How we verify
Every fact on ChillNomad is checked verbatim against an official source, rated on a tier 1–4 scale, and corroborated by a second independent source before it publishes. Numbers are copied digit-for-digit, safety-critical claims are checked twice by separate reviewers, anything unconfirmed is flagged rather than guessed, and each page shows when it was last verified and when it's next due for review.
The source tier system
Not all sources carry equal weight. Every source we cite is registered with a tier, shown on each page next to the claim it backs, so you can see the strength of the evidence — not just that a citation exists.
| Tier | What it is | How we use it |
|---|---|---|
| Tier 1 | Official primary The regulation text itself; the country's agriculture ministry or veterinary authority | The only thing we treat as the basis for a claim. |
| Tier 2 | Official secondary Another government's summary of a country's rules; embassy pages | Usable as a basis, but tier 1 wins any conflict. |
| Tier 3 | Industry Pet-relocation companies, IATA, aggregator sites | Cross-checking and leads only. |
| Tier 4 | Community Forums, personal blogs | Leads only — always taken back to an official source. |
How each claim is checked
- Find the primary source. We start from the authority's current page or the regulation text — original-language versions over English, which are often out of date.
- 2-of-3 corroboration. A claim is only marked verified when at least one tier 1/2 source and one independent source agree. Two aggregators copying the same article do not count as two votes.
- Resolve conflicts in the open. When sources disagree we check effective dates (usually an old-vs-new version issue), defer to the latest tier 1 text, and record the conflict and our reasoning on the page rather than silently picking a side.
- Zero tolerance on numbers. Day counts, fees and thresholds are copied verbatim from the official text — never paraphrased. A build-time check fingerprints every reader-facing number and fails the build if a digit moves.
- Flag the unknown. Anything we can't confirm against a tier 1/2 source is marked pending and shown as unconfirmed. Blank beats a guess.
- Record the metadata. Every record carries its official source URLs, what it was cross-verified with, the last-verified date and the next-review date.
Safety-critical facts get a second reviewer
Some mistakes are worse than others. Getting a breed ban or a titer-test rule wrong can leave a dog detained at the border or an owner turned away. These claims go through an adversarial second pass: a separate reviewer — working from a clean slate, with the explicit goal of disproving the claim — re-opens the official source and re-checks it. The producer of a fact never signs off on their own work.
The claim ledger
Each country page ends with a ledger: every claim on the page, its confirmation status, and the exact source IDs backing it. It's the moat of this site — the difference between "trust us" and "here is precisely what we checked, and what we haven't." Rows we couldn't confirm show as unconfirmed, in the open.
Audit rounds
Beyond per-claim checks, the whole dataset goes through audit rounds where an independent reviewer samples rendered numbers across the built pages and traces each back to its provenance. A number with no backing source is flagged for adjudication — the standard is that a false flag costs one review line, while a missed one costs the moat, so we over-flag on purpose.
Staying current
Rules change. We re-check on a schedule tied to the regulations themselves — for instance, the EU's pet-certificate changes have fixed transition dates that trigger a full review — plus quarterly spot-checks across the dataset. Each page's "last verified" and "next review due" dates tell you exactly where it stands.
What AI does, and doesn't, do
We use AI to speed up first-pass extraction from official pages and to translate non-English source documents. We do not let a model answer a regulatory question from memory — training data mixes old and new versions of the law. Every verified mark means a human read the corresponding passage of the official source.
What does a 'verified' badge actually mean?
That the specific number or rule was read, verbatim, in the official source cited next to it — and corroborated by a second independent source. It is not an AI summary; a person checked the source text.
Why do some fields say 'pending official confirmation'?
Because we could not confirm that detail against a tier 1 or tier 2 source. Rather than guess or repeat an aggregator, we show it as unconfirmed until we can verify it. Blank beats a guess.
How many countries and sources does this cover?
15 countries so far, each with its own claim ledger mapping facts to registered tier 1–4 sources, plus the cities and airlines linked from them. Coverage expands country by country, not by mass generation.