Real-world datasets served as JSON for prototypes, demos, tests, and anything else that needs believable data without a backend. There's no key and no sign-up, and any origin can read them.
https://hypothesis.sh/datasets
https://hypothesis.sh/datasets/periodic_table
https://hypothesis.sh/datasets/periodic_table/26
https://hypothesis.sh/datasets/periodic_table/fe| Path | Returns |
|---|---|
/datasets | Every dataset: name, title, record count, license, source |
/datasets/{name} | One dataset: license, metadata, field definitions, records |
/datasets/{name}/{key} | One record under data, with the dataset's license |
Anything else returns a 404 with a JSON error and a list of the dataset URLs that do exist.
The datasets are published by hypothesis.sh under CC BY 4.0. You can use them for anything, commercial use included, as long as you credit hypothesis.sh.
Every successful response body includes license, placed before data. A client that only reads the body still gets the terms, including agents that never look at headers. The license comes with an attribution line you can paste as-is:
"license": {
"name": "CC BY 4.0",
"url": "https://creativecommons.org/licenses/by/4.0/",
"attribution": "Periodic table of elements by hypothesis.sh (https://hypothesis.sh/datasets/periodic_table), licensed under CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/)."
}If you ship something built on a dataset (an app, a page, a generated file), put its attribution text where the data appears or in your credits. Responses also carry a Link: <…>; rel="license" header pointing to the license.
A dataset response looks like this, with data holding the records:
{
"name": "periodic_table",
"title": "Periodic table of elements",
"count": 118,
"license": { "name": "CC BY 4.0", "url": "…", "attribution": "…" },
"source": {
"name": "Wikidata",
"url": "https://www.wikidata.org",
"license": "CC0 1.0",
"retrieved_at": "…"
},
"primary_key": "atomic_number",
"fields": [
{
"name": "atomic_number",
"type": "integer",
"nullable": false,
"unit": null,
"description": "…"
}
],
"notes": ["…"],
"data": [{ "atomic_number": 1, "symbol": "H", "name": "Hydrogen", "…": "…" }]
}A single record comes in the same wrapper, with the record itself under data:
{
"dataset": "periodic_table",
"url": "/datasets/periodic_table/fe",
"license": { "name": "CC BY 4.0", "url": "…", "attribution": "…" },
"source": { "…": "…" },
"data": { "atomic_number": 26, "symbol": "Fe", "name": "Iron", "…": "…" }
}fields describes every key in a record, in order: its type (integer, number, string, boolean, string[], or integer[]), whether it can be null, and its unit. A missing value is always null, never an empty string or 0, and every record carries every key. notes explains how values were chosen.
source records where the data came from. It's provenance only; the license that applies is the one in license.
All 118 elements, 32 fields each:
Look up a single element by atomic number, symbol, or name, ignoring case. Names use IUPAC spelling (Aluminium, Caesium, Sulfur), but aluminum, cesium, and sulphur also work.
/datasets/periodic_table/26
/datasets/periodic_table/fe
/datasets/periodic_table/iron
/datasets/periodic_table/aluminumUnits: temperatures in kelvin, density in kg/m³ (the same as g/L), energies in eV, radii in picometres. Each field's unit says which. To draw the table, use grid_row and grid_column, which follow the IUPAC 2021 layout: Lu and Lr in group 3, with La–Yb and Ac–No on rows 9 and 10 below an empty row 8.
The records are built from Wikidata, whose data is released under CC0. That puts no conditions on reuse, which is why hypothesis.sh can publish the result under its own license. Wikidata is community-edited, so every value goes through these checks before it's served:
null.The upshot: a value that's present has passed every check, and a value that's uncertain is `null` rather than a guess. That trades coverage for trust. Density and specific heat are present for about 45 elements, and boiling point for about 80.
null.null.Four real activity recordings in Garmin's FIT format, served exactly as recorded:
| File | Activity |
|---|---|
fort-worth-trek-100-texas-2025-09-20.fit | Road ride, Garmin Edge |
friday-ski-2026-01-16.fit | Alpine skiing |
halfway-house-morses-mill-2026-06-21.fit | Hike |
old-west-austin-deep-eddy-2026-05-03.fit | Run |
/datasets/fit_activities lists them with the license, and each file downloads from its file_url:
https://hypothesis.sh/datasets/fit_activities
https://hypothesis.sh/datasets/fit_activities/friday-ski-2026-01-16.fitDownloads are application/octet-stream (FIT has no registered media type). They're readable from any origin and carry the license in a Link: <…>; rel="license" header.
Every response sends Access-Control-Allow-Origin: *, so fetch() works from any page. Datasets only change with a deploy, so responses are cached at the CDN for a day and in browsers for an hour.
const res = await fetch("https://hypothesis.sh/datasets/periodic_table");
const { data, license } = await res.json();
console.log(license.attribution); // show this wherever the data appears
const gases = data.filter((e) => e.standard_state === "gas");