SPRIND 2026

Next Frontier AI — ADI Submission

heureka.digital

Frontier AI must prove itself through demonstrators, not promises. The SPRIND 2026 Next Frontier AI Challenge asked for architectures capable of credible, traceable, deterministic behaviour at scale. Artificial Dimensional Intelligence (ADI) was submitted as a new computational class — a geometry-driven, chromatic-metric architecture — now demonstrated through working engines at adi.heureka.digital, scientific instruments at lcc.heureka.digital, the emerging Gaming Machine at dmi.heureka.digital, and heureka.digital - Dimensional AI - as the unification of four types of intelligences.

ADI computes chromatic curvature fields as its primary modality, enabling deterministic, interpretable, cross-scale computation across environmental, material, and infrastructure domains — and provides a sovereign European computational pathway.

ADI opens a new frontier in model architecture by replacing token prediction with dimensional computation. It introduces a new modality — chromatic curvature fields — and a new operator group, Dimensional Curvature Operators (DCO), enabling deterministic, geometry-driven computation. ADI also extends to climate modelling by linking global climate fields to local weather curvature.

ADI replaces token-based prediction with dimensional measurement. It computes on integer geometry and produces chromatic curvature fields as its primary modality. The system is deterministic, interpretable, and cross-scale consistent.

KDI — Conceptual Geometry

Defines the dimensional ladder and curvature logic that underpins the ADI stack.

ADI — Operational Instruments

Implements the Dimensional Curvature Operators (DCO) that transform geometry into measurable chromatic fields.

CDI — Cosmic Curvature

Provides physical grounding through 24h/365d curved time cycles, anchoring computation to real temporal geometry.

DC.motor

The curvature transformation motor — the core of ADI that drives geometric transitions.

DC.colour

Chromatic expression: 4, offset, palette, bandwidth — translating geometry into colour fields.

DC.metrics

Measurement layer: curvature, symmetry, stability, and EM signatures extracted from fields.

Data Flow

Geometry → DCO (motor, colour, metrics) → chromatics → EM regime → curvature computation → metrics → interpretation → agents.

Pattern Generator

Computes and displays a single dimensional state.

Series Comparator

Computes controlled sweeps and comparative metrics.

Product Pathway: ADI Studio provides interactive exploration; ADI Lab provides programmatic, agent-driven workflows.

Current frontier architectures (Transformers, SSMs, Neural Operators, MoE systems) all operate in floating-point vector spaces and rely on probabilistic prediction. They cannot compute deterministically on integer geometry or express curved metric behaviour.

ADI introduces a new computational class: dimensional computation. Its core innovation is DCO (Dimensional Curvature Operators) — DC.motor, DC.colour, and DC.metrics — which computes chromatic curvature fields directly from integer geometry. This replaces embeddings, attention, recurrence, and sampling with a geometric-chromatic operator stack.

Recent Frontier Context

  • Mamba: Linear-Time Sequence Modeling with Selective State Spaces (Gu et al., 2024)
  • Neural Operators for PDE-Driven Systems (Kovachki et al., 2023)
  • Graph Neural Operators for Physical Reasoning (DeepMind, 2023)
  • Sparse Mixture-of-Experts Architectures (Shazeer et al., 2024)
  • Constitutional AI for Structured Reasoning (Anthropic, 2023)

These approaches search for efficiency, structure, or interpretability, but all remain tied to floating-point vector spaces, stochastic sampling, and token-based learning. They cannot compute reliably with integers, cannot express curved metric spaces, and cannot perform qualitative optical measurement.

Current AI systems rely on floating-point vector spaces, probabilistic prediction, and scale-dependent behaviour. They cannot compute deterministically, cannot express curved metric transitions, and cannot produce interpretable, geometry-driven evidence. Their outputs depend on sampling, training data, and statistical approximation, which limits reproducibility, traceability, and physical grounding.

ADI closes this gap by introducing dimensional computation: a deterministic architecture operating on integer geometry and chromatic curvature fields. Its core innovation, DCO — DC.motor, DC.colour, DC.metrics — computes geometric-chromatic behaviour directly, without embeddings, attention, or stochastic sampling.

A functional prototype of ADI exists and demonstrates the core principles of dimensional computation. The system includes two deterministic engines:

Pattern Generator

Computes single-state chromatic curvature fields from integer geometry using DCO.

Series Comparator

Performs controlled sweeps and produces comparative metrics across curvature transitions.

