We pulled five years of daily weather for the capital of every country we rate and checked whether our best/avoid months match the actual sky. Verdict: 92% agreement, one coherent cluster of dissent, one rating we now think is wrong, and one lesson about capitals.
Our when-to-visit guides rate every month in 199 countries as best, ok, or avoid: 2,388 ratings. Those ratings drive the value calendar, the country pages, and a fair number of people's actual bookings, so this week we did the thing you're supposed to do with a model: tested it against ground truth.
The ground truth is the Open-Meteo historical archive, which serves the ERA5 reanalysis. We pulled daily rainfall and temperatures for the capital of all 199 countries, every day from January 2021 through December 2025. That is 363,374 country-days, collapsed into a monthly climate profile per country, scored for discomfort (rain days, downpour days, heat above 33°C, cold below 15°C), and compared against our ratings.
The sky agrees with us 92% of the time. The other 8% is the interesting part.
The score
In 144 countries the ratings include both a "best" and an "avoid" month, which makes a testable claim: avoid months should have measurably worse weather. In 132 of the 144 they do. The margin is widest exactly where getting it wrong hurts most: Myanmar, Nepal, Costa Rica, Sierra Leone and Laos top the agreement list, all monsoon countries where the gap between a January and an August is not a preference but a flooded trail. Conakry, Guinea, our single wettest capital-month, gets 853mm of rain in an average July — more than London receives in an entire year (745mm, same dataset). We rate Guinea's July "avoid". Good.
The dissent is one region wearing twelve flags
The twelve countries where the thermometer sides against us: Algeria, Morocco, Palestine, Malta, Israel, Vatican City, Italy, Spain, Libya, Uzbekistan, Tunisia, Jordan. That is not a random scatter; it is the Mediterranean and its neighbours, and the pattern is identical everywhere. We rate their July–August "avoid" and their spring/autumn "best". The raw weather numbers prefer their summer: Madrid in August has 1.8 rain days, April has 8. Rome in August is drier than Rome in May by five rain days.
What the reanalysis cannot see is that Madrid's August afternoons average a 34.6°C daily maximum, that Rome's queues triple, and that August is when hotel prices spike across southern Europe. Our avoid ratings there were never meteorology; they are heat-plus-crowds-plus-price calls, and we stand by every one of them. A dry 35°C afternoon in a two-hour Vatican queue is not "good weather" in any sense a traveler cares about.
One number makes the case: August is simultaneously the driest month in half these capitals and the most avoid-rated month in our entire dataset. Dry is not the same as pleasant.
The capital is not the country
The audit's sharpest lesson was about weather stations, not ratings. Peru shows it best: we rate Peruvian summer (December–February) "avoid" (wet), yet Lima in January records two rain days and 24°C. Both facts are true. Lima is a coastal desert; the rating exists to keep you off the Inca Trail in flood season, 570km away and 3,400m higher. Same story in Morocco, where Rabat's August is a mild 27°C while Marrakech cooks; in Eritrea, where Asmara sits at 2,300m and June is a pleasant 25°C while the Danakil lowlands are among the hottest places on Earth; in Sri Lanka, where Colombo drowns through a "best"-rated August (28 rain days) because the season belongs to the east coast, two monsoons away.
Any country-level weather advice — ours, a guidebook's, an airline magazine's — has a resolution floor. This is why the individual guides carry region notes, and why the ratings are the start of the answer, not the end of it.
What we actually got wrong
Two findings survive every excuse above.
Bhutan's May is rated "best" for the rhododendron bloom. Thimphu's May, per five years of data: 29.6 rain days and 352mm, monsoon numbers in all but name. The bloom is real but it peaks March–April; May is the door closing. That rating is queued for demotion in the September review pass, and per our re-verification policy we'd rather print the correction than defend the error.
Second, the Sahel. South Sudan's February is rated "best" at a mean daily maximum of 38.6°C; twenty of our "best" months across the Sahara–Sahel belt run 35°C or hotter. They are still the right months; they are the dry, marginally cooler ones — but "best" there means "least punishing", and the glossary should say so more plainly than it does.
Assorted findings from 363,374 days
Baghdad's July averages a 45.4°C daily maximum, the hottest capital-month in the dataset, with zero rain. Palau's four "best"-rated months still carry about 28 rain days each — in the equatorial Pacific, rain is not a season, it is a baseline you learn to swim in. And the five capitals where our best months most outperform our avoid months are all in Asia's monsoon belt, which is statistically the region where checking a when-to-visit guide pays best.
Method
ERA5 reanalysis via the Open-Meteo archive API, daily precipitation sum and 2m max/min temperature, 2021-01-01 to 2025-12-31, one series per capital (capitals geocoded via Open-Meteo, four fixed by hand). Monthly profiles are five-year averages. Discomfort score: rain days ×0.6, 10mm+ days ×0.8, each °C of mean max above 33 ×1.2, each °C below 15 ×1.0 — crude, but symmetric, and it can't see crowds or prices, which is precisely what makes the Mediterranean dissent informative. Caveats we accept: capitals under-represent tourist geography (see Peru), reanalysis smooths microclimates, and the window includes the 2023–24 El Niño. The ratings under audit are the best-months dataset (v2026-05-04). Bhutan aside, they survived contact with the sky better than we expected when we started pulling the data.