· TRAVEL KEEPERS
Methodology

How the Keeper Score works.

The Keeper Score turns three government travel advisories into one of four ratings: Walk freely, Generally safe, Know before you go, or Plan carefully. We name your options, not your fears.

How the score works

Six dimensions, one composite, four ratings.

A single “is it safe” number flattens too much. Walking the market at noon is not the same question as catching a cab at midnight. So we score every place across six dimensions. Some sit on government frameworks; some are our editorial calls for trip planning. Tap any dimension below to see its rubric: what counts as safe / caution / avoid, where it's anchored, and the prescriptive language we use on briefing cards.

For each dimension, an AI extraction step (Claude Sonnet) reads the three government advisories and assigns a tier — “safe,” “caution,” or “avoid.” Tiers map to point values (95 / 75 / 50), and the Keeper Score is the average of the six numbers. That average translates into one of four plain-language ratings:

  • Walk freely (90+)Few or no flags from government advisories. Most travelers move freely day and night.
  • Generally safe (75–89)A handful of caution flags. Standard precautions are usually enough.
  • Know before you go (50–74)Several advisories flag caution. Worth reading the full briefing before you book.
  • Plan carefully (Below 50)Multiple serious flags. Plan deliberately — neighborhoods, transit, timing all matter.
Choices

Choices we made along the way.

  • Tier assignment is AI-mediated.The AI follows our rubric; it does not invent ratings, and it does not introduce facts that aren't in the source. But the rubric itself, what counts as “safe” vs. “caution” for each dimension, is ours. Tap any dimension card above to read it.
  • The points scale (95 / 75 / 50) is a Travel Keepers operational choice, not derived from external instrument calibration.
  • The six dimensions are equally weighted. Future work will test whether literature-derived weighting changes city rankings.
  • When the AI has no signal for a dimension, the rubric instructs “default to caution.” This introduces a small downward bias on under-documented cities. Disclosed.
  • The prescriptive language on every briefing card — “Use apps,” “Vetted only,” “Food/water care,” “Stay alert” — is Travel Keepers editorial language. It translates the underlying tier into action language specific to the dimension. The full lookup table is shown on each dimension card above.
What we don't measure

Three things this score does not cover.

  1. Harassment and sociocultural risk.Government advisories don't formally name harassment, sexual or otherwise — and currently, neither do we. Planned for v1.x via community-reported data, where Keepers report what advisories don't surface.
  2. Digital safety and cyberstalking. Data breaches, identity theft, cyberstalking, and digital harassment are documented as a distinct risk dimension in current solo-women research. Not in Travel Keepers v1.
  3. Within-city variance by event, season, or day of week. A city's score is a single number; real safety varies by what's happening when you arrive. Briefings include neighborhood-level day/night detail, but the headline score does not.
Limitations

The honest caveats.

Six things to know about our own score, in plain language.

  • Who the score is for vs. who wrote the source data. The score reflects advisory-reported risk written for a generic, implicitly-male traveler audience, applied with a solo-women lens. Government advisory writers are not the population the score is for. The lens is editorial, not empirical.
  • Why three advisories aren't really three independent sources. The US, UK, and Canadian advisories share liability-driven over-warning patterns and a shared blind spot for harassment. Their agreement is not statistical triangulation.
  • AI extraction reliability is not yet measured. The same advisory text run through the AI twice may produce slightly different per-dimension ratings. We have not yet quantified this. Planned for v1.x.
  • Solo dimension (already corrected). We tested scoring solo travel as a separate dimension and found government advisories rate every destination uniformly poorly on this measure — the dimension was potentially indexing source bias, not destination variance. Removed from the composite. Solo safety is a candidate for community-reported data in v1.x.
  • Prescriptive language is editorial.The actionable phrases on every briefing card are Travel Keepers writing, not direct advisory quotes. Some encode infrastructure assumptions that may not hold in every city (e.g., “Use apps” assumes ride-share is available and appropriate).
  • Source reporting biases.UNODC sexual-violence rates have severe underreporting bias (UNODC-acknowledged). Government advisories systematically over-warn. We've baked these biases into our composite — we have not corrected for them.
Sources

Everything we cite, with links.

