independently. The Fractal Market Hypothesis (FMH) describes self-similar price patterns across time scales. Music theory has long recognized self-similar hierarchical structure (Schenker, GTTM). But I found zero papers that formally bridge the two with a shared mathematical framework.

This is a genuine gap. The connection we made in Parts 1-4 — that Schenkerian analysis of musical hierarchy maps to multi-scale market structure analysis — appears to be novel. Nobody has published on context-free grammars applied simultaneously to both music and markets.

Status: We are ahead. This is original.

3.3 AI Music Generation vs. Music Cognition

The bulk of 2024-2025 music+AI research focuses on generation (Transformer-based composition, emotional recognition via CNN+BiGRU, AI-assisted music education). Very little focuses on the cognitive machinery of musical understanding — the prediction/surprise dynamics we drew from in Parts 1-2.

The APA's March 2026 feature on music and the mind focuses on therapeutic applications, not computational models of musical cognition. The computational music cognition community remains small.

Sources:
- IDyOMpy: Journal of Neuroscience Methods, 2025, doi:10.1016/j.jneumeth.2024.110297
- Pearce, "IDyOM," 2005 (original)

4. Tree-Structured / Hierarchical AI (2024-2026)

4.1 Tree of Thoughts: The Mainstream Version

Yao et al., "Tree of Thoughts: Deliberate Problem Solving with Large Language Models," NeurIPS 2023, continued impact through 2025.

Tree of Thoughts (ToT) allows LLMs to explore multiple reasoning paths simultaneously, evaluate viability, and backtrack. Game of 24 accuracy went from 4% (chain-of-thought) to 74% (ToT). This has become a standard prompting technique.

2025-2026 extensions include feedback loops where models learn from past decisions and adjust in real-time.

For us: ToT is tree-structured reasoning within a single LLM. Our architecture is tree-structured reasoning across multiple organisms. The key difference is that ToT is a prompting technique for a monolithic model, while Kayle is a multi-agent developmental system. Related but architecturally distinct.

4.2 TreeRL: Tree-Based Reinforcement Learning for LLMs

TreeRL, ACL 2025: "LLM Reinforcement Learning with On-Policy Tree Search"

Entropy-guided tree search for RL training of LLMs, parameterized as (M, N, L, T)-trees (parallel trees, fork points, iterations, branching factor). This is RL training using tree-structured exploration, not tree-structured inference.

4.3 Hierarchical Reinforcement Learning Revival

Active research surge in 2025:

- Causal discovery for subgoal structure: Modeling subgoal hierarchy as a causal graph, learning it with causal discovery algorithms. Remarkable improvements for tree structures and Erdos-Renyi random graphs (arxiv 2507.04373, July 2025)
- Temporal structure discovery: HRL methods ranging from online experience to LLM-guided decomposition (survey: arxiv 2506.14045, June 2025)
- Macro actions: Abstracting micro-action sequences into macro-actions to simplify decision-making (Complex & Intelligent Systems, 2025)
- Human causal perception-driven HRL: Using human-like causal reasoning to structure hierarchies (Neural Networks, 2025)

For us: The HRL community is converging on exactly the kind of hierarchical temporal decomposition we built into Kayle's behavior trees. The causal discovery approach is particularly interesting — they are trying to learn the tree structure from data, which is what our developmental process does organically.

Status: Converging. They are formalizing what we built intuitively. We should read the causal discovery work carefully.

4.4 Context-Free Grammars + Neural Networks

OpenAlex returned generic deep learning surveys for this query. The intersection of CFGs with neural network learning remains niche. Nobody appears to be using CFGs to model either market structure or music structure in a neural network context, which was our core Part 4 insight.

Sources:
- Yao et al., NeurIPS 2023 (Tree of Thoughts)
- TreeRL, ACL 2025
- arxiv 2507.04373 (causal HRL, July 2025)
- arxiv 2506.14045 (HRL survey, June 2025)

5. Autonomous AI Agents in Markets (2024-2026)

5.1 TradingAgents: The Multi-Agent LLM Trading Firm

TradingAgents (Tauric Research), ICML 2025; v0.2.2 released March 2026.

