Methodology
How we work.
The process is always the same, whatever the context. The documents change, the actors change, the question changes. The structure is the constant.
Only what bears on the question
TNT receives the relevant documentation, in any format. The system does not absorb it wholesale: a selection stage carries forward only what bears on the work at hand. Less material means a tighter perimeter, less noise and a firmer result.
A closed and checkable context
The selected material is organised into a structured, interlinked knowledge base, the only informational universe the system may operate within. It is also the record of what the system actually used: if an output states something, the document behind it can be found.
Where we part company with document review
Up to this point, the process resembles rigorous document management. From here, it does not.
The relevant parties are represented as distinct agents. Each has its own objectives, constraints and available information, and a behavioural profile reconstructed from the documentation: what it has done before, what it responded to, where it gave ground and where it did not.
The question we then put to the system is no longer what the documents say, but how the people and organisations described in those documents move as the situation develops.
That is the difference between knowing what happened and understanding what can happen. It is not a statistical forecast built on historical series: it is a simulation of the behaviour of the parties involved, returning the trajectories consistent with what we know about them.
Trajectories side by side
The system generates alternative trajectories and sets them side by side, spelling out for each one the conditions that make it possible, the turning points, and what each one implies.
Human validation and delivery
Before delivery, outputs are reviewed and validated by our professionals. The deliverable includes structured documentation of the analysis and a summary written for the people who have to decide.
Limits
What this method is not.
No system built on language models is free from error. We work on reducing that risk and on making the result checkable: a closed information perimeter, traceable statements, human oversight at the control points.
The scenarios we produce support a decision. They are not a guaranteed forecast, and they do not replace the judgement of the decision-maker or the advice of the professionals assisting them.
Outputs are generated with the assistance of artificial intelligence systems, under human supervision.