Report · Methodology and findings · 2026

What we can evidence about visitor-stay rules, and what we cannot

A methodology and findings report on how long visitors may stay, built only from the routes in our own rulebook that carry a government source for the claim being made. Of the 1,555 passport-to-destination routes we generate a page for, 228 carry no length of stay at all and 35 carry no official government source. This report says what those gaps are, where the counting models actually differ, and how to re-derive every figure in it.

By David Toniolo, Do a Stretch · Updated 2026-08-23 · Prices and facts checked 2026-08-23 · 19 min read

What this report measures

Start with what we cannot tell you, because that is the part most reports leave out and it is the part that decides how much the rest is worth.

We generate a rules page for 1,555 passport-to-destination routes. Three things are true of that set at once:

228 of 1,555 routes carry no length of stay. A visa sets it, or no source we trust records one, or the stay is positively recorded as unlimited, which is true of 1. For those routes we can tell you that you may go, and not how long you may remain.
35 of 1,555 routes carry no official government source for the entry rule, and 35 carry none for the length of stay. Whatever those counts read on the day you open this page, those routes are excluded from every finding below rather than hedged into it. The counts move fast — this corpus is under active sourcing and they have shifted by hundreds of routes inside a single day — which is why the build is cited alongside the date.
Coverage is not uniform across passports. The per-passport table below runs from complete coverage down to well short of it, and an average across the corpus would hide which readers it is worst for.

What is left after those exclusions is the corpus this report is built from: 1,520 of 1,555 routes carrying an official government source for both the entry rule and the length of stay.

Everything here describes the Do a Stretch rulebook as generated from build 2026-08-30. It is not a census of global visa policy and should not be cited as one.

Coverage and exclusions

The population. One projection produces every route in this report — the same function the site calls before it writes a page. It covers 22 passports against the destinations those passports travel to, which is a launch catalogue rather than the world.

Three classes of route, counted separately. Of the 1,555 routes, 1,160 carry a positive recorded day count, 167 are treaty free-movement rights with no day count to record, and 228 carry no day number for one of the reasons above. Those three classes sum to the whole and none is obtained by subtracting the others.

That is not pedantry. Deriving "routes with a day count" by removing free-movement routes from the total is how this project once promised a day count for the 228 routes that have none, and how a guide came to publish a denominator no projection here produces.

Evidence coverage, stated separately for each claim. A route makes two separable claims — you may enter without a visa, and you may stay this long — and a source proves one of them, not both. We count them apart:

The claim	Rule the source has to cover	With an official source	Without one	Of
You may enter without a visa	the entry rule for this route	1,520	35	1,555
You may stay this long	the rule governing duration, which may be a zone rule	1,520	35	1,555
Both, on the same route	both of the above	1,520	35	1,555

They can differ, and the machinery exists so that we would notice if they did. A stay allowance can come from a shared zone rule while the entry permission is recorded on the country's own row, and each carries its own sources; measuring them as one number would report the average of two facts and describe neither.

A trust label is not a source. Our bank also carries a review label on each entry rule. It records our own verification state rather than the tier of the evidence, and the two do not move together:

Recorded access-trust label	Routes	Of those, with an official source
verified	1,228	1,228
free-movement zone rule	167	167
reported	131	125
conflict	21	0
single-source	8	0

Read that table in both directions.

0 of 1,555 routes are labelled verified while carrying no official-tier source. Those pages are generated and then withheld from search until the contradiction is resolved, because a page saying "verified" with no government link behind it is the failure this report exists to be honest about.

Running the other way, 125 routes carry an official government source while our label still reads something short of verified — the evidence landed ahead of the review. Which of those two numbers is larger changes week to week as sources arrive.

A single "percentage verified" is therefore wrong in both directions at once, which is why this page publishes the cross-tab and no headline percentage.

Coverage by passport, because an average hides who it is worst for:

Passport	Published routes	Official source for entry	Official source for duration
US passport	71	69	69
Canadian passport	71	70	70
New Zealand passport	66	65	65
UK passport	71	70	70
Singapore passport	71	70	70
Australian passport	71	70	70
Brazilian passport	71	68	68
French passport	71	70	70
German passport	71	70	70
Japanese passport	71	70	70
South Korean passport	71	70	70
Dutch passport	71	70	70
Irish passport	71	70	70
Mexican passport	70	68	68
Portuguese passport	71	70	70
South African passport	70	68	68
Spanish passport	71	70	70
Italian passport	71	70	70
Swedish passport	72	71	71
Swiss passport	71	70	70
Indian passport	70	67	67
Philippine passport	71	64	64

What we exclude, and why. Routes failing the relevant source test are excluded from the findings rather than counted as zeros or hedged in prose. A route with no recorded stay length is not a route with a short one. Free-movement rights are excluded from every duration statistic, because a treaty right of residence has no allowance to measure; they are counted as their own class instead.

