Load any dataset into your own code

The id of a dataset is its address with dots for slashes, so /power/states.csv is power.states. Its current rows are at https://jesseahawkins.com/data/v1/datasets/power.states.json. A field of numbers holds numbers, and an empty cell holds null.

curl -s https://jesseahawkins.com/data/v1/datasets/power.states.json | jq '.rows[0]'

import pandas as pd, requests

r = requests.get("https://jesseahawkins.com/data/v1/datasets/power.states.json").json()
df = pd.DataFrame(r["rows"])
df.shape  # (51, 16)

const r = await fetch("https://jesseahawkins.com/data/v1/datasets/power.states.json").then((x) => x.json());
console.table(r.rows.slice(0, 5));

df <- read.csv("https://jesseahawkins.com/power/states.csv")
dim(df)  # 51 16

/llms.txt lists every section and dataset in plain text. https://jesseahawkins.com/data/mcp is a Model Context Protocol server over Streamable HTTP. Add it to an assistant as a remote server; it needs no key and keeps no session. It reads the same files as this API, and it cannot write.

• A dataset keeps its id, and a field keeps its name and its type. A field of numbers stays numbers.
• New fields can appear, so read fields by name and ignore the ones you do not use.
• A release file never changes after it is published. It carries its own field types, read from its own rows, and an old release can type a field differently from the current rows.
• A dataset that stops is listed under retired in the index, with the day it stopped.
• A change that would break any of these gets a new version at a new address, and version 1 stays.

Each field in a dataset file has a tier. A field that is an identifier or a place, and most text fields, have null. When one field holds values of two tiers, tier is null and tierBy names the field of the same row that decides, with the tier of each of its values.

quoted

The source publishes this value, and this site copies it. A change of unit or of rounding is allowed. Nothing else is.

derived

This site calculates the value from quoted values with fixed arithmetic: a sum, an average, a share, a change, a count or a rank.

modelled

An estimate from a model, made by the source or by this site. A value calculated from a modelled value is modelled.

Each dataset file also has freshness: the state that /status/ shows, with a notice when the data is due or late. The files are static, so the state is true for asOf, the day of the build. The dataset is due from next and late from late; compare those dates with the day you read the file.

Each dataset carries its terms in terms, in the words of its own sources. Credit the source a dataset names when you reuse it. The files are static, so fetch them as often as you need, and use the release files when a result must not change.