The relevant NMCA disclosure is the 46-page blueprint dated April 20, 2025, titled:
The document identifies Derek Van Derven as inventor and states a conception date of April 20, 2025.
A public copy of the 46-page blueprint is available at:
April 20, 2025 NMCA Blueprint — PDF
The same document was subsequently preserved as a Zenodo record:
Zenodo — NMCA Original Blueprint (46 pages, April 20, 2025)
The Zenodo record preserves April 20, 2025 as the publication date of the original blueprint. The later technical metadata associated with the repository record does not change the underlying date of the disclosed document.
The important issue is not whether individual ingredients such as 3D memory, world models, planning, belief revision, or simulation existed independently. Several of those concepts clearly predate April 20, 2025.
The relevant question is whether the blueprint disclosed a particular integrated architecture in which persistent internal spatial representations are used as an active medium for cognition, simulation, planning, observation, contradiction handling, and updating.
The blueprint describes an internal 3D scene builder capable of constructing and manipulating internal representations of environments, objects, and abstract concepts.
Visualization is described not merely as an output but as part of the cognitive process: thoughts and concepts can be represented internally and simulated visually.
Visual scenes can be stored and linked to verbal or symbolic information for later recall and use.
The April 20, 2025 blueprint describes internal expectations tied to the scene, detection of conflict when new observation disagrees with that state, and revision of the internal representation so later planning uses the updated state.
Explicit belief graphs, certainty scores, and timestamps appear in later NMCA writing and are not treated here as part of the April 20 disclosure.
New observations can conflict with existing internal representations, triggering recognition of the contradiction and modification of the relevant state.
Internal spatial representations can be used to simulate routes, interactions, actions, and possible outcomes before or during physical execution.
External sensory information is compared against the internal representation, with discrepancies feeding back into the system.
The architecture includes reflective processes intended to monitor reasoning, identify contradictions, adjust internal reasoning, and iteratively improve future decisions.
The significance of the disclosure is therefore the relationship among these functions rather than any individual component in isolation.
The central architectural distinction is:
A persistent, re-enterable spatial scene is maintained internally and used as a medium of thought.
This is different from merely:
The architecture described by NMCA connects the persistent internal scene to cognition itself: the system can construct, revisit, simulate, compare, modify, and reason through an internal spatial representation.
The following table compares the principal architectural elements against representative publicly available work that predates the April 20, 2025 cutoff.
| Work / lineage | Date | Persistent spatial / 3D memory | Internal simulation / imagination | Belief / state updating | Embodied planning / action | Contradiction / revision | Meta-cognitive loop | Full NMCA combination? |
|---|---|---|---|---|---|---|---|---|
| Early internal-world / inner-world research | Pre-2025 | ✓ | ✓ | ◐ | ✓ | ◐ | ◐ | — |
| World Models / learned environment simulation | 2018+ | ◐ | ✓ | ◐ | ✓ | ◐ | — | — |
| LERF | 2023 | ✓ | — | — | ◐ | — | — | — |
| ConceptGraphs | 2023 | ✓ | — | ◐ | ✓ | — | — | — |
| SayPlan | 2023 | ✓ | ◐ | ◐ | ✓ | ✓ | — | — |
| RoboDreamer | Apr. 2024 | ◐ | ✓ | — | ✓ | — | — | — |
| Explore-EQA | 2024 | ✓ | ◐ | ◐ | ✓ | ◐ | — | — |
| OpenEQA | 2024 | ◐ | ◐ | ◐ | ✓ | — | ◐ | — |
| OV-PWM | July 2024 | ✓ | ✓ | ◐ | ✓ | ◐ | — | — |
| KARMA | Sept. 2024 | ✓ | ◐ | ✓ | ✓ | ◐ | — | — |
| Embodied-RAG | Sept. 2024 | ✓ | — | ✓ | ✓ | ◐ | — | — |
| EVA — Embodied World Model | Oct. 2024 | ◐ | ✓ | ◐ | ✓ | — | ◐ | — |
| Belief-revision research | 2024 | — | — | ✓ | — | ✓ | ◐ | — |
| Embodied VideoAgent | Dec. 2024 / Jan. 2025 | ✓ | ◐ | ✓ | ✓ | ◐ | ◐ | — |
| 3D-Mem | Nov. 2024 | ✓ | ◐ | ◐ | ✓ | ◐ | — | — |
| WorldMem | Apr. 16, 2025 | ✓ | ✓ | ◐ | ◐ | ◐ | — | — |
| NMCA — April 20, 2025 disclosure | Apr. 20, 2025 | ✓ | ✓ | ✓ | ✓ | ✓ | ✓ | ✓ INTEGRATED ARCHITECTURE DISCLOSED |
This table is not intended to claim that the individual technologies listed before NMCA were absent from the literature. They were not.
