Open research note · Dataset v2026-07-13
A public sample set for checking AI travel plans
Fifty-four curated itinerary cases make the validation problem inspectable: messy place names, duplicates, closed days, ticket pressure, day trips, and neighborhood fit.
This is not a statistical study.
The rows are hand-curated demonstrations from public sample itineraries. They contain no TripSapien user data, do not compare model accuracy, and do not support product-performance claims.
Page reviewed July 13, 2026 · Dataset snapshot modified July 13, 2026
01 · Contents
Six views of the same planning problem
Each of the nine cities contributes the same six case shapes, making the data predictable to browse without pretending it is a randomized sample.
- Full mixed-source pasteA long list combining AI, friends, articles, social saves, and map links.
- AI day-plan subsetA shorter day-by-day draft focused on extraction, hours, and map grouping.
- Booking-ahead subsetTimed-ticket, reservation, and likely sell-out cases.
- Closure-risk subsetClosed days, seasonal openings, and uncertain-hour cases.
- Day-trip subsetParent/child attractions, distance, and out-of-town grouping.
- Neighborhood subsetA larger set for deduplication and geographic grouping.
02 · Method
What the rows mean
1. Start from public samples
Nine city itineraries were written as realistic mixes of AI drafts, friend notes, article ideas, social saves, and map links.
2. Derive six case shapes
Each source is represented as one full paste and five focused subsets for repeatable demonstrations.
3. Describe checks to exercise
Validation tasks name the intended workflow: extraction, hours, bookings, dedupe, day trips, neighborhoods, and maps.
4. Record expected characteristics
Expected-outcome counts describe the hand-authored case. They are not observed model scores or product accuracy results.
Appropriate uses
- Parser and interface demonstrations
- Schema and integration examples
- Regression-case design
- Travel-validation vocabulary research
Unsupported conclusions
- Assistant or model accuracy rankings
- TripSapien precision or recall estimates
- Traveler-behavior generalizations
- Current venue status without a live check
03 · Download
Inspect or reuse the data
The repository and Hugging Face mirror use the same CC BY 4.0 license. Attribute TripSapien and link to the public repository when you reuse the data.
- JSONL · 54 complete casesFull paste text, validation tasks, expected outcomes, provenance, and license.
- CSV · compact row indexA smaller table for browsing IDs, cities, case shapes, and expected-count fields.
- Dataset schema · field definitionsMachine-readable definitions for the public data fields and expected-outcome object.
04 · Try the workflow
Check a real itinerary
The dataset explains the problem. To work on your own trip, paste the actual draft and set the dates you plan to travel.
Browse every planning tool