Primary data sources (drive the score)

Authority frameworks (cited for face validity)

  • CDC Yellow Book — US CDC Health Information for International Travel; structures the health dimension and traveler-perception-of-risk anchor.
  • WHO International Travel and Health — vaccination requirements and disease-risk anchor for the health dimension.
  • UNODC Statistics — country-level homicide, theft, and violent-crime data; cited for petty-crime face validity, with the reporting-bias caveats noted in Limitations.
  • OSAC Country Security Reports — US State Department / Diplomatic Security industry-facing security reports. Report content is largely member-gated; cited institutionally.
Research that informed this thinking

Current academic literature (2024–2025)

  • Abdul Shukor, S., & Kattiyapornpong, U. (2024). Solo female travelers: a systematic literature review and future research agenda. Consumer Behavior in Tourism and Hospitality, 19(3), 366–382. doi.org/10.1108/CBTH-08-2023-0125 — PRISMA review of N=25 empirical studies; identifies risk perception as the second most-studied topic in solo-female-traveler research and names gendered risk (including sexual harassment) as dominant.
  • Dubey, S. (2025). Women's perceived safety in public places and public transport: A narrative review of contributing factors and measurement methods. Cities, 156. doi.org/10.1016/j.cities.2024.105534
  • Ghaderi, Z., Bagheri, F., Esfehani, M., Beal, L., & Houanti, L. (2025). Exploring Cybersecurity Threats to Solo Female Travelers. Journal of Travel Research. doi.org/10.1177/00472875251361466
  • Maiurro, A., & Brandão, F. (2025). Motivations, Needs, and Perceived Risks of Middle-Aged and Senior Solo Travelling Women: A Study of Brazilian Female Travellers. Journal of Population Ageing. doi.org/10.1007/s12062-024-09450-z

Foundational literature (still cited in current work)

  • Wilson, E., & Little, D. E. (2008). The solo female travel experience: Exploring the ‘geography of women's fear.’ Current Issues in Tourism, 11(2), 167–186.
  • Yang, E. C. L., Khoo-Lattimore, C., & Arcodia, C. (2017). A systematic literature review of risk and gender research in tourism. Tourism Management, 58, 89–100.
  • Yang, E. C. L., Khoo-Lattimore, C., & Arcodia, C. (2018a). Constructing space and self through risk taking: A case of Asian solo female travellers. Journal of Travel Research, 57(2), 260–272.
  • Yang, E. C. L., Khoo-Lattimore, C., & Arcodia, C. (2018b). Power and empowerment: How Asian solo female travellers perceive and negotiate risks. Tourism Management, 68, 32–45. (Maps the harassment risk taxonomy for Asian solo female travelers — a direct anchor for the harassment-dimension gap.)
What's next

The score will get more sound over time.

Here's the next layer of work, in sequence.

  • v1.x — Reliability test. Running the extraction multiple times to measure how consistently the AI applies our rubric. Publishing the result.
  • v1.x — Keeper community data.Real solo women travelers reporting back on accuracy, plus the dimensions advisories don't surface (harassment, digital safety, sociocultural fit).
  • v2 — Literature-derived weights. Moving from equal-weighted dimensions to weights informed by solo-women-traveler risk research.
Where this is going

V1 today. Keepers next.

The 2027 roadmap brings a second layer: Keepers— real women who've been there — reporting back on whether the briefing matched their experience. That community-data layer is where harassment, digital safety, and other dimensions advisories don't formally name will eventually live.

Until then, this page is the truth of what V1 is: a working method with cited sources and documented limitations.

Want the bigger picture?Read our mission