Skip to main content
V3 is in beta. General availability (GA) is planned for 1 December 2026.
A fourth-party logistics coordinator plans a Monday of pallet transport across Belgium for a client. Nine orders have to move between a container terminal, suppliers, a cross-dock and distribution centres, each with a pickup window and a delivery window. One container has to clear a customs scan on the way out of the port before it goes anywhere, and if the scan cannot happen the container stays. One order is dangerous goods and needs a certified truck. One is a spot-market load the coordinator will take only if a truck is passing anyway. The three trucks are not alike, and their drivers’ driving time is regulated. This page models that day, solves it with POST /v3/routing/solve, and reads the plan: how a shipment appears as two stops, how a chain of three stops is kept together and served as a whole or not at all, where the driving-time breaks fall, and why two orders are left out with a number attached to each. The response is real output for the request shown. Yours can differ in the travel figures, if the road data has changed since, and in timings.

The business rules

The chain rule is the one that needs three relations, and each does one thing. consecutive forbids anything between the three stops when they share a route. same_route makes them share a route. all_or_none makes them stand or fall together, and it is allowed on this group because the group is exactly one consecutive chain; that is the form that lets a shipment id stand for both its legs. Drop any one of the three and the rule changes: without same_route, the scan could go on another truck; without all_or_none, the scan could be served and the container left behind.

The data

All times are local time on 17 November 2026, written with the +01:00 offset. Coordinates are [longitude, latitude]. Every pickup and delivery takes 25 minutes. The nine orders are 140 pallets; the three trucks have 84 places between them, so the day only works with trucks delivering and loading again, and some orders may not fit.

The request

The complete body, as sent. POST /v3/routing/validate/solve checks the relations before anything else: it rejects a relation that names a shipment id where only a leg id is allowed, an all_or_none group that is not one consecutive chain, and a consecutive relation that gives a stop two successors.
Three things to notice before reading the answer:
  • A shipment’s legs are named {id}:pickup and {id}:delivery. Those are the stop ids the relations use and the ids the response reports on the two stops. The shipment’s own id appears in unassigned[].task and in all_or_none, where it stands for both legs.
  • required_skills and drop_cost sit on the shipment, not on a leg. They apply to the pair and are priced once.
  • limits sits on the vehicle, max_overtime_s on the shift. The duration cap is a property of the truck; the overtime allowance belongs to a working day. Both could be moved to the other level if the fleet needed it, since a shift’s limits override the vehicle’s.

The response

The response below is the real output for that request, trimmed in three ways to keep it readable: each stop keeps its times, waits, travel and load and drops location, snapped_location and service_s; the top-level search_progress and warnings are left out; and plan is shown whole because it is what you send back. Response times are in each shift’s own offset, +01:00.

The plan in one paragraph

All three trucks go out and six of the nine orders move, 86 pallets, plus the customs scan. truck-a does the port chain and nothing else: terminal, scan, Zaventem, home by 09:46. truck-b runs the two big western pickups, Zeebrugge and Kortrijk, to Namur and Mechelen, with its driving break in Namur. truck-c, the small ADR truck, runs a loop of four orders through Ghent, Roeselare, Sint-Niklaas, Turnhout and Leuven, and is the last home at 15:17. Two required orders, Aalst to Liège and Ghent to Hasselt, are unassigned, each with a relaxation that says what would have let it in. The spot load to Mons is declined at its drop_cost.

Summary

  • tasks reads served: 7, unassigned: 3, unassigned_required: 2. A shipment is one task; the seven served are six shipments and the scan job. Of the three unassigned, the spot load has a drop_cost and does not count as required. HTTP 200 all the same.
  • distance_m 1,009,462: a thousand kilometres over three trucks. time.travel_s 43,308 s is twelve hours of driving between them; time.wait_s 14,059 s is time spent at an address before its window opened; time.break_s 5,400 s is two 45-minute breaks. time.overtime_s is 0: nobody needed the hour past 17:00.
  • search used the five-second budget in full (termination: "time_limit") over 116,596 iterations, far more than the parcel example in the same time, because a problem with ten tasks has few positions to try.

The chain on truck A

truck-a’s stops are start, ord-port-zaventem:pickup, customs-scan-port, ord-port-zaventem:delivery, end. Nothing sits between the three, which is what consecutive demands, and all three are on one route, which is what same_route demands. The times show the windows at work: That last line is worth a look. truck-a is free from 09:46, and two orders are unassigned. The solver did not find a way to use it: the Aalst pickup closes at 09:00, while the truck is still in Zaventem; the Ghent cross-dock closes at 10:00, and the response’s relaxation says the truck would have been 14 minutes too late. The unassigned section below reads both. The scan could not have gone elsewhere. Had the request listed only consecutive and all_or_none, the solver would have been free to put the scan on truck-c as a separate job, and consecutive would have been satisfied trivially, because it binds only members that share a route. same_route is what makes “right after” mean what the planner meant.

