Intelligence

Different questions need different engines

There is no single model that answers everything. Aurant runs rules, statistics, graph algorithms, time series, geospatial models, forecasting, machine learning, solvers and simulation, and routes each question to the engine that can actually answer it.

Routing the question

                     QUESTION
                        |
      +-----------------+-----------------+
      |                 |                 |
 descriptive        predictive       optimization
      |                 |                 |
What happened?    What happens next?  What should we do?

The Problem object

A problem is not a dashboard tile. It is a first-class object with a lifecycle, evidence, candidate causes, an owner and a resolution record.

PROBLEM  #<id>

Type       <problem type>
Location   <administrative unit>
Severity   <level>
Started    <timestamp>
Affected   <estimated population>

Evidence
   <sensor / telemetry / report streams>

Probable causes
   cause A   <probability>
   cause B   <probability>
   cause C   <probability>
   unknown   <probability>

PROBLEM -> CAUSE GRAPH -> IMPACT GRAPH
        -> AVAILABLE ACTIONS -> SIMULATION
        -> RECOMMENDATION

Because the problem is an object, it can be assigned, escalated, merged with a duplicate, reopened when the impact returns, and measured after the action.

Causal graphs, honestly labelled

Instead of the sentence 'inflation is increasing', the platform shows the chain, and marks how each edge is known.

Fuel imports  ^
      |
Fuel price  ^ ------------+
      |                    |
      v                    v
Transport cost ^     Electricity cost ^
      |                    |
      +---------+----------+
                v
        Production cost ^
                |
                v
           Food price ^
                |
                v
          CPI inflation ^

observed

Observed relationship

Both series are measured, and the link is recorded in the data itself.

correlation

Statistical correlation

Movement co-occurs. No direction and no mechanism is claimed.

estimated

Model-estimated causal effect

A named model with a stated method and interval produced the edge.

hypothesis

Human hypothesis

An analyst asserted it. It is visible as an assertion, with the author attached.

A decision system that blurs these four is dangerously overconfident. Aurant keeps them separate at the data model level, not in a footnote.

Simulation

Policy questions are branch questions. The platform builds scenarios off the current state model and compares their modelled consequences side by side.

CURRENT WORLD
      |
  +---+-----------+-----------+
  |               |           |
Scenario A    Scenario B   Scenario C
  |               |           |
transport      transport   transport
cost           cost        cost
  |               |           |
inflation      inflation   inflation
  |               |           |
budget         budget      budget
impact         impact      impact

Result table: each column reports the modelled
outcome, the assumptions behind it, and the
confidence interval. The AI explains the model.
It does not invent the numbers.

Assumptions are visible

Every scenario lists the parameters it holds fixed. A result whose assumptions are hidden is not a result.

Ranges, not single numbers

Outputs carry intervals. A point estimate presented alone invites false precision in a political room.