Lab
One continuing workspace: build a route, schedule the dinners and coordinate shared airspace.
Every week's case in one workspace. Pick a week on the slider, press the one change that week asks for, then open Change inputs to try your own.
Week 4 · Payload, service, return energy and reserve
The fastest route may not bring you home
Which time–energy labels must survive until the return is checked?
Starting case ready.
#07: 472 s from take-off to landing; 80443 J including service and the unloaded return. 14557 J remain on landing.
Constraints: Complete plan; checks passed. Delivery quality: 1/1 on time. Search: Selected method completed; no global optimality claim.
More views
Map legend and layers
Kitchen = dispatch. Numbered homes mark delivery addresses. Teal doorstep markers are the exact delivery points; dotted paths link them to their houses. Grey streets are landmarks, not extra flight edges. Solid arrows fly out; dashed lines return. Select a table row or map marker. In 3D, drag to orbit and scroll to zoom; camera buttons work with a keyboard. 3D height is shown at 3× scale; labels use real metres.
Candidate colours show the outbound approach; the dashed line is the selected unloaded return. Candidate costs cover the complete trip, including service.
Route elevation · actual waypoint heights
Read these lines with the route legend above. Distance includes the complete outbound and return trip.
Play or step through the recorded events.
Inspect time and shared resources
Time and shared resources
NowThe vertical line marks the current time; outlined intervals are active. Select an interval to jump to its start. Intervals include their start and exclude their end.
0 s 3138 s
Inspect the evidence tables
Evidence from this run
| Inspect | Candidate | Out (s / J) | Service (s / J) | Back (s / J) | Total (s / J) | Reserve check |
|---|---|---|---|---|---|---|
| 0 | 201 / 56539 | 45 / 4500 | 201 / 23708 | 447 / 84747 | over budget | |
| 1 | 222 / 53212 | 45 / 4500 | 201 / 23708 | 468 / 81420 | over budget | |
| 2 | 226 / 52235 | 45 / 4500 | 201 / 23708 | 472 / 80443 | fits |
Correct Pareto strategy, same input 472 s / 80443 J
Run identifiers
Engine lab-1.0 · hilltop · 5.82 ms · input fnv1a32:509d191a · model fnv1a32:0cb5e08e. Times are seconds unless this is a symbolic case.
Change this experiment
Strategy code and presets
Change a supplied rule or edit its function body. Run evaluates the whole experiment with this strategy.
Scenario and manual planning inputs
Model assumptions
One drone; complete trips · No shared charging in this model · Corridor reservations not active
Saved experiments
Check your explanation
Record full-trip seconds, kJ and the usable budget. Explain why a faster label cannot dominate a slower one that uses less energy.
How five model upgrades change the same instance
Five upgrades, one instance
The same kitchen, the same twenty orders, the same hill. Each stage changes the model, breaks something the previous stage relied on, and keeps the rest. What was right under an earlier model is not wrong in hindsight; it is labelled with its model.
Find a route Weeks 1–3
- What changed
- The evening became a graph: streets as edges with a time cost, buildings deciding which connections are legal.
- What it broke
- The straight line from the kitchen to the customer.
- What survived
- The map. Everything after this searches it.
Dijkstra from the kitchen to #07 on the hilltop: 201 s one way over 8 streets, 43 expansions.
Find a route the drone can fly and return from Week 4
- What changed
- Energy is counted along the whole round trip, loaded out and unloaded back, and a steep climb costs more per metre of rise.
- What it broke
- Week 3's fastest route to #07: it does not fit the light drone's budget.
- What survived
- The search loop, now over (time, energy) labels that are kept until the whole trip is costed.
Fastest round trip 447 s at 84.7 kJ; budget 80.8 kJ; flown: 472 s at 80.4 kJ.
Order many deliveries for one drone Weeks 5–6
- What changed
- One order became six, each with a ready time and a promised time, and an objective: total lateness, then the time the drone is back, then energy.
- What it broke
- "Each order is feasible on its own" as a plan.
- What survived
- Each task's cost from week 4; the timetable only adds them up in an order.
#01–#06 on one light drone: FIFO 0:00 late, back 18:42:38; earliest deadline back 18:45:38; swaps stop at back 18:42:38; the best of 720 permutations is back 18:42:38 — the same as the swaps found.
Divide the work between drones that share two charging pads Weeks 7–8
- What changed
- Five drones of two types, and two charging pads that only two of them can use at once.
- What it broke
- "Available again" as "back plus turnaround": a drone's next start is when a pad is free.
- What survived
- The recurrence, with "available" set by the pad queue instead of the drone's own clock.
Make five plans executable in one airspace Weeks 9–12
- What changed
- The corridor is a shared resource with capacity one; a state is a place and a time.
- What it broke
- Week 2's shortest paths for two drones that want the corridor at the same time.
- What survived
- Each drone's route, as the proposal the reservation table answers.
See the corridor case above: two shortest routes, one conflict, and the two arrangements that pass the validator.