The operating picture is not the hard part. The next move is.

Every fusion programme of the last fifteen years has ended in the same place: a very good map. Tracks correlate, symbols are standard, latency is acceptable, the operator can see everything at once. This is a real achievement and it is also where the work usually stops.

The problem is that a complete picture of the present is not what anyone is actually asking for. The question in the room is never "what is happening". It is "what happens next, how sure are you, and how long do I have to decide".

Why the gap persists

Three reasons, in rough order of how often we see them.

Fusion is scored on the wrong axis. Programmes are evaluated on track purity, correlation rate and time to display. All of those measure fidelity to the present. None of them measure whether the system told you anything useful about the next interval. A system can be perfect on every acceptance criterion and still be silent on the only question that matters.

Prediction gets treated as a model problem. It is mostly a data problem, and specifically an outcome problem. To forecast anything you need a record of what actually happened, joined to what the system believed at the time. Very few operational systems keep that join. They keep the picture and they keep the logs, and the two are never reconciled, so there is nothing to learn from.

Nobody wants to be wrong on the record. A picture is a statement of fact and carries no risk. A forecast is a claim that can be graded. Institutions optimise against being graded, so the forecast gets softened into a description, and the description is what ships.

What we think is required

A foresight system has to do three things the picture does not.

It has to commit. Every projection is stored with a timestamp, a horizon, and a confidence, before the outcome is known. No retrospective adjustment.

It has to score itself. When the horizon passes, the projection is compared against what happened and the result goes on a permanent scoreboard. Calibration is measured, not asserted. A system that claims eighty percent confidence should be right about eighty percent of the time, and the operator should be able to check.

It has to carry its own uncertainty into the display. An operator who can see that this class of projection has been reliable and that class has not will use both correctly. An operator handed a single number with no history will either over trust it or ignore it, and both failures are expensive.

The scoreboard comes first

The instinct is to build the model and add measurement later. We think that ordering is backwards, and it is why so many of these systems end up unfalsifiable. Build the scoreboard first, run it against the dumbest possible baseline, and only then start improving. If a new model cannot beat persistence on the record, it has not earned its place in the operating picture.

That constraint is uncomfortable and it is the entire point. A foresight claim that cannot be graded is not foresight. It is a map with an opinion attached.