Five more situations. The same instrument in all of them.

The policy scale is not a switch. It runs from +3 to −5, and it is asymmetric on purpose: the penalty for an option we want avoided is heavier than the reward for one we prefer. That is how a city reasons about restriction, and it was in the constants a decade ago.

Every point carries its own transfer policy, including one that is not a pickup point at all. Toompea is refused, and the refusal is a policy value rather than an omission.

Two weightings are scored side by side and the authority takes B. Punctuality falls from 0.91 to 0.78. The cost is recorded next to the choice, because the decision has to survive being asked about in public, and a decision whose price was never written down cannot.

Current demand is measured against the historical curve for that hour and destination, and the projection is turned into vehicles: three buses, four shuttles, five taxis, each sized by its own capacity and expected occupancy.

One vehicle type is refused. The nineteen-seat minibus would fit the arithmetic and is not available in the window, so it does not appear in the plan. A capacity estimate that ignores availability is a wish.

The output is not a warning. It is an instruction an events department can act on before the peak arrives, which is the difference between a dashboard that reports a problem and a system that hands you its solution in units.

Four processes read from the incumbent: vehicles, geofences, bookings, executed passengers. A fifth is refused, because the incumbent does not expose drivers, and the screen says so rather than hiding the gap.

And in the other direction, one field. The unit assigned to a reservation. That is the entire write-back, and it was specified that way in a versioned connectivity document in 2017, not decided later as a reassurance.

This is the answer to the fear an operator will not say out loud. The layer above does not want the customer, the record or the dispatch. It wants to read what you already have and hand back an assignment. Their console keeps the record, and the count of records created here is zero.

Two options are refused and one is held. The carpool repair cannot take nine, the public line is suspended, and the partner shuttle is forty minutes away and stays on the bench. The taxi pool is taken at 96.40.

Then the interesting half: the cost splits four ways under a contract rule, and the four amounts reconcile exactly to the cost. Operator, authority, passengers at a capped contribution, and the driver who accepted the detour.

This is what a procurement officer actually worries about. Not whether the system optimises on a good day, but who pays when it does not, and whether the answer was agreed in advance or argued afterwards.

Three pressures are multiplied into one number. Multiplied, not added, so trouble in two dimensions at once compounds instead of averaging away. Two sheets cross 1.00 and both are told to slow down; one is far below it and is told to speed up.

Read the instruction itself, because it is the strongest sentence we own: Aumentar velocidad promedio con seguridad. Increase average speed safely. All three values end that way. The system never told a driver to go faster; it issued an instruction to a person, carrying a duty of care, and stored it as a controlled vocabulary with a code and a readable label.

And the instruction is written to the route sheet, not shown on a screen for somebody to notice. Forty-three sheets receive nothing, because nothing was wrong with them. A dashboard measures. This measured, decided, and wrote the decision back into the operation.