Architecture: LLM-powered agents in specialized roles:
- Fundamental analysts, sentiment analysts, technical analysts
- Bull and Bear researcher agents debating market conditions
- Risk management team monitoring exposure
- Traders synthesizing all inputs

Open-source (GitHub), supports GPT-5.4, Gemini 3.1, Claude 4.6. Performance improvements in cumulative returns, Sharpe ratio, and maximum drawdown over baselines.

For us: This is the closest thing to what we tried to build with Kayle's trading organisms. Key differences:
- TradingAgents uses LLMs as the agent substrate; we use specialized lightweight organisms
- TradingAgents has fixed roles; our organisms can differentiate and specialize during development
- TradingAgents has no developmental process; our tree grows
- TradingAgents has no biological inspiration; it is inspired by human trading firms

The LLM-as-agent approach is fundamentally different from our biology-inspired approach. They are simulating a trading desk. We are growing an ecosystem. But they have production code and benchmarks, and we should study their coordination protocols.

5.2 AI-Trader Benchmark

Fang et al., "AI-Trader: Benchmarking Autonomous Agents in Real-Time Financial Markets," arxiv 2512.10971, December 2025.

First fully-automated, live, data-uncontaminated evaluation benchmark for LLM trading agents. Key findings:
- General intelligence does NOT automatically translate to effective trading
- Risk control capability determines cross-market robustness
- AI strategies achieve excess returns more readily in highly liquid markets
- Most agents exhibit poor returns and weak risk management

For us: This validates our skepticism about pure LLM approaches to trading. Their finding that risk control is the critical differentiator aligns with BLAKCAT's risk-first architecture. The "general intelligence does not equal trading skill" finding supports our specialization thesis — you need organisms evolved for specific market niches.

Status: We are philosophically aligned with the benchmark findings. Our architecture is designed to solve exactly the problems they identify. But we need to benchmark against AI-Trader.

Sources:
- TradingAgents, ICML 2025, arxiv 2412.20138
- AI-Trader benchmark, arxiv 2512.10971

6. Biological Computation and Unconventional Computing (2024-2026)

6.1 Organoid Intelligence: Brains in a Dish

The organoid intelligence field has exploded:

- Johns Hopkins (2025): Lab-grown brain organoids show synaptic plasticity — the molecular machinery for learning and memory. After chemical stimulation, increased expression of immediate early genes associated with memory formation.
- PhyChip project: Building hybrid bio-electronic chips using living neural organoids
- arxiv 2503.19770: Comprehensive overview of brain organoid computing (March 2025)
- Ethical frameworks (2025): Active debate about consciousness, privacy, and moral status of organoids

Key challenge: writing, reading, and interpreting information to and from organoids remains unsolved. The complexity is enormous.

For us: Organoid intelligence is computation using actual biological substrate. We are using biological principles in silicon. Different approach, same inspiration. Their work validates the idea that biological organization has inherent computational value.

6.2 Slime Mold Computing: Still Going

Physarum polycephalum continues to deliver:

- Solves Traveling Salesman Problem in linear time (Lanzhou University)
- PhyChip EU project aims for functional biomorphic computing devices
- Physarum-inspired adaptive optimization for microtubular network design (Science China Chemistry, 2024)
- Non-equilibrium steady state computations tracked and distinguished (arxiv 2504.03492, 2025)

For us: Physarum's tube network optimization maps directly to how our tree prunes and grows connections. The slime mold does not compute through neurons — it computes through network topology itself. This is what we mean when we say the tree IS the computation.

6.3 Reservoir Computing Renaissance

Yan et al., "Emerging opportunities and challenges for the future of reservoir computing," Nature Communications, 2024 (251 citations).

Reservoir computing is experiencing a revival. The core idea — that a fixed, high-dimensional dynamical system can be used for computation with only the readout weights being trained — is gaining traction as an energy-efficient alternative to deep learning.

The ESA fly connectome work (Section 2.2) is part of this wave. Other developments include memristor-based physical reservoirs and brain-inspired adaptive E/I balance control (Nature Communications, 2025).

Sources:
- Johns Hopkins organoid study, 2025
- PhyChip (phychip.eu)
- Yan et al., Nature Communications, 2024, doi:10.1038/s41467-024-45187-1
- arxiv 2503.19770 (brain organoid computing overview)

7. Elliott Wave Theory + Formal Mathematics + AI (2024-2026)

7.1 ElliottAgents: LLMs Meet Elliott Waves

Vorobiev et al., "Large Language Models and the Elliott Wave Principle," Applied Sciences (MDPI), 14(24), 2024, doi:10.3390/app142411897.

