How we verify Travel AI Rescue
⚙️ Study-design preview — the questions and scoring key are published first. Runs: September 10–17, 2026; scoring: September 18–25; results go live on Monday, September 28, 2026 at 11:00 JST. Until then the table stays empty (this page is not indexed yet).
We publish our verification method, not comparative claims. We do not name other services — only journalists who re-run the test do. Everything below is designed to be reproduced independently.
1. Pre-registered protocol
We fix the test design before running it, using a dual-subject design:
- Inbound travelers to Japan — prompts in English, Chinese and Korean.
- Japanese travelers abroad — prompts in Japanese.
Pre-registration date: September 8, 2026 (fixed and published before scoring begins).
Question set (30 × 2)
- Set A — inbound travelers (English / Chinese / Korean): 30 emergencies inside Japan — theft, sudden illness, lost passport, earthquake alerts, an eSIM that will not connect.
- Set B — Japanese travelers abroad (Japanese): 25 emergencies abroad — theft, illness, embassies, local emergency numbers — plus 5 connectivity questions (emergency numbers, roaming / APN settings, what to do with no signal).
- Every question is tied in advance to at least one official published source (URL) that defines the correct answer.
Scoring axes (3 levels each)
- ① Factual accuracy (phone numbers, addresses and agency names exist and are correct)
- ② Procedural accuracy (order and requirements)
- ③ Freshness (no discontinued schemes or old numbers)
- ④ Availability (does it assume a connection?)
Model names, versions and run dates are published together with the results. Our own product is scored on the same questions, and the items where it loses are published as they are.
2. Method — machine scoring
Answers are scored by agreement with official published information (embassies, police and other authorities), not by subjective judgment. The scoring key is defined in advance and applied identically to every answer, so the evaluation cannot favor any single service.
3. Results
Other services are shown anonymously as “major general-purpose chat AIs (as of September 2026, Models A / B / C)”. Only our own product is named. (Results will be published on September 28, 2026 at 11:00 JST.)
| Scenario | Model A | Model B | Model C | Travel AI Rescue |
|---|---|---|---|---|
| Passport lost — right first-response order | — | — | — | — |
| Traffic accident — emergency numbers & record | — | — | — | — |
| Scam / overcharge — what not to do | — | — | — | — |
| Sudden illness — nearest care & insurance | — | — | — | — |
| Card lost — freeze & fraud dispute | — | — | — | — |
4. Reproduction package
We publish the full prompts, the scoring key and the run scripts so anyone can reproduce the test:
GitHub link — to be published.
5. 30-minute replication guide for journalists
- Request a free review account: email info@jagproject.com (issued the same day).
- Sign in at https://panam.travelsim-japan.com/my/.
- Run the pre-registered prompts from the reproduction package and score them against the published key.
- Use synthetic data (fictional people, orders and itineraries); do not enter real personal information.
6. Disclaimer
These are results confirmed under pre-specified conditions and do not guarantee the same results in all environments. AI output is a draft; in a real emergency, always prioritize local police, ambulance services, medical care and your embassy.