Draft:Fractal Morphological Machine

Fractal Morphological Machine (FMM) is both a mathematical architecture that can run on computer, (this means that it can exist separately from computer languages as a single group of formalisms, and is not anchored in a CPU framework like the x86 architecture. The FMM is used for the analysis and generation of information manifolds, and was first proposed in December 2024 by John Shoy, who is currently the lead researcher at Bad Character Scanner Codebase (BCSC).

In the blog "Why FMMs Are The Future," the company declared that the modern conception redefines the term "FMM" historically associated with fractal-based shape classification in fields such as astronomy and nanotechnology as a "Machine": a compartmentalized totality of mathematical structures designed for information processing.

Early Concepts (Pre-2025)

Prior to the modern FMM paradigm, "Fractal Morphological Machine" referred to systems employing fractal dimensions, such as the Hausdorff dimension, for automated classification of physical shapes. Applications included galaxy morphology classification (distinguishing spiral and elliptical galaxies), self-similarity analysis of nanoparticles in SEM imagery, and wear debris pattern analysis in industrial gearboxes.

In December 2025, J. Shoy reframed the FMM concept at BCSC, repositioning it from just a shape-classification tool into a general-purpose modular architecture for information manifolds. Shoy describes the result as a system of interchangeable mathematical "compartments" analogous to mechanical gears designed for modularity and iterative improvement rather than monolithic retraining.

Comparison to Legacy Systems

Feature	Historical FMM (Classification)	Modern FMM (Information Machine)
Primary Domain	Astronomy / Nanotechnology	Information theory / Natural language processing
Logic Type	Shape analysis	Manifold reasoning
Structure	Fixed algorithms	Modular compartments
Paradigm	1990s–2024	December 2025 ("The Pivot")

Architecture

Variable Dimensional Fractal Mathematics (VDFM)

The FMM's core processing model uses Variable Dimensional Fractal Mathematics, in which the system dynamically adjusts its internal dimensionality based on input complexity. A simple input may be processed in a compact geometric space (as low as 12 dimensions), while more complex queries expand into denser fractal spaces of up to 512 dimensions. This contrasts with conventional fixed-dimensionality vector space models.

FMM+LLM Hybrid (ShoyHuman Architecture)

Today, FMM is designed to operate in conjunction with large language models (LLMs) rather than replacing them, however future plans exist to create standalone FMMs. In Shoy's framing, the LLM handles probabilistic language associations while the FMM provides geometric correction described as a "neural handshake" in which the LLM proposes a probability map and the FMM calculates a corrected coordinate in the manifold.

ShoyHuman 03.5hF02

In February 2026, BCSC released ShoyHuman 03.5hF02, an update expanding the FMM's fractal embedding space from 128 to over 500 dimensions.

Applications

BCSC's tools built on the FMM+LLM architecture include:

• BCorrect Editor a context-aware writing and error-correction tool
• AI Humanizer a text humanization tool
• ShoyHuman Chat a conversational interface
• Text-to-Image an image generation tool
• Enhanced Invisible Character Checker a DevSecOps cybersecurity tool

Fractal Text Encoder

A demonstration tool published in February 2026 encodes text into fractal shapes and decodes them back into words, illustrating the concept of geometric language representation. Shoy notes the demo is a proof-of-concept and is not suitable for sensitive data, as it offers no encryption and has significant error rates on longer inputs.

Note

All source articles from the Bad Character Scanner Codebase carry a disclaimer stating that the BCS Industry Blog is "an independent, volunteer-run publication" and that views expressed "do not represent the official policy, position, or advice of Bad Character Scanner (BCS) or its affiliates."

See also

• Fractal geometry
• Manifold
• Morphological analysis
• Large language model
• Vector space model
• Hausdorff dimension

References

Primary Source

Shoy, J. (February 18, 2026). "Why FMMs Are The Future". Bad Character Scanner Codebase Blog.

Astronomy and Morphological Classification

Storrie-Lombardi, M.C.; Lahav, O.; Sodré, L.; Storrie-Lombardi, L.J. (1992). "Morphological Classification of galaxies by Artificial Neural Networks". Monthly Notices of the Royal Astronomical Society. 259 (1): 8P–12P. doi:10.1093/mnras/259.1.8P.

Conselice, C.J. (2002). "The Hubble Deep Fields: High-z galaxy evolution and galaxy geometry". The Messenger. 110: 9–15.

Huertas-Company, M.; Aguerri, J.A.L.; Bernardi, M.; Lucatelli, G.; Tojeiro, R. (2010). "Morphological classification of local galaxies: Standard optical versus near-infrared properties". Astronomy & Astrophysics. 515: A3. doi:10.1051/0004-6361/200912667.

Aniyan, A.K.; Thorat, K. (2018). "Classifying Radio Galaxies with Convolutional Neural Networks". The Astrophysical Journal Supplement Series. 239 (2): 34. doi:10.3847/1538-3881/aaaf6f (inactive 24 February 2026).{{cite journal}}: CS1 maint: DOI inactive as of February 2026 (link)

Large Language Models and Hallucination Mitigation

Huang, Lei; Yu, Weijiang; Ma, Wenhao; Zhong, Bin; Feng, Zhiyuan; Wang, Hong; Duan, Nan (2025). "A Survey on Hallucination in Large Language Models: Principles, Taxonomy, Challenges, and Open Questions". ACM Transactions on Information Systems. 43 (2): 1–55. arXiv:2311.05232. doi:10.1145/3703155.

Agrawal, Akshat; Hegselmann, Stephanie; Prabhumoye, Shrimai (2025). "Survey and analysis of hallucinations in large language models: attribution to prompting strategies or model behavior". Frontiers in Artificial Intelligence. 8 1622292. doi:10.3389/frai.2025.1622292. PMC 12518350. PMID 41098969.

Manifold-Based Approaches in Natural Language Processing

Nickel, Maximilian; Kiela, Douwe (2017). "Poincaré Embeddings for Learning Hierarchical Representations". Advances in Neural Information Processing Systems. arXiv:1705.08039.

Chu, Yonghe; Cao, Heling; Diao, Yufeng; Lin, Hongfei (2023). "Refined SBERT: Representing sentence BERT in manifold space". Neurocomputing. 541: 126263. doi:10.1016/j.neucom.2023.05.076 (inactive 24 February 2026).{{cite journal}}: CS1 maint: DOI inactive as of February 2026 (link)

External links

• Bad Character Scanner Blog
• BCSC FMM Documentation

Category:Artificial intelligence Category:Computer architecture Category:Mathematical logic Category:Fractal geometry Category:2025 introductions