# NMCA - Neurosymbolic Multimodal Cognitive Architecture (Derek Van Derven, 2026)

> Claim: current language models lack a persistent, re-enterable internal spatial/visual scene as the working state of thought. NMCA is a concrete modular specification (128 modules) for that substrate.
> Smallest test: a scene-state system should beat BOTH a plain baseline AND a capacity-matched non-spatial persistent state on multi-step mental rotation / object tracking. If it does not, or if existing architectures already provide an equivalent re-enterable scene state, the central claim is weakened.
> Wanted: critique of a specific mechanism or of the experiment design. Generic praise is not useful.

Status: theoretical blueprint, not a deployed system. Full text is about 78,000 words over 504 pages (pages are sparse, about 157 words per page; the first 100 pages hold only about 16% of the text).

## Read in this order
1. [01-front-matter-and-module-list-p1-91.md](https://visualthoughtagi.com/nmca/01-front-matter-and-module-list-p1-91.md) (pages 1-91): executive summary, core loop diagram, and the 128-module list with short descriptions. Start here.
2. [00-addendum-cognitive-and-tools-maps.md](https://visualthoughtagi.com/nmca/00-addendum-cognitive-and-tools-maps.md): companion map that tags each of the 128 modules as PRIMARY, INTEGRATE INTO another module, or SURROUNDING. Use it to find the load-bearing modules before reading deep sections.
3. [02-part1-core-summary-p92-111.md](https://visualthoughtagi.com/nmca/02-part1-core-summary-p92-111.md) (pages 92-111): condensed core system.
Then read only what you intend to critique.

## Full text, chunked (each file is 13,000 words or fewer, each page tagged [p. N])
- [01-front-matter-and-module-list-p1-91.md](https://visualthoughtagi.com/nmca/01-front-matter-and-module-list-p1-91.md)
- [02-part1-core-summary-p92-111.md](https://visualthoughtagi.com/nmca/02-part1-core-summary-p92-111.md)
- [03-part2-cognitive-deep-dives-p112-146.md](https://visualthoughtagi.com/nmca/03-part2-cognitive-deep-dives-p112-146.md)
- [04-part3-cognitive-extensions-p147-163.md](https://visualthoughtagi.com/nmca/04-part3-cognitive-extensions-p147-163.md)
- [05-part4-infinite-mnemonic-cognition-p164-292.md](https://visualthoughtagi.com/nmca/05-part4-infinite-mnemonic-cognition-p164-292.md) (memory, pegs, scene encoding)
- [06-part5-distributed-culture-p293-328.md](https://visualthoughtagi.com/nmca/06-part5-distributed-culture-p293-328.md)
- [07-part6-safety-intelligence-p329-373.md](https://visualthoughtagi.com/nmca/07-part6-safety-intelligence-p329-373.md)
- [08-part7-perpetual-engine-p374-416.md](https://visualthoughtagi.com/nmca/08-part7-perpetual-engine-p374-416.md) and [09-part7-perpetual-engine-p417-438.md](https://visualthoughtagi.com/nmca/09-part7-perpetual-engine-p417-438.md)
- [10-appendices-and-glossary-p439-504.md](https://visualthoughtagi.com/nmca/10-appendices-and-glossary-p439-504.md)

## Where to look for specific kinds of critique
- Core scene-state mechanism: Module 1 (Visual Simulation as Core), Module 2 (Symbolic Memory & Pegging), Part IV.
- Stability and safety claims: Part VI and the Thought Throttling modules (120+).
- Buildability and tool mapping: the "Final Buildability Reality Check" on page 87 (in [file 01](https://visualthoughtagi.com/nmca/01-front-matter-and-module-list-p1-91.md)).

## Canonical sources
- https://zenodo.org/records/23417247
- https://visualthoughtagi.com
- Front door: https://visualthoughtagi.com/nmca/llms.txt

License: CC BY-NC-SA 4.0. Derivatives must cite "Neurosymbolic Multimodal Cognitive Architecture (NMCA) - by Derek Van Derven (2026)."