Method

One projection, no second walk. Every number here comes from the two functions that write the site's own pages. We have made the opposite mistake and paid for it: a hand-written second walk of the same data silently omitted a destination the projection already handled, and the figures taken from it were published as a description of our corpus. There is now a test that refuses to let a second walker exist.

Predicates that say what they test. Has an official source and is labelled verified are different predicates over different fields. So are treaty free-movement right and shape is unlimited — one route in this corpus sits between those two, an entry waiver with no time limit and no treaty behind it, and counting free movement by shape would sweep it in.

Evidence must cover the claim it is attached to. A government page establishing visa exemption is evidence for access. A government page publishing a maximum stay is evidence for duration. Attaching one and calling the row sourced is how a corpus comes to look better evidenced than it is, and it is why the two coverage figures above are derived from two different rules rather than from one flag.

Every count states its denominator — in the sentence, not a footnote. And a maximum is not a typical case: where a destination's routes disagree, this report carries the spread and the number of routes that actually reach the maximum, because "up to 365 days" can mean every route or one of them.

Dates. Every figure on this page is substituted from the build that produced it, not typed, so a number here and a number on a rules page cannot disagree. The official sources behind these routes were retrieved between 2026-08-16 and 2026-08-26.

Findings

Rolling windows are not a European peculiarity

The traveller's heuristic is that Schengen counts on a rolling window and everywhere else counts per entry. It fails, and it fails in the direction that produces an overstay rather than a wasted trip.

Among the 1,153 routes whose duration carries an official government source, 322 use a rolling window and 831 use a per-entry maximum. The rolling ones span 30 destinations, of which 13 are outside the Schengen area entirely: Albania, Barbados, Brazil, Cyprus, Ecuador, India, Kiribati, Montenegro, Peru, Serbia, South Korea, Türkiye, the United Arab Emirates.

Some are European and some are nowhere near Europe. The counting shape does not follow the map, which is what makes the heuristic worse than no heuristic at all. What a rolling window does to a border run is worked through in detail here.

The window is not always 180 days

Across the 322 officially-evidenced rolling routes, the window lengths are 90 in 180 (273 routes), 90 in 365 (26 routes), 180 in 365 (23 routes):

Window	How it reads	Routes
90/180	90 days in any rolling 180	273
90/365	90 days in any rolling 365	26
180/365	180 days in any rolling 365	23

"90 in 180" is the shape people know. A 180-day allowance inside a rolling 365 is twice the allowance over twice the window, which is not the same rule and does not behave the same way when you plan a year around it. A 90-day allowance inside a rolling 365 is stricter than either. Reading the allowance without the window is how a traveller books a year that does not fit.

Identical headline numbers, opposite behaviour

Georgia publishes a 365-day visa-free stay; India publishes 180 days. Georgia's is counted per entry, so leaving and returning opens a fresh allowance. India's runs across a rolling 365-day window, so it does not.

Two numbers of the same order, two completely different planning consequences. A table of maximum stays that omits the counting shape is not a simplification of that fact; it is the loss of it.

A maximum is not a typical case, and the evidence test proves it

The longest officially-evidenced allowance in our set is 365 days, reached by Georgia on 20 of its 20 routes. Now read the same ranking without the evidence test applied, where the top is reached by Albania on 1 of its 22 routes, Georgia on 20 of its 21 routes.

The difference between those two lines is the whole argument. A destination can head a table of maximum stays on the strength of a single route, while every officially-sourced route into it records a shorter allowance under a different counting shape — and "up to 365 days" is true of both cases and useful in neither.

So the honest presentation needs four columns rather than two: the maximum, the spread, how many routes reach the maximum, and the denominator.

