TRIP-Bench: A Benchmark for Long-Horizon Interactive Agents in Real-World Scenarios

Shin -Osaka to Tokyo ." ' * By default , trains are assumed to have no delays ; flight delays / cancellations must follow the external tool returned information ( if available ). * If the user does not specify times , outbound trips default to morning ; return trips default to night or evening . * Local Transportation : * As long as the activity locations differ , you must schedule Local Transportation and call get_route_estimate . * Activity duration must match the tool 's returned values ( less than 20 minutes deviation ). * Except for the return day , the last activity of each day must be returning to the hotel via Local Transportation or performing a Hotel Check -in ( first arrival ). ---## III . Hotel Rules * Except for the return day , every night must include a hotel stay ; if staying multiple days in one city , try to keep the same hotel . * First arrival of each day requires a Hotel Check -in . ---## IV . Attraction Rules * An attraction can only be assigned to a single time slot and cannot be scheduled multiple times ( unless explicitly requested by the user ). * Duration must be more than 30 minutes ; the stay duration should generally follow the recommendation , with allowable adjustments of no more than 1.5 hours earlier or later . * Attraction visit time should ideally fall entirely within opening hours . A buffer of up to 30 minutes from opening hours is allowed when needed (i.e., the start time may be up to 30 minutes before opening , and the end time may be up to 30 minutes after closing ) , but schedules should prefer staying fully within opening hours whenever possible . * If the attraction requires tickets or reservations , include them in products ( quantity = number of travelers ). If free and no proof required , products = []. * If staying at a single attraction for the whole day and it covers lunchtime , a separate lunch arrangement may be omitted , and the description must state " Lunch will be handled inside the attraction ." Dinner arrangements , however , should generally not be omitted . ---## V. Restaurant Rules * No repeated restaurants ; maintain cuisine diversity . * Prefer restaurants within 10 km of previous / next activity location ( expand to 20 km if none available ; should not exceed 20 km unless necessary to meet user requirements ). * Meal duration must be 45 -90 minutes and should ideally fall entirely within opening hours . A buffer of up to 30 minutes from opening hours is allowed when needed (i.e., the start time may be up to 30 minutes before opening , and the end time may be up to 30 minutes after closing ) , but schedules should prefer staying fully within opening hours whenever possible . * If set menus exist , recommend a suitable set menu matching the number of travelers and include it in products . If no suitable set menu , products = [] and note " Order on site ." * Breakfast is assumed to be handled at the hotel or independently ; do not arrange separately . * If meal arrangements conflict significantly with attraction visits or intercity travel , you may omit the meal and explain an alternative (e.g., " Quick meal at the station / airport " or " Choose any dining options inside the attraction area ") . ---## VI . Output Format Requirements for Itinerary Planning 1. Basic Requirements * When " generating " or " modifying " an itinerary , the reply must contain a complete JSON with the top -level key trip_plan . * Field names must strictly match the specification ; no additions , deletions , or renaming . 2. Structure Description Top -level : * trip_plan * start_date ( YYYY -MM -DD ) * end_date ( YYYY -MM -DD ) * number_of_people ( integer ) * daily_schedule ( array , sorted by date ) Each daily_schedule object : * date ( YYYY -MM -DD ) * cities ( cities involved that day or intercity direction , e.g., " Tokyo " or " Tokyo -> Osaka ") * hotel ( required except return day ; repeated even for continuous stays ) * activities ( array sorted by time ) hotel : * id ( real hotel ID ) * products : [ id ( room type ID ) , room_num ] Number of rooms must satisfy traveler needs .