Persistent spatial memory, world models, belief revision, embodied planning, simulation, semantic scene representations, and active exploration all have substantial prior literature.
The question is whether those elements were publicly disclosed together in the particular functional relationship described by the April 20, 2025 NMCA architecture.
A collection of earlier references can demonstrate that individual components were known. It does not, by itself, establish that a single earlier reference disclosed the complete integrated architecture.
Several pre-April-20-2025 systems provide important pieces of the broader architecture.
These are significant predecessors and should be acknowledged rather than ignored.
The distinction under investigation is whether any one of them already disclosed the complete combination of:
persistent internal spatial scene + visual thought / internal simulation + memory + belief/state updating + embodied planning + contradiction handling + reflective/meta-cognitive feedback
On the sources examined for this investigation, no single pre-April-20-2025 document has been identified that discloses that complete combination.
The blueprint provides a concrete type of scenario that makes the architecture easier to distinguish from ordinary retrieval systems.
The system has previously observed a refrigerator and associated it with a location in its internal visual memory.
It receives the instruction:
“Go to the fridge and get a banana.”
The system loads the relevant spatial information, internally simulates the route and interaction, and executes the action.
New sensory information then reveals that the refrigerator is empty.
The discrepancy becomes a new reasoning event: the existing internal belief is no longer consistent with observation. The system can register the contradiction, revise its internal state, and generate a new subgoal.
The important point is not the refrigerator itself. It is the closed relationship between memory, spatial simulation, action, observation, contradiction, and belief update.
Several later publications and projects substantially expand techniques related to persistent 3D world modeling, belief inference, generative scene completion, and embodied reasoning.
A particularly relevant example is:
arXiv submission: May 12, 2026
The work describes maintaining and updating an agent's belief about an unobserved 3D environment from observations, including spatially consistent scene memory, multi-hypothesis belief sampling, sequential belief updating, prediction of unseen regions, and actionable 3D beliefs.
Project:
3D-Belief project page
Paper:
arXiv:2605.11367
The later work contains substantial technical material that is not claimed here to have appeared in identical form in the 2025 NMCA blueprint. Its importance to this record is that it demonstrates how closely related the broader architectural direction became in subsequent embodied-AI research.
Later technical implementations should therefore be evaluated on their own technical merits. They should not be retroactively treated as evidence that the April 20, 2025 disclosure contained every later implementation detail.
3D-Mem becomes publicly available on arXiv, providing an important pre-cutoff example of 3D scene memory for embodied exploration and reasoning.
Embodied VideoAgent introduces persistent scene memory derived from egocentric video and embodied sensors.
WorldMem is publicly dated shortly before the NMCA blueprint, making it an important chronological comparison point.
NMCA 46-page blueprint dated and publicly disclosed.
The document describes visual thought simulation, internal scene construction, visual memory, embodied simulation, contradiction handling, iterative learning, and meta-cognitive feedback as parts of a unified cognitive architecture.
3D-Belief is submitted to arXiv, describing explicit 3D belief maintenance, scene memory, unseen-region prediction, and embodied reasoning.
Document: NMCA Original Blueprint — 46 pages
Disclosure date: April 20, 2025
Author / inventor identified in document: Derek Van Derven
Zenodo DOI: 10.5281/ZENODO.21972901
Zenodo record: https://zenodo.org/records/21972901
Original PDF: April 20, 2025 NMCA Blueprint
The provenance record is intended to preserve the chronological existence of the disclosure independently of later development, terminology, implementation details, or subsequent research literature.
Based on the sources examined in this investigation:
No pre-April-20-2025 public document has been identified that discloses the complete integrated architectural combination described in the April 20, 2025 NMCA blueprint.
The investigation does identify numerous earlier works that disclose important individual components or neighboring architectural ideas.
Accordingly, the defensible chronological statement is:
The April 20, 2025 NMCA disclosure remains the earliest public disclosure identified in this investigation of the particular integrated combination described above.
The investigation separates three questions that are sometimes incorrectly collapsed into one:
These questions have different evidentiary significance. Finding an earlier example of one component does not answer the second question by itself.
This page records the state of the investigation and the chronological provenance of the April 20, 2025 NMCA disclosure.
Its purpose is preservation of the public technical record: what was disclosed, when it was disclosed, what earlier literature was identified, where substantial overlap exists, and where the investigated sources differ.
The record intentionally distinguishes documented facts from interpretation and does not depend upon a claim that every later implementation is technically identical to NMCA.