Destination	Longest officially-sourced stay	Across those routes	Routes	At that length
Georgia	365 days	every route agrees	20	20
Peru	183 days	90–183 days, depending on passport	19	1
Antigua & Barbuda	180 days	every route agrees	15	15
Barbados	180 days	28–180 days, depending on passport	16	9
Canada	180 days	every route agrees	16	16
Costa Rica	180 days	30–180 days, depending on passport	21	20
Dominica	180 days	21–180 days, depending on passport	17	14
Ecuador	180 days	90–180 days, depending on passport	19	1
Hong Kong	180 days	14–180 days, depending on passport	22	1
India	180 days	every route agrees	21	21
Jamaica	180 days	30–180 days, depending on passport	19	6
Mexico	180 days	every route agrees	17	17
New Zealand	180 days	90–180 days, depending on passport	17	1
Panama	180 days	90–180 days, depending on passport	20	2
South Korea	180 days	30–180 days, depending on passport	19	1
the United Kingdom	180 days	every route agrees	17	17
the United States	180 days	90–180 days, depending on passport	16	1
Albania	90 days	every route agrees	21	21
Argentina	90 days	every route agrees	20	20
Australia	90 days	every route agrees	15	15
Austria	90 days	every route agrees	10	10
Brazil	90 days	30–90 days, depending on passport	20	19
Brunei	90 days	14–90 days, depending on passport	18	12
Bulgaria	90 days	every route agrees	10	10
Chile	90 days	30–90 days, depending on passport	20	19
Colombia	90 days	every route agrees	20	20
Croatia	90 days	every route agrees	10	10
Cyprus	90 days	every route agrees	10	10
Czechia	90 days	every route agrees	10	10
Estonia	90 days	every route agrees	10	10
France	90 days	every route agrees	10	10
Germany	90 days	every route agrees	10	10
Greece	90 days	every route agrees	10	10
Guatemala	90 days	every route agrees	20	20
Hungary	90 days	every route agrees	10	10
Ireland	90 days	every route agrees	9	9
Italy	90 days	every route agrees	10	10
Japan	90 days	every route agrees	20	20
Kenya	90 days	every route agrees	21	21
Kiribati	90 days	every route agrees	8	8
Lesotho	90 days	14–90 days, depending on passport	13	2
Malaysia	90 days	30–90 days, depending on passport	12	9
Malta	90 days	every route agrees	10	10
Montenegro	90 days	every route agrees	19	19
Morocco	90 days	every route agrees	20	20
Nepal	90 days	every route agrees	20	20
the Netherlands	90 days	every route agrees	10	10
Poland	90 days	every route agrees	10	10
Portugal	90 days	every route agrees	10	10
Romania	90 days	every route agrees	10	10
Rwanda	90 days	30–90 days, depending on passport	22	2
Serbia	90 days	every route agrees	19	19
Singapore	90 days	30–90 days, depending on passport	4	3
South Africa	90 days	30–90 days, depending on passport	18	17
Spain	90 days	every route agrees	10	10
Switzerland	90 days	every route agrees	10	10
Taiwan	90 days	30–90 days, depending on passport	18	16
Tanzania	90 days	every route agrees	21	21
Thailand	90 days	60–90 days, depending on passport	22	2
Türkiye	90 days	30–90 days, depending on passport	20	18
the United Arab Emirates	90 days	30–90 days, depending on passport	11	9
Uruguay	90 days	every route agrees	20	20
Vietnam	90 days	45–90 days, depending on passport	21	11
Sri Lanka	60 days	30–60 days, depending on passport	22	2
the Philippines	59 days	14–59 days, depending on passport	21	1
Cambodia	30 days	every route agrees	21	21
the Dominican Republic	30 days	every route agrees	20	20
Egypt	30 days	every route agrees	21	21
Indonesia	30 days	every route agrees	22	22
Laos	30 days	15–30 days, depending on passport	21	19
Maldives	30 days	every route agrees	22	22

71 destinations carry an officially-sourced day count for at least one of our routes. Where a row says every route agrees, every route agrees; where it does not, the spread is in the row rather than flattened into the headline. The unrestricted version of this ranking is the published maximums in full.

Visa information is organised around a handful of instruments

This is the structural fact that surprised us most, and it is the reason the coverage figures above moved as fast as they did.

170 of our routes take their stay allowance from a single shared zone rule — the Schengen 90-days-in-any-180 window — across 17 destinations and 10 passports, and that rule's recorded source is one EU regulation. Read the publisher distribution below and the same shape appears again: 77 distinct government hosts carry the whole corpus, and the row at the top of it is a single instrument rather than a national portal.

Publisher	Routes it evidences
eur-lex.europa.eu	394
gov.uk	28
cancilleria.gov.co	22
clang.cancilleria.gob.ar	22
dominica.gov.dm	22
embassyofpanama.org	22
eta.gov.lk	22
foreign.gov.bb	22
gub.uy	22
igm.gob.gt	22
image.mfa.go.th	22
imigrasi.go.id	22

One research pass against an instrument of that reach corrects many destinations at once. The same concentration means one stale page degrades many at once, and a report mentioning only the first half of that would be selling rather than measuring.

Where the counting model is an approximation, we say so

21 routes carry a stay that the authority states per calendar year and that our evaluator stores as a rolling 365-day window, across these destinations: India.

That approximation is deliberately stricter than the rule at a year boundary — it can warn earlier than the authority requires, never later. Nothing in the shape of the stored rule distinguishes it from a window the source really does state as rolling, so the count comes from a sentence written onto the rule at sign-off rather than from the shape. Each of those pages says so. We would rather publish the number of routes where our model is an approximation than let it disappear into a table of exact-looking figures.

