The tabs that are not part of the system

The background

A simulator can show every part of itself working and never answer what would I do differently if this were fresher? A system that measures the world and also decides is two systems; merging them hides the boundary an evaluator most wants to question.

The requirement

Three notional systems consuming the forecast to reach a decision: unmistakably not part of the simulator, and checkably consuming it rather than illustrating it.

The options considered

Build them elsewhere. The cleanest boundary, rejected on the demonstration: the moment worth showing is a forecast published in one tab and a recommendation going stale in another.

Keep them inside and mark them. What was built — yellow, under a strip that cannot be dismissed, declared in configuration so a fourth cannot arrive without it. ADR-0039 settles the conflict with drogna's rule against asserting what is not running: a consumer may synthesise its own inputs, never drogna's, and says so.

What makes it checkable. A new forecast triggers no recalculation. The tab grows a halo and waits; on the click, the previous answer stays as a ghost. If the recommendation barely moves, that forecast was not decision-relevant. A tab faking its inputs would have nothing to go stale.

The demo

Sampling — a hex grid coloured by observation coverage, routed by value per mile. The grid covers what is in view, making fine resolutions affordable; 3 hours against 24 changes the route's shape, not its length.

Courses — three classes of vessel that may be present, each moving by its own model rather than a multiplier. Drag the weighting; the ranking reorders.

Feasibility — no map, deliberately. Ten lanes wearing their provenance, and the maximal feasible sets: A and B, or B and C, never A and C.