ElliottAgents (2025), arxiv 2507.03435 and arxiv 2506.16813.

A multi-agent system combining:
- LLMs for natural language understanding and analysis
- RAG (Retrieval-Augmented Generation) for real-time market data
- Deep Reinforcement Learning for backtesting and pattern recognition
- LangGraph framework for inter-agent dialogue

Agents engage in natural language dialogue to collaboratively refine Elliott Wave analysis. Tested on stock and cryptocurrency markets with improvements in pattern recognition and trend forecasting.

For us: Someone built exactly the kind of system that combines Elliott Wave pattern detection with multi-agent AI. However:
- They use LLMs as the analytical engine (heavyweight, general-purpose)
- They do not connect Elliott waves to context-free grammars or musical structure
- They have no biological/developmental component
- Their agents do not grow or evolve — they are fixed roles

The gap we fill: the formal mathematical connection between Elliott wave structure, context-free grammars, Schenkerian analysis, and behavior trees. They automated the pattern detection. We are building the generative grammar.

7.2 Fractal Market Structure: Active but Not Connected

The Fractal Market Hypothesis continues to generate research:
- Laplace-Mittag-Leffler distributions for fractal financial analysis (ScienceDirect, 2025)
- 34-year empirical study of Nifty 50 self-similarity (SSRN, 2025)
- Bank of England stability analysis using FMH framework

Everyone agrees markets are fractal/self-similar. Nobody has connected this to tree grammars or musical hierarchy in a formal system.

7.3 Has Anyone Formalized Elliott Waves Mathematically?

OpenAlex returned no highly-cited papers on mathematical formalization of Elliott waves since 2020. The theory remains largely informal — rules of thumb about wave counts, Fibonacci ratios, and nesting patterns. Our observation that Elliott waves are a context-free grammar (with specific production rules) appears to be genuinely novel.

Status: The formal grammar connection is ours. ElliottAgents is the closest competition but operates at a completely different level of abstraction.

Sources:
- Vorobiev et al., Applied Sciences, 2024
- ElliottAgents, arxiv 2506.16813, 2507.03435

8. Free Energy Principle + AI Systems (2024-2026)

8.1 Active Inference Implementations

The gap between Friston's theory and practical AI remains wide, but it is closing:

- pymdp: Open-source Python library for active inference in discrete state spaces. Under active development. Agents maximize "epistemic value" (curiosity) while simultaneously learning environment structure and maximizing reward.
- ActiveInference.jl (January 2025): Julia reimplementation of pymdp, integrated with Julia's cognitive modeling ecosystem for fitting POMDP active inference models to empirical behavioral data.

Key capability: Active inference agents naturally balance exploration and exploitation through expected free energy minimization. This is exactly the uncertainty-reduction mechanism we discussed in Parts 1-3.

8.2 Has Anyone BUILT a Real Active Inference System?

The honest answer: mostly toy problems and proof-of-concept demos. No production active inference trading system exists. No large-scale active inference agent has been deployed in the wild. The theory-to-implementation gap remains large.

Most implementations handle small state spaces. The combinatorial explosion of real-world environments (like financial markets) makes pure active inference intractable without approximations.

For us: Active inference remains the theoretical gold standard for what Kayle's organisms should be doing internally. But nobody has solved the scaling problem. Our approach — specialized organisms with small generative models, coordinated through tree structure — may be a viable decomposition strategy for making active inference tractable at scale.

Status: Theoretical alignment, implementation opportunity. Nobody has solved scaling. Our decomposition approach may be the answer.

Sources:
- pymdp: github.com/infer-actively/pymdp, arxiv 2201.03904
- ActiveInference.jl: Entropy, 27(1), 2025, doi:10.3390/e27010062

9. Emergent Intelligence and Artificial Life (2024-2026)

9.1 ASAL: Automating the Search for Artificial Life

Sakana AI, MIT, OpenAI, IDSIA. Published December 2024.

