SkillCompiler/data/skills-bench/tasks/paratransit-routing/environment/model-and-data.md

Paratransit Routing Model And Data

This document defines the data conventions, feasibility rules, objective, and route-output schema for the full-day dial-a-ride instance.

Data Conventions

• requests.json: passengers in file order. Each passenger has a passenger_id and one or more trip requests.
• t_matrix.csv: square integer travel-time matrix in minutes. Negative entries are invalid arcs.
• instance_config.json: nb_passengers, nb_trips, nb_vehicles, vehicle_capacity, and time_window_width.
• Times are integer minutes after midnight; for example, 300 means 5:00 AM.
• requests.json and t_matrix.csv are intended to be parsed from disk with code. Avoid opening, printing, or copying the full request list or matrix into the model context; compact summaries are enough for debugging.

Let n be the number of trips after flattening requests.json by passenger order, then trip-list order.

• Node 0: start depot.
• Nodes 1..n: pickups.
• Nodes n+1..2n: dropoffs.
• Node 2n+1: end depot.

For flattened trip index i starting at 0, pickup node is 1 + i and dropoff node is 1 + n + i. Matrix rows and columns follow this node order.

Feasibility Rules

• A served trip has exactly one pickup and one dropoff.
• Pickup and dropoff for a served trip are on the same vehicle, with pickup before dropoff.
• Trips for the same passenger form one request set. A request set counts as served only when every trip in that passenger's set is served; otherwise all trips in that set are unserved.
• Pickup service time is pickup_service_time; dropoff service time is dropoff_service_time.
• arrival_time is the service start time after waiting. departure_time must be at least arrival_time + service_time.
• Vehicle load increases by passenger_count at pickup and decreases by passenger_count at dropoff.
• Vehicle load must stay between 0 and vehicle_capacity.
• Dropoff time window: expected_arrival_time - time_window_width <= arrival_time <= expected_arrival_time.
• Pickup time window: expected_arrival_time - time_window_width - direct_travel_time <= arrival_time <= expected_arrival_time - direct_travel_time, where direct_travel_time is the matrix value from the pickup node to the dropoff node.
• Each used route starts at node 0 at an hourly shift start from 300 through 840 inclusive.
• Each used route ends at node 2n+1 no later than 1320.
• Each vehicle shift is 8 hours, so a route must end within 480 minutes of its start time.
• Routes must not use negative travel-time arcs.

Objective

Maximize the number of served trips; higher is better. The verifier recomputes arrivals, departures, loads, served trips, route metrics, and feasibility from the submitted node sequences.

Output Schema

Write /root/report.json as a JSON object with this structure. Placeholder values below show the required shape only.

{
  "routes": [
    {
      "vehicle_id": "V0",
      "start_time": 300,
      "node_sequence": [0, 1, 516, 1031]
    }
  ]
}

Each route has a vehicle_id string, a start_time integer in minutes after midnight, and a node_sequence list of integer node IDs in visit order. The first node must be 0; the last node must be 2n+1. Omit unused vehicles.

Exact route sequences are not unique. Solutions are judged by feasibility and served trip count.