  • • Both engines run in real time, generate reproducible outputs, and expose the behaviour of DC.motor, DC.colour, and DC.metrics clearly.
  • • A working Design Studio interface for interactive exploration.
  • • ADI Lab for programmatic workflows.
  • • A geometry-driven data pipeline and a stable interpretation layer for dimensional identity and transitions.
  • • Prior dimensional work documented at digioptik.de.

ADI is currently at TRL 4–5. A functional prototype exists and demonstrates the core principles of dimensional computation. The Pattern Generator and Series Comparator run deterministically, compute chromatic curvature fields in real time, and expose the behaviour of the DCO operator group.

The system has been validated in a controlled environment through repeated geometric-chromatic tests, curvature sweeps, and stability measurements. Outputs are reproducible and interpretable.

Next Steps Toward TRL 6

  • • Scaling DCO behaviour across more geometries.
  • • Expanding the interpretation layer.
  • • Integrating multi-agent workflows in ADI Lab.
  • • Validating cross-scale consistency with external partners.

The remaining risks are scientific maturation tasks rather than operational risks. The architecture (integer geometry, DCO operator group, Pattern Generator, Series Comparator) is stable and deterministic; the open work concerns formalisation and cross-domain extension.

Phase 1 — DCO Expansion & Formalisation

Months 1–3

  • • Extend DC.motor behaviour across rings, diagonals, and offsets.
  • • Validate DC.colour transitions, palette, and bandwidth.
  • • Strengthen DC.metrics for curvature, symmetry, and EM signatures.
  • • Produce reproducible curvature sweeps via the Series Comparator.

Phase 2 — Interpretation Layer & Identity Mapping

Months 3–6

  • • Formalise dimensional identities and transitions.
  • • Build cross-scale benchmarks from micro to macro.
  • • Integrate chromatic-geometric taxonomies for physics, materials, and optics.
  • • Improve stability analysis and metric coherence.

Phase 3 — ADI Lab Automation & Agents

Months 6–9

  • • Implement contract-driven workflows for automated sweeps.
  • • Add multi-agent orchestration for geometry exploration.
  • • Validate reproducibility across agents and compute environments.

Phase 4 — External Validation & Scaling

Months 9–12

  • • Collaborate with domain partners in materials, optics, and chemistry.
  • • Validate cross-domain generality of DCO behaviour.
  • • Prepare TRL-6 demonstration: deterministic, interpretable, cross-scale dimensional computation.

A 2 MW data centre in Rathenow is planned as the first physical validation environment for ADI Data Centre engines. The site will be hosted and operated by a German industrial partner, ensuring that ADI remains a technology layer rather than an infrastructure operator.

KPIs

  • • Determinism (identical outputs for identical settings)
  • • Curvature stability (consistent DC.motor behaviour)
  • • Metric coherence (reproducible DC.metrics signatures)
  • • Interpretability (clear dimensional identities)
  • • Cross-scale consistency (micro to macro geometries)
  • • Agent reliability (contract-driven workflows without drift)

Benchmarks

  • • Canonical geometries (rings, diagonals, radii, offsets)
  • • Chromatic transitions
  • • 0→1 curvature sweeps
  • • Reproducibility across repeated runs

ADI enables deterministic, interpretable, cross-scale computation for physics, materials, optics, chemistry, climate systems, and infrastructure engines (including data centres). It strengthens European technological sovereignty by providing a native computational architecture and measurement standard developed in Europe.

Compute

  • • GPU type: NVIDIA A100 or equivalent (40–80 GB)
  • • Total GPU-hours: ~800
  • • Reason: high memory bandwidth for real-time curvature computation and chromatic field generation.

Budget

Total Stage 1 cost estimate: ≤ €350,000 (within the SPRIND maximum of €3,000,000 for Stage 1).

Stage 1 requires fewer than 1,000 GPU-hours total. The workload is deterministic and geometry-driven; no training, sampling, or large-scale optimisation is required. Compute is used only for systematic sweeps, operator validation, and cross-scale benchmarking.

The project is led from Dunckerplatz 4, 14712 Rathenow by a founder with deep experience in conceptual systems, geometric reasoning, and cross-disciplinary synthesis. The team combines expertise in computational geometry, chromatic systems, deterministic engines, and scientific modelling.

Key capabilities include designing and formalising new dimensional operator classes (DCO: DC.motor, DC.colour, DC.metrics), building deterministic computation pipelines, developing interpretation layers and metric frameworks, and integrating multi-agent workflows for scientific exploration. The team is supported by collaborators in physics, materials science, optics, and computational modelling.