ASAL uses vision-language foundation models to automatically discover artificial life simulations. Three search modes:
- Supervised target: Find simulations matching text descriptions
- Open-endedness: Find simulations that generate persistent novelty
- Illumination: Discover maximally diverse simulation sets

Discoveries include exotic Boids flocking (Fibonacci spirals), self-organizing Lenia cells, dynamic Particle Life ecosystems with cell division, and open-ended cellular automata exceeding Conway's Game of Life.

For us: ASAL automates the discovery of emergent behavior in simple substrates. We are doing something related but different — we are not searching for interesting rules in cellular automata, we are growing a computational organism in a market environment. But the principle of using AI to discover emergence is the same.

The open-endedness search mode is particularly relevant to Kayle's developmental process. We want our tree to generate persistent novelty — new trading behaviors that remain informative over time.

9.2 Emergent Behaviors in LLM Agent Networks

ResearchGate, 2025: "Emergent Behaviors in LLM-Driven Autonomous Agent Networks"

Collections of LLM-powered agents display complex behaviors that are not explicitly programmed:
- Coordination and specialization
- Collective memory formation
- Spontaneous strategy development

This is emergence in multi-agent LLM systems. The agents are heavyweight (full LLMs), but the coordination patterns resemble what we hope to achieve with lightweight organisms.

9.3 Michael Levin: Collective Intelligence Across Scales

McMillen & Levin, "Collective intelligence: A unifying concept for integrating biology across scales and substrates," Communications Biology, 2024 (84 citations).

Levin argues that the same principles of collective intelligence operate across:
- Cells organizing into tissues (morphogenesis)
- Neurons organizing into brains (cognition)
- Organisms organizing into societies (swarm intelligence)

The unifying mechanism: bioelectric networks that bind computational units into collective agents. The "cognitive glue" is the same regardless of substrate.

Levin (2025): "The Multiscale Wisdom of the Body: Collective Intelligence as a Tractable Interface for Next-Generation Biomedicine," BioEssays.

Ertle, Levin & Scheutz (2025): "Low-Cost Intelligence Through Pattern-Guided Exploration," IEEE ICDL 2025.

For us: Levin's framework is the most complete theoretical justification for what we are building. When we say Kayle "grows" intelligence through organism coordination, we are describing exactly what Levin means by collective intelligence across scales. His 2025 paper on "low-cost intelligence through pattern-guided exploration" directly maps to how our organisms should explore market structure.

Status: Levin is the theoretical godfather of our approach. We should be reading everything he publishes. But he has not applied his framework to markets or music.

Sources:
- ASAL: arxiv 2412.17799 (Sakana AI)
- McMillen & Levin, Communications Biology, 2024, doi:10.1038/s42003-024-06037-4
- Levin, BioEssays, 2025, doi:10.1002/bies.202400196
- Ertle, Levin & Scheutz, IEEE ICDL 2025

10. Kuramoto Synchronization in AI (2024-2026)

10.1 AKOrN: The Kuramoto Neuron

Miyato et al., "Artificial Kuramoto Oscillatory Neurons," ICLR 2025 (Oral). Autonomous Vision group.

This is a landmark paper. AKOrN replaces traditional threshold neurons with coupled Kuramoto oscillators. The update rule:

Each neuron is an oscillator on an (N-1)-dimensional sphere. Coupling is through the Kuramoto differential equation, with data-dependent conditional stimuli providing symmetry-breaking input. Networks alternate between Kuramoto layers (oscillatory update) and readout modules (feature extraction).

Key results:
- Object discovery (PascalVOC): Competitive with slot-attention models — oscillators spontaneously bind object features through synchronization
- Sudoku reasoning (out-of-distribution): 89.5% accuracy, outperforming transformer-based reasoning
- Adversarial robustness (CIFAR-10): 51-59% adversarial accuracy with surprisingly good calibration (1.3-1.4% ECE) WITHOUT explicit robustness training
- Energy-based voting: The energy landscape of Kuramoto dynamics correlates with prediction confidence, enabling test-time compute scaling that improves reasoning by up to 33%

The deep insight: early layers capture high-frequency local features, deeper layers generate global low-frequency waves. This is exactly the multi-scale frequency decomposition we discussed in Part 2 when analyzing music structure.

