SnowVerdict answers one question for eighteen Alpine ski resorts: will there be snow when I am there? Not a forecast widget and not an opinion, but a dated verdict backed by fifteen winters of measured weather history, with the method shown on the page. We designed it, built it and run it. It is our own property, which is rather the point: it is where we test what actually earns a citation from a search engine or an AI assistant, on our own money instead of a client’s.
Built to be quoted
Assistants like ChatGPT, Perplexity and Google’s AI Overviews do not return ten links. They return a paragraph and cite two or three sources, and they lean towards sources that are independent, specific and checkable. Most sites are none of those things.
So every resort page opens with a verdict, a number and a date, in that order. “Twelve of the last fifteen Decembers had a 30cm base by the 15th” is a sentence a model can lift whole, because it is a claim with a quantity and a scope attached to it. Hedged seasonal marketing copy is not.
The methodology is a public page rather than a footnote. Anyone can read which winters were sampled, which dates the snapshots were taken on, and what counts as a usable base. That is the difference between a site that asserts and a site that can be checked, and being checkable is what makes a source safe to cite.
Fifteen winters of real data, not estimates
The snow probability figures come from the Open-Meteo historical archive, computed on fixed snapshot dates so the statistics match the sentences on the page exactly. Nothing is rounded into a nicer number and nothing is modelled where a measurement exists.
Around that sits live data: a seven-day forecast, avalanche levels from the European CAAML bulletins where a regional feed exists, public transport times, and three live webcams per resort served through a Cloudflare Worker so the API key never reaches the browser. Resort facts come from official sources and open reference data rather than from memory.
Where a feed does not exist, the page says so instead of guessing. A resort with no wired avalanche region shows nothing rather than a plausible-looking level, and a resort with no practical rail connection says exactly that. Refusing to fill a gap is unglamorous, and it is the reason the numbers on the rest of the page can be trusted.
The verdict is a state machine, not a template
Snow reporting is full of edge cases that break naive logic. A year-round glacier is open in August with zero modelled snowfall at village level. A resort’s opening day is an announced calendar date, not something to infer from a forecast. Data goes stale between scheduled runs.
Each of those is handled explicitly. The glacier verdict comes from a flag on the resort, never from a modelled zero. Opening dates are curated each autumn. When data is old, the page says how old it is rather than presenting last week’s numbers as today’s. Getting that wrong once is how a source stops being cited.
Static, cheap, updated daily
The site is Astro, statically rendered, on Cloudflare Pages. A scheduled job refreshes the data every day, commits it and redeploys. There is no database and no server to keep alive, so running costs stay near zero as the page count grows, and every page ships as HTML that a crawler or an AI agent can read without executing JavaScript.
Themes are token-level light and dark, and the hero shifts between day and dusk based on the actual sun position in the Alps rather than the visitor’s clock.
Stack: Astro, Tailwind, Cloudflare Pages and Workers, GitHub Actions, Open-Meteo, EAWS CAAML avalanche bulletins, Transitous.