In people
We settle on the most familiar reading.
Experience, emotion, and the stories an organization tells itself decide which futures it can even see.
Driven by the Unusual
Reconfiguring inferential boundaries
to expand reasoning under uncertainty.
When the future stops behaving like the past, the costliest mistake is settling on the obvious answer too soon. I research how people and AI systems can keep more options open before they decide, and how to turn that wider view into decisions a team can explain and defend.
Beyond the probable
A language model is built to find the most probable continuation. Strategy often depends on the improbable one.
The next frontier of generative AI is defined by how we dialogue with each model: extending inferential exploration beyond probabilistic weights and controlling generative outputs.
Most companies now have access to the same models. The advantage moves to how you use them: which questions you ask, how many readings of a situation you allow before deciding, and whether you can trace how a recommendation was produced.
Delay the closure, open the responses
Current systems are optimized to resolve uncertainty as quickly as possible. They converge toward statistically dominant configurations, closing the space of possibilities before it can be meaningfully explored.
People do the same. Under pressure, teams lock onto the familiar reading of a market, a crisis, or a customer. The cost shows up later, as the scenario nobody modeled and the opportunity nobody saw.
In language models
Models converge toward the statistically dominant configuration.
A more capable model or a longer prompt fills that closure with more detail. It does not reopen it.
This is not about improving answers. It is about expanding what can be answered.
Two lines of research
The first gives your models, agents, and analytical processes a way to explore more before they conclude. The second helps organizations build capabilities with AI that a software license alone cannot buy.
1Inside AI reasoningReasoning OrchestrationAn architecture that expands exploration and keeps every result traceable.Six frameworks, from scenario modeling to issue dynamics, examine the same challenge through distinct analytical logics. You can use them as apps, run them with your own API, or have the laboratory execute the complete architecture.What neither logic can see alone appears where one hands its output to the other.Go to Reasoning Orchestration2Between people and AIHybrid CompetitivenessA new organizational architecture for hybrid competitiveness.Organizations using the same models get very different results. Generative Change Management explains where the difference comes from and how to design it into roles, workflows, and culture.The expectations you place in an AI system shape what you ask it, and so what it can become for you.Go to Hybrid Competitiveness
Reverse Extrapolation
We model the future to redefine the present.
The future is not a prediction. It is a design space, shaped by variable interactions, structural tensions, and transformative forces that conventional analysis doesn't reach.
Instead of projecting last year's trend line forward, the method maps the futures a situation could reach and works backward to the signals and decisions that shape them today. You get options with triggers, not a single forecast.
Explorealternative future states
Generate structurally different configurations that a situation could reach.
Reverseconditions and signals
Identify the forces, tensions, and emerging signals that can activate each configuration.
Actpresent decisions
Translate possible futures into criteria, interventions, and strategic choices for today.
Epistemological Orchestration
Can controlled differences in reasoning produce analytical configurations that cannot be recovered through synthesis alone?
We work on a reasoning architecture at the inferential control layer.
CODHZ, the laboratory, is where the research is tested, measured, and made usable. It publishes open-access papers, offers an environment where anyone can verify the results, and runs the complete architecture for organizations and AI labs.
Make the invisible visible
Questions about the research and proposals for academic or laboratory collaboration are welcome.
Write to Marcelo