ETR technology

From black-box prediction to causal, traceable and adaptive intelligence

ETR (Emotional Theory of Rationality) is a CELL-based cognitive architecture that combines adaptability, resource efficiency, native traceability and explainability for ML, RL and MoE approaches. Its current version, ETR 0.5, is deployed in pilots across industrial, logistics, real estate and traffic-control use cases.

ETR 0.5 Engine Capability levels

01The challenge

The Black Box problem, addressed by architecture

The Black Box is one of the most significant open problems in AI: models whose decision-making processes are difficult to predict and interpret.

ETR does not add an explanation layer on top of an opaque model. Events, relations, decisions and consequences are stored in an inspectable, glass-box knowledge graph as the system operates.

Opacity

We don’t know what, where or how statistical AI stores its knowledge (correlations).

Trustworthiness

Black Swans: statistical models can produce nonsensical responses in the long tail of the unknown.

Adaptability

They struggle to conceptualise and extrapolate available knowledge to new problems.

Efficiency

Learning requires substantial amounts of data, time, energy and compute.

02Inside one CELL

Bounded local processing, shared consequences

Each CELL performs regulation, prediction, learning and evaluation, and shares its events, decisions and consequences with the rest of the system. Watch how complexity is divided, shared and traced back to its source.

Small data in, cognitive response out. The system works from its current state, not from a big-data training set.
Memory, prediction and learning. Short- and long-term predictions are compared with outcomes; the prediction model is updated from experience.
Emotional evaluation. In ETR, emotions are functional: they evaluate criticality and allocate scarce cognitive resources.
Cognitive branch and decision trail. Model-based response; events, decisions and consequences are written to the Shared Knowledge Graph.

03Architecture

Five differentiating principles

Bio-inspired by the founder’s neuroscientific research since 2005, peer-reviewed (Garcés & Finkel, 2019) and funded by €1.4M in R&D public grants since 2021: NEOTEC, INVESTIGO, public-private collaboration, EIC Accelerator Step 2 and NextGenerationEU.

Energy and time as organising constraints

Intelligence treated as efficient resource allocation under energy/time scarcity: what deserves attention, computation and action.

CELL-based architecture

Small adaptive units process bounded information locally and interact through a common mechanism.

Emergent complexity

Global behaviour is not fully programmed in one model; it emerges from the dynamic interaction of multiple CELLs.

Shared Knowledge Graph

Events, relations, decisions and consequences are stored in an inspectable shared space. Traceability is native, not reconstructed after the fact like post-hoc XAI (SHAP, LIME…).

Emotions as functional optimisation

In ETR, emotions are not feelings: they evaluate criticality, prioritise stimuli and allocate scarce cognitive resources (Garcés & Finkel, 2019).

04Evolutionary insight

Knowing the parts is not understanding the system

302

neurons in C. elegans

Wiring mapped in 1986 (White et al.), behaviour still not predictable from structure alone.

86Bn

neurons in a human brain

Orders of magnitude more components…

20W

…on the power of a light bulb

Efficiency comes from regulation, not from count. Scaling parameters is not the only road.

05Capability-gated roadmap

Evolution-inspired capability levels

Evolution-inspired, not biologically simulated: each level is a measurable capability, a benchmark and a commercial use case. The upside is capability-gated, not a binary bet.

Productised · ETR 0.5
WORM
Reactive regulation
Maintain critical variables and prioritise responses under changing inputs.
Control / monitoring
Productised · ETR 0.5
FLY
Adaptive learning
Learn non-linear action–outcome associations and update behaviour from experience.
Adaptive automation
Next gate · 2027 · ETR 0.6
BEE
Operational causality
Test action–consequence relationships, adapt continuously, remain efficient and traceable.
Industrial critical systems
2028
MOUSE
Context + planning
Episodic context, persistent goals, multi-step planning and anticipation.
Unlocks autonomous systems
2029
PRIMATE
Abstraction
Transfer useful structure across tasks; cooperate and coordinate across CELL groups.
NPCs & interactive agents at consumer scale
AGI target · gated, not dated
HUMAN
General cognition
Open-ended generalisation, language, metacognition and cross-domain reasoning.
General cognition

Current platform: ETR 0.5 covers WORM and FLY in productised form, with native traceability across both. Next gate: BEE, operational causality (ETR 0.6, 2027). AGI remains the destination; each gate is the investable milestone.

06Why now

Auditability has deadlines

EU AI Act high-risk requirements turn traceability, logging, technical documentation and human oversight into procurement issues, not optional extras.

Traceability
Logging
Technical documentation
Human oversight

27 Jul 2026
Digital Omnibus in force
Reg. (EU) 2026/1744 sets the high-risk timetable.
Today
The buying window is open
Buyers must adapt procurement and technical documentation before the deadlines.
2 Dec 2027
Standalone high-risk systems
Annex III systems must comply.
2 Aug 2028
AI embedded in regulated products
Annex I: medical devices, machinery, vehicles.

07Applications

One architecture, infinite applications

ETR ships as a software engine integrated into third-party systems, and complements statistical and GenAI components where decisions must remain adaptive, auditable and controllable.