strikeoo/steam-launch-data

Steam Launch Window Data

Data behind steam-launch-window, a tool that helps indie developers pick a Steam launch week. Updated weekly (the raw gp/ files daily). Last update: 2026-11-09.

Not affiliated with or endorsed by Valve. The data comes from Steam's public store pages and endpoints, collected politely (rate-limited, no logins), and the gp_ configs from Games Popularity. No personal data: no Steam IDs, usernames or review texts, only counts.

What is here

Config	Kind	Rows	What
history	today's view	88,076	games launched on Steam since 2022: launch date and type, price, tags, languages, demo, reviews in the first day, week and 30 days
upcoming_snapshots	raw, daily	1,858,339	one row per coming-soon game per day (2026-10 to 2026-11): release date and its precision, Early Access, wishlist rank, price if set
upcoming_apps	latest	67,286	name, developers, publishers, tags with vote weights, languages, demo and its release time; last_seen is the last day the game was coming soon
release_changes	derived	80,729	each game's first day in the snapshots, every change of its release date text or range, each day it left the coming-soon list, and each day it came back (returned)
gp_followers	raw	588,863	Steam follower history from Games Popularity, about one point a day since 2024-03-15: appid, followers, observed_at (their time, UTC), fetched_at (ours); one file per fetch month
gp_players	raw	1,293,392	concurrent players, hourly, for the first month after launch, from Games Popularity; same columns with players
events	curated	308	Steam sales, Next Fests, themed fests, major releases, regional holidays, each with a source; starts_at/ends_at (UTC) give the exact window: Steam events run from 10:00 Pacific on start to 10:00 Pacific on end
tags	latest	446	Steam tag id to name, in English and in Steam's Simplified Chinese (name_schinese)
runs	log	31	one row per daily collection: games, requests, whether the wishlist crawl worked

from datasets import load_dataset
history = load_dataset("<this repo>", "history")

When each value was observed

• Observed at the time: upcoming_snapshots (one row per day, snapshot_date in UTC), runs. Use these for "what was known before launch". The gp_ configs were recorded at the time by Games Popularity (observed_at) but fetched by us later, often long after (fetched_at). Each fetch only asks for points newer than the ones already fetched, so a game's series is spread over several fetch days, and so over several fetch-month files.
• Frozen before launch: a launched game's upcoming_apps row stops updating when it leaves the coming-soon list, so its tags, publisher, demo and languages are pre-launch values when last_seen is before the launch date.
• Today's view: history. Price, demo, Early Access status and tags are as they are now, and tags keep being voted after launch. Review counts cover a fixed window after launch but only reviews that still exist. first_listed/last_listed say when a game was first and last seen in Steam's list of released games; rows of games that disappear are kept.

Things to know before using it

• Dates are Pacific time. Fuzzy release dates ("Q3 2027") have a precision and a date to date_end range.
• Reviews of paid games are from people who bought the game on Steam (key activations excluded), like the store's own count. Free games count all reviews, since nobody buys them; reviews_purchase says which (steam or all; empty means steam). It is set per row when the row is counted, so trust it over is_free. "Free" is the game's status at counting time, so a game that was paid at launch and went free later counts all reviews, including key activations. One exception: a free game's reviews_before_launch above 0 may still be a Steam-purchases count, which is a lower bound. Delisted games that left before collection started are missing.
• launch_type: full, early_access, or ea_exit (a 1.0 launch of a game that was already on sale in Early Access; treat these apart).
• Wishlist rank only covers Steam's top ~5,300 wishlisted games.
• Upcoming prices are almost always missing: paid games set them at launch.
• Missing days: runs shows which days were collected.

Data rights

No license is granted here for the data. The code that builds it is MIT licensed (LICENSE), and that license covers the code only. The facts in this dataset (release dates, tags, review and follower counts) come from Steam, owned by Valve Corporation, which has not licensed or endorsed this dataset. Game names are trademarks of their owners. Whether you may reuse the data depends on Valve's terms for Steam data and on your local law. The gp_ follower and player histories come from Games Popularity and are republished with its owner's permission (2026-10-12); credit them if you use them. The gp/ folder holds the same points as gzip CSV, one file per fetch day, updated daily; the gp_ configs are a weekly Parquet copy of them. It also holds latest_followers.csv, the newest follower count per game (appid, followers, observed_at), and state.csv and runs.csv, the collector's working files (what was fetched when, and with what result).

Full methods: README and endpoint notes.