Real signals. A little gallows humor.
We combine current Steam activity, historical momentum and publisher announcements. Labels make the directory easier to scan; the underlying counts, time periods and source links stay visible. A population signal does not measure game quality, server uptime or matchmaking speed.
How the health tags work
| Signal | Evidence required |
|---|---|
| † Certified DEAD | A reviewed official announcement confirming closure of the game’s service. Low counts or an automated headline match cannot award this label. |
| RIP · content ended | Reviewed evidence that development or seasonal content ended. The report explains which modes remain playable. |
| Resurrection alert | A separate momentum tag: average concurrency grows at least 50%, with a previous average of at least 100, a current average of at least 250, and an absolute gain of at least 100. The population label remains visible. |
| Cooling off | A separate momentum tag: a comparison-period decline of at least 40%, from at least 250 average players, with an absolute drop of at least 100. A large audience can be Thriving and Cooling off simultaneously. This is not evidence of a dying game. |
| Life support | A fresh count below 250 Steam clients and sustained low, declining averages. We require a decline of at least 50% and at least 100 players from a baseline of at least 250, plus the persistence checks below. A single month’s drop is insufficient. |
| Thriving | At least 25,000 clients in the latest fresh Steam observation. |
| Healthy | 2,500–24,999 clients in the latest fresh Steam observation. |
| Surviving | 1–2,499 clients in the latest fresh Steam observation. |
| No pulse detected / Signal unavailable | A valid zero / missing or stale data. Neither establishes a shutdown. |
Reviewed lifecycle status takes precedence. For other games, stale observations withhold automated health labels. Population health and momentum are shown separately: a post-launch or league-cycle drop does not erase a large current audience. We do not infer the cause of a change or predict that a game will shut down. Thresholds are editorial rules, not a scientifically validated survival prediction.
When “Life support” is justified
The current count must be below 250. With 28 complete observed days, we compare the latest 14-day average with the preceding 14. Both of the latest two weekly averages must remain below 250, and the latest week cannot be higher than the week before. The earlier fortnight must average at least 250; the decline must be at least 50% and 100 players. All 28 days need observations in at least 18 distinct UTC hours.
Without enough daily coverage, the fallback requires three consecutive completed months. The latest two averages must both be below 250, each month must decline by at least 10% from its predecessor, and the total decline from the oldest month must be at least 50% and 100 players from a baseline of at least 250. If a fully covered daily window shows recovery, we do not override it with older monthly data. A small but stable community remains Surviving.
Trend windows and coverage
Our preferred signal compares the last seven completed UTC days with the preceding seven. Every day in both windows must have observations in at least 18 distinct UTC hours. The average of the seven daily sample averages is used for each window. Missing days, zero baselines and insufficient coverage produce no weekly percentage.
Until the internal archive has enough coverage, we can compare the two most recent completed calendar-month averages from Steam Charts. The report explicitly names those months. Trend classifications require consecutive months, a latest month starting within 70 days, and source data retrieved within 48 hours. We never present a monthly comparison as a daily surge or merge external monthly summaries into our own daily records.
Current activity: directly from Steam
The server calls the Steam player-count API. The tracked directory is sampled every 30 minutes while the server runs. Individual reports can refresh using a five-minute cache. A directory count becomes stale after 60 minutes; on-demand lookup counts become stale after ten minutes. Failed requests retain the last successful timestamp and mark the result stale.
These are app-wide connected Steam clients. They can include solo, local, idle and bot sessions. Offline players, consoles, other launchers and mobile editions are not included. A multiplayer listing does not mean every running client is available to join your mode, region or lobby.
The internal daily archive
We store actual timestamped observations in SQLite. Each UTC day retains its sample average, minimum, maximum, number of observations and distinct hours sampled. Daily averages are averages of samples, not time-weighted estimates; additional on-demand readings can affect them. Charts show gaps and mark limited coverage. Raw observations are retained for 30 days; daily aggregates, archived source snapshots and lifecycle events are retained long term. CSV downloads expose the underlying records.
The short-term chart connects actual observations only and breaks at gaps over 45 minutes. “24h observed high” means the highest recorded sample, with the sample count shown. A 24-hour comparison requires a current sample no more than ten minutes old and a baseline within 15 minutes of 24 hours earlier.
Historical sources and recorded milestones
Steam Charts describes hourly sampling beginning in July 2012; individual game coverage may begin later. We display up to 12 completed calendar months separately from its rolling “Last 30 Days” row. Monthly averages are sampled concurrency, not monthly active users. A game’s first month may be partial. Highest-recorded peaks can miss earlier activity or spikes between observations; record months appear only when the source contains a matching peak.
Historical pages use a six-hour cache. The daily collector archives retrieved source snapshots; a failed refresh preserves the old retrieval date. Source HTML is parsed conservatively, and an unrecognized format is treated as unavailable.
News, patch announcements and new releases
We use the Steam news API with the game’s community/publisher announcement feed. A patchnotes tag marks a patch announcement in the latest 20 feed items. Publication dates do not independently verify deployment dates. The on-demand feed cache is 30 minutes.
The daily collector retrieves announcements, historical tables and recent multiplayer listings from Steam’s store search. New releases require an actual past or current date in the returned listing. The ticker links to the original announcement or store page. A quiet feed does not prove support ended, and a release listing does not certify service availability.
Automation and the Graveyard
The daily job starts from 00:15 UTC, with a catch-up attempt after a restart. It archives sources, records lifecycle changes and queues shutdown-related headlines for review. Interrupted runs can retry; successful daily runs are not repeated automatically. Time missed while the host was offline stays missing.
The Graveyard is a sardonic Hall of Fame with receipts, including games outside Steam. Its past-12-months filter uses the actual shutdown or final-content date, not the announcement date. Confirmed server closures are separate from retired content roadmaps such as Destiny 2’s Monument of Triumph update. Only reviewed official evidence can change a certification. Corrections go to hello@deadgames.gg. Inspect collection status.
Independent by design
Sponsorship does not change ordering, thresholds or conclusions. Missing data never creates synthetic counts, viewers, queue times or historical curves. Game names and artwork belong to their respective owners. We are not affiliated with Valve, Steam Charts or game publishers.
Single-player assessments
Steam store categories determine whether a game lists single-player only, multiplayer only, both, or unknown modes. Only single-player-only listings replace the population badge with a review-based assessment. At least 20 reviews and store/review checks within 48 hours are required: 80% or more positive is well received, 50–79% is mixed reception, and below 50% is poorly received. These are our descriptive bands, not Steam's rating labels. Certified publisher lifecycle evidence still takes priority. Quiet