For us: AKOrN validates our core Part 2 thesis — that Kuramoto synchronization is a viable computational primitive for AI. We discussed oscillator synchronization as a coordination mechanism for our organisms. The AKOrN team proved it works as a neuron replacement. The question is whether it can work at the inter-organism level (our use case) as well as the intra-neuron level (their use case).

The adversarial robustness finding is especially relevant for markets, where adversarial conditions are the norm. A system that is naturally calibrated under adversarial perturbation is exactly what BLAKCAT needs.

10.2 Swarmalators: Synchronization Meets Swarming

Anwar et al., "Collective dynamics of swarmalators with higher-order interactions," Communications Physics, 2024 (56 citations).

Tunable colloidal swarmalators, Nature Communications, 2025.

Swarmalators are systems where each agent both oscillates (has phase dynamics) and moves (has spatial dynamics), with the two coupled. The field achieved its first fully tunable experimental realization in 2025 — tiny self-propelled particles that simultaneously coordinate motion and synchronize internal rhythms through hydrodynamic coupling.

Four collective states: async, phase wave, mixed, and sync. Applications include autonomous robotic swarms with spontaneous coordination.

For us: Swarmalators formalize what we are trying to build. Our trading organisms have both spatial position (in market structure space) and oscillatory phase (in time/cycle space). The swarmalator framework could provide the mathematical underpinning for inter-organism coordination in Kayle.

Status: This is a direct mathematical tool for us. We should be using swarmalator dynamics as the coordination model.

Sources:
- Miyato et al., ICLR 2025 (AKOrN)
- Anwar et al., Communications Physics, 2024, doi:10.1038/s42005-024-01556-2
- Colloidal swarmalators, Nature Communications, 2025, doi:10.1038/s41467-025-66830-5

11. Synthesis: The Competitive Landscape

11.1 What Others Have Done That We Should Know About

Connectome as computation: ESA/ACT (Costi 2025) — Fly connectome as reservoir for time-series prediction. Validates our substrate.

Kuramoto in AI: Miyato/AKOrN (ICLR 2025) — Kuramoto oscillators as neurons. Validates our coordination mechanism.

Multi-agent trading: TradingAgents (ICML 2025) — LLM agents as trading firm. Similar goal, different architecture.

Elliott + AI: ElliottAgents (2024-2025) — LLM agents detecting Elliott patterns. Automated detection, no formal grammar.

Collective intelligence: Levin (2024-2025) — Multi-scale collective computation. Theoretical framework for our approach.

Swarmalators: Anwar (2024), Konstanz (2025) — Coupled oscillator-swarming dynamics. Mathematical model for our organisms.

Active inference: pymdp, ActiveInference.jl — Implementation of free energy minimization. Internal model for organisms.

IDyOM Python rewrite: IDyOMpy (2025) — Accessible music prediction model. Tool for Ballad integration.

Artificial life search: ASAL/Sakana (2024) — Automated discovery of emergence. Methodology for growing intelligence.

Architecture > training: JHU/Bonner (2025) — Brain-like AI without training. Validates architecture-first design.

11.2 What Nobody Has Done (Our Genuine Contributions)

1. Formal grammar connecting Elliott waves, Schenkerian analysis, and behavior trees. Nobody has published a context-free grammar that generates both musical hierarchical structure and market fractal structure. This is original.

2. Multi-scale biological architecture for markets. TradingAgents uses LLMs. ElliottAgents uses LLMs. Nobody is growing a tree-structured ecology of specialized lightweight organisms for market computation. This is original.

3. Music-to-market mapping through shared hierarchical formalism. The fractal self-similarity of both domains is well-known independently. The formal bridge through GTTM/Schenker production rules has not been published. This is original.

4. Developmental AI for trading. Nobody is building trading systems that grow through a developmental process analogous to biological ontogeny. All existing systems are trained (static architecture, learned weights) or prompted (LLMs with instructions). A system that develops, differentiates, and matures is unprecedented in the trading literature.

5. Swarmalator-coordinated active inference organisms on a behavior tree. The combination of swarmalator dynamics (inter-organism coordination), active inference (intra-organism computation), behavior trees (organizational hierarchy), and developmental growth (ontogeny) has no precedent in the literature.

