The reasoning loop
A closed reasoning loop, not a collection of features.
Theralon doesn't just model the operation. It continuously reasons over its changing state — and that reasoning isn't a set of independent analytics running side by side. It's a single loop, where each stage hands its work to the next.
Establish what is actually true.
Large operations routinely hold competing versions of the same reality — a shipment shows delivered in the ERP, the warehouse hasn't received it, the carrier says delayed, and telemetry shows the vehicle still moving. Theralon determines the most defensible current state without overwriting the disagreement or pretending one system was simply wrong.
Determine which way it's moving.
What direction is the operation heading, how fast, and is that movement actually meaningful — or just noise around a normal range?
Follow what it causes downstream.
Not "Supplier A connects to Plant B", but: Supplier A is deteriorating; Plant B depends on it for a component with a two-day buffer; Plant B's schedule will become constrained; those orders serve customers C, D and E. Dependency, timing, capacity to absorb and recovery, modelled together.
Consolidate what's actually at stake.
Risk answers how likely and how severe. Exposure answers what that puts at stake — and groups it. Because the same underlying dependency surfaces through different units and categories, the honest statement is rarely "47 high-risk items". It's that 47 observed conditions consolidate into 11 real concentrations.
Weigh the action against the whole network.
The best decision is frequently not the one with the biggest direct benefit. Expediting a shipment may reduce one risk while creating transport cost, warehouse congestion, and a new bottleneck downstream.
Learn from what actually happened.
What was predicted, what was decided, what was executed, and what followed — kept as one connected record, so the next recommendation is better than the last.
Each stage enriches what came before rather than overwriting it, so the chain from a recommendation back to the observations that produced it stays intact and inspectable. Underneath sit shared capabilities any stage can draw on — forecasting, simulation, optimisation, machine learning, rules and causal inference. These are techniques, not stages: forecasting matters because Trajectory, Propagation and Intervention use it, not as a destination in its own right.
The output of all this isn't a report. It's a continuously better representation of reality.
Confidence
A confidence score you can interrogate.
A single confidence number is easy to produce and almost impossible to trust. Theralon's confidence is a structure: evidence quality and source agreement beneath Truth, baseline strength beneath Trajectory, dependency and timing beneath Propagation, valuation beneath Exposure, feasibility and historical effectiveness beneath Intervention.
These are never simply averaged. Some are floors rather than terms — an outstanding score in five dimensions doesn't compensate for a critical weakness in one. If alternative supplier capacity is essentially unobserved, no amount of freshness elsewhere makes the conclusion sound.
Cause: alternative supplier capacity is only partially observed.
Because confidence carries provenance the same way facts do, a recommendation can be traced back through exposure, propagation and trajectory to the individual claims beneath it. An operator can establish that a recommendation is weak because one specific fact came from a stale source — rather than being asked to accept a score.