Limitations

The catalogue is not the world. 22 passports and the destinations they travel to, chosen for a travel-tracking product. Absence from this report is not evidence about a route we do not publish.
Coverage is a statement about our evidence, not about the law. A route without an official source is not a route with no rule. It is a route we have not sourced to the standard we publish at, and many are very likely correct.
Recency is not correctness. An official source retrieved this month is evidence that a page said something on the day we read it. Rules change between readings, and the authority at the border decides.
We do not publish the underlying rows, and that is a licensing constraint rather than a preference. The bulk of the compilation derives from Wikipedia's visa-requirement tables, which are CC BY-SA. Share-alike is copyleft: republishing the compilation would oblige us to license the result the same way. Derived aggregate analysis — what this report is — is our own work and is not encumbered. So you get counts, distributions and method, and no row export.
A generated page is not a live page. Every count here describes what this build generates, including pages deliberately withheld from search. It is not a measurement of what any search engine has indexed.
Review labels lag evidence. The cross-tab above is published because the label and the source disagree in both directions. We would rather show the disagreement than average it away.
This is a snapshot, and the corpus moves quickly. Rulebook build 2026-08-30. A separate workstream is actively sourcing rules, and figures on this page have moved by hundreds of routes inside a single day. Cite the build, not just the date.

Reproducing these numbers

Every figure above is substituted at build time from the generator's own fact pack. Two committed tools re-derive them from the same functions:

node docs/organic-growth-program/tools/derive-authority-report.mjs
node docs/organic-growth-program/tools/derive-authority-report.mjs --receipts <dir>

The second writes a count-receipt/v1 record for each headline figure: the population, the predicate, the denominator, the sibling classes, a membership digest, the repository commit and the bank snapshot identity. The receipt is the unit of evidence, not the sentence.

There is an honest gap in this offer. The tools are public; the rules bank they read is not distributable, for the licensing reason above. A reader outside the project can audit the derivation — what was counted, what was excluded, whether each partition closes — but cannot re-run it against our data. We would rather say that than imply an independence the licence does not allow.

The procedure the derivation follows is committed too, and it exists because we got twelve numbers wrong in a single day: docs/organic-growth-program/COUNTING_DISCIPLINE.md.

Corrections

If a figure here is wrong, we want to know, and we would rather correct it loudly than quietly. Write to hello@doastretch.com with the figure, the page and what you think it should be.

Corrections to a rule change the rulebook, which changes this page at the next build. Corrections to the method are recorded in the programme documentation with the date and what was observed. A citable wrong number outlives its correction, so the correction has to be findable from the same page as the number.

Frequently asked questions

How much of your rulebook carries an official government source?

1,520 of our 1,555 generated routes carry an official government source for the entry permission, and 1,520 carry one for the length of stay. Those are two different claims about the same route, counted separately and against the rule that governs each; 1,520 routes carry a source for both.

Why do you publish routes you have not officially sourced?

We publish them because withholding a rule a traveller needs is not obviously safer than publishing it with its evidence labelled. Every page states its evidence tier, the date it was checked and whether an official source exists, and pages carrying an internal contradiction are withheld from search. 35 of 1,555 routes carry no official access source today, and each of those pages says so in plain words.

Is "verified" the same as "officially sourced"?

No — "verified" is our review label and "officially sourced" is a property of the evidence, and this report keeps them apart on purpose. 0 of 1,555 routes are labelled verified while carrying no official-tier source, and 125 carry a government source while still labelled something short of verified. That is why we publish the cross-tab instead of a headline percentage.

Does a border run reset a rolling window?

No — under a rolling window, leaving and returning does not clear the count, and it is the single most expensive misunderstanding in visitor-stay rules. Days leave the window only by ageing out of it. 322 of our 1,153 officially-evidenced counted routes work this way, across 30 destinations.

Can I cite these numbers?

Yes — cite the figure, this page and rulebook build 2026-08-30, and note that it describes the Do a Stretch rulebook rather than global visa policy. The corpus moves, so the build matters as much as the date. The aggregate analysis is ours to share; the underlying compilation is not, for the licensing reason set out under Limitations.

Sources

Regulation (EU) 2018/1806 — the visa-exemption instrument recorded against the shared Schengen rule
Directive 2004/38/EC — free movement of EU/EEA citizens, recorded against our treaty-right routes
Common Travel Area guidance — the UK and Ireland arrangement
Special Category visa (subclass 444) — the Trans-Tasman arrangement
Legislative Herald of Georgia — recorded against Georgia's 365-day allowance
Indian visa portal — recorded against India's 180-day rolling allowance

Rulebook build 2026-08-30 (generated 2026-08-30T14:25:39.441Z). Derived through the site's own projectRoutes() and countingFacts(); every figure on this page is substituted from that pack at build time rather than typed.

Count it, don't carry it

Do a stretch. Don't snap.

The counter, the alerts and your documents are free, forever. Coming to the App Store and Google Play — one email when it lands.