Trucks B and C

truck-b picks up at Zeebrugge (20 pallets) and Kortrijk (12), for a load_peak of 32 on a 33-pallet truck, delivers Namur at 10:12 and Mechelen at 12:21, and is home at 13:02 after 434 km. Its break stop is at Namur, 10:37 to 11:22, listed before the Mechelen delivery: the truck had driven 3 hours 36 minutes by then, and the next leg would have taken it past 4 hours 30, so the break goes before the leg. The counter resets after it. truck-c is the busy one: four orders, 18 pallets on board at the peak, which is its whole capacity, 430 km and ten hours. Two things to read on it:
  • Waiting at the receiver. It reaches Sint-Niklaas at 08:52 with the Roeselare order, and the receiver opens at 10:00: wait_s: 4046. It uses the stop to pick up the ADR order at the same address straight after (ord-chem-turnhout:pickup at 10:25), which is why the plan is willing to wait: the two stops share an address and the second is the only ADR order, which only this truck may carry.
  • The break at Turnhout. 12:25 to 13:10, after the delivery and before the long leg to Leuven, which arrives at 14:16, inside its window.
Both breaks are triggered breaks and have no window of their own; their position is a consequence of the route. Neither truck’s route exceeds max_route_duration_s: truck-c’s span is 37,023 s against a cap of 46,800.

The unassigned orders

Three entries, three different stories:
  1. ord-aalst-liege is blocked by a time_window. Some truck could do Aalst to Liège on its own; none can alongside what the plan already carries. The relaxation says the pickup window, which closes at 09:00, would need to be 26 minutes 41 seconds longer. truck-a is the truck in question: after the scan it reaches Zaventem at 08:00, and Aalst is on the other side of Brussels. A supplier that loads until 09:30 would have made the day whole.
  2. ord-gentxd-hasselt is blocked by capacity and a time_window, and gets a relaxation for each. The capacity reason is truck-c: 24 pallets on an 18-pallet truck, so reduce the load by 6 or add 6 places. The window reason is the two big trucks: the cross-dock closes at 10:00, and the nearest of them is 14 minutes 19 seconds short of it. Each figure is the measured shortfall for that vehicle; re-solve to confirm the order is placed.
  3. ord-spot-mons is priced_out. It fits, it has a drop_cost, and serving it at the cheapest position found would cost at least 180, so the solver paid the 180 instead. The single reason has no constraint: no rule kept it out. The 180 shows up in score.components.drop_costs, and unassigned_priority does not count it, because a task with a drop_cost is not a required task.
A relaxation is a measurement, not advice. Widening a supplier’s window by 27 minutes is a phone call the coordinator can make; splitting a 24-pallet order is a different conversation. The response makes both sizes visible so the right conversation can happen. Note too what the response does not suggest: using truck-a’s idle afternoon. It cannot, because every unassigned pickup closes before truck-a can reach it, and the relaxations are the proof.

Score

score.cost is the sum of score.components, in the currency you labelled: unassigned_priority is 2: the two required orders left out, at the default priority of 1 each. Had the Hasselt order carried priority: 5, the solver would have ranked a plan that serves it and drops two others above this one, whatever the cost difference, because unassigned priority is minimised before money.

One KPI line

Cost per delivered pallet: 29.98 (2,578.10 over 86 pallets), with 86 of the 126 required pallets delivered and two orders, 40 pallets, left for tomorrow or a subcontractor. The figure includes the 180 paid to decline the spot load; take drop_costs out of cost to compare the fleet’s own efficiency across days.

What the solution cannot prove

The response proves what the solver was told to respect. It cannot prove what it was not told:
  • That the windows are the ones the sites will keep. A pickup window of 06:00 to 09:00 that the supplier reads as “we open at 06:00 and load until noon” is a 26-minute phone call away from a fuller plan, and the response cannot make it.
  • That 25 minutes per pallet exchange is right. A terminal that takes an hour to release a container turns truck-a’s 26-minute wait into lateness at the DC, which the plan cannot absorb because the DC window has no price. wait_s per stop shows where the plan has slack; truck-c has almost none after 10:25.
  • That the 4 hours 30 rule is the only rule. A triggered break models one break after one accumulator. Daily and weekly driving limits, split breaks and rest periods are outside what this request says, and a plan that keeps the one rule can still breach the others.
  • That the trucks can load together. The plan has truck-b carrying Zeebrugge and Kortrijk pallets at once. Whether 32 pallets from two shippers can share a trailer in practice is something the request did not state, and capacity alone does not say.
  • That the spot load should stay declined. priced_out means “at 180, no”. The cheapest way to serve it is somewhere above 180; the response does not say how far above. To find out, send the plan to /v3/routing/suggest with the spot load left out of plan: the first insertion’s delta.cost is the price of taking it.
What it does prove, and how to check it in code: every shipment on a route has its pickup stop before its delivery stop on the same route; ord-port-zaventem:pickup, customs-scan-port and ord-port-zaventem:delivery are three consecutive stops of one route; the ADR order is on truck-c; each route’s load_peak is within its vehicle’s capacity; every route’s time.span_s is within max_route_duration_s; every task stop’s service_start is inside its window; and every task is either in some route’s stops or in unassigned[]. Relations and Shipments list the fields rule by rule.

Next steps

Relations

All ten relation types, what each binds, and the common 400s.

Suggest

Price the spot load: ranked insertions into this plan.

Handling infeasibility

Reading categories, reasons and relaxations, and the ranked fixes.

Waste collection

The same loop with mid-route reloads and depot intake caps.