How we know this

A directory is only worth reading if you can see where its facts come from. Here is ours, including what we deliberately do not do.

Where the data comes from

• Places — public Google business listings: name, location, category, rating count, opening hours and the attributes Google itself publishes.
• Facilities — inferred by a language model from visitor reviews, then stored with the number of mentions and the date of the most recent one.
• Aerial imagery — USGS NAIP, public domain, with the survey year shown. Tree cover in the image is our second check on shade.
• Quiet times — Google's hourly visit patterns.

Why site plans, not photographs

We would rather show you a photograph of every playground. We checked what is actually available and legally reusable, and the answer is: nothing at scale. Wikimedia Commons has geotagged pictures near most large parks, but they are of birds and monuments, not play areas. The open street-level archives, KartaView and Panoramax, return nothing for this region. Google's photos may not be stored at all — their terms state a photo reference cannot be cached — so every visitor would trigger a fresh paid request. Commercial satellite basemaps carry licences that forbid exporting the imagery for a site like this.

So each page opens with a plan drawn from OpenStreetMap: the play area outline, the paths, the green space, at a fixed scale. Below it sits public-domain federal aerial photography. Both can be shown honestly and cover every playground, which a scraped photograph could not. Where visitors send us pictures, we will use those.

What we hold back

We also dropped a step-free filter. Google lists a wheelchair accessible entrance on 972 of the 992 playgrounds we track — 98% — which tells you nothing about which one to choose, and nothing at all about whether the equipment itself can be used. A family that needs accessible play deserves better than a list of almost everything, so we would rather publish nothing there until we have real data.

Fencing shows how our quality gate works. The first sample check caught the fence classifier at 67% precision: it counted the fence around a dog run, and a roof described as an enclosure, as a fenced play area. A parent choosing a playground for a child who bolts cannot use a number like that, so we switched fencing off, rewrote the classifier with those exact errors as counter-examples, and re-tested it: 10 out of 10 on a fresh sample. It is back on — shown only when at least two visitors mention it, and always as what visitors report, never as bare fact. An attribute that fails the check is withheld, fixed and re-measured; it is never quietly published.

What we do not do

• We do not republish review text or visitor photos. Those belong to the people who made them. We publish counts and dates.
• We do not state a fence or a splash pad as bare fact. Those are safety-relevant, so we always phrase them as what visitors report, with how many said it.
• We do not fill gaps with guesses. A playground with too little evidence gets a page that says so, or no page at all.

How accurate is it?

Before publishing, a random sample of claims is checked by hand against the underlying reviews and the aerial image. If precision on any attribute falls below 90%, the build is blocked and the classifier is fixed rather than the number.

Corrections

Playgrounds change: equipment is replaced, fences come down, splash pads close. If something here is wrong, tell us which playground and what changed and we will correct it in the next monthly run.