11.3 Where We Are Behind

1. Benchmarking. TradingAgents has ICML publication, open-source code, multi-LLM support. AI-Trader has live evaluation. We have none of this yet.

2. Implementation maturity. The ESA fly connectome reservoir is published and replicated. Our use of connectome-inspired architecture is still architectural.

3. Formal proofs. The JHU team proved architecture matters with controlled experiments. We have the intuition but not the controlled validation.

4. Community. AKOrN has an ICLR oral, open-source code, a project page. Active inference has pymdp with an active developer community. Our work is internal.

12. Immediate Action Items for the Project

12.1 Papers to Read in Full (Highest Priority)

1. Miyato et al., AKOrN (ICLR 2025) — full technical details of Kuramoto as neurons
2. Costi et al., Drosophila connectome as reservoir (Biomimetics, 2025) — methodology for connectome-as-computation
3. McMillen & Levin, Collective Intelligence (Communications Biology, 2024) — theoretical framework
4. TradingAgents (ICML 2025) — competitive intelligence on multi-agent trading
5. Bonner et al., convolutional architectures cortex-aligned de novo (Nature Machine Intelligence, 2025) — architecture > training

12.2 Tools to Integrate

1. IDyOMpy — Python-accessible music prediction for Ballad
2. conn2res — connectome-based reservoir computing toolbox
3. pymdp — active inference implementation
4. Swarmalator math — coordination model from Anwar et al.

12.3 Connections to Formalize

1. Elliott wave rules as context-free grammar production rules (write this up)
2. Swarmalator dynamics as the mathematical model for inter-organism coordination in Kayle
3. Active inference as the internal computation model for individual organisms
4. Behavior tree as the grammar's parse tree

13. The Verdict: Are We Reinventing Wheels?

No. But we are independently discovering components that others have proven in isolation.

The situation is this: different research groups have validated individual pieces of our architecture:
- Kuramoto works as computation (AKOrN)
- Connectome topology has computational value (ESA)
- Architecture matters more than training (JHU)
- Active inference provides the right internal model (pymdp community)
- Collective intelligence operates across scales (Levin)
- Multi-agent trading improves over single-agent (TradingAgents)
- Elliott wave detection can be automated (ElliottAgents)
- Swarmalator dynamics describe coupled oscillator-agent systems (Anwar et al.)

Nobody has assembled all of these into a single coherent architecture. The synthesis is ours. The growing tree — Kayle — is novel not because its components are novel, but because the integration is novel.

The risk is that someone else will see the same convergence. The walls are closing in. AKOrN + swarmalators + active inference + connectome reservoirs = someone else's version of Kayle, potentially. We need to publish or build faster.

The opportunity is that validation of each component reduces our technical risk. Each paper in this survey proves that one piece of our architecture works. What we need to prove is that they work together, and that the developmental growth process adds value over static assembly.

Final assessment: We are building something genuinely novel at the integration layer. We are not reinventing wheels — we are building a vehicle nobody else has designed. But the wheels are now off-the-shelf, and other mechanics are in the shop.

14. Source Count

New sources identified in this survey: 42

Major papers (read or reviewed): 22
Web resources / project pages: 12
Conference proceedings / workshops: 8

Cumulative source count (Parts 1-6): 218 + 42 = 260

15. What Fundamentally Changes

Nothing in this survey contradicts our vision. Several things validate it strongly. The key shifts:

1. Architecture-first is proven. JHU showed untrained CNNs match billion-parameter trained models. This is the strongest evidence yet that our design-over-training philosophy is correct.

2. Kuramoto as computation is proven. AKOrN moved Kuramoto from a coordination metaphor to a working computational primitive. This upgrades our confidence in using oscillator dynamics for inter-organism binding.

3. The connectome IS a useful computer. ESA proved the fly brain topology outperforms random networks as a computational substrate. This validates our Part 3 intuition at a level we had not achieved.

4. Swarmalators give us the math. We needed a mathematical framework for coupled spatial-oscillatory dynamics. Swarmalator theory provides exactly this.

5. The music-market bridge is ours to claim. After extensive search, nobody has formalized the shared hierarchical grammar connecting musical structure and market structure. This is genuinely novel territory.

6. We need to move faster. The component validation is accelerating. If we wait too long, someone else will see the integration opportunity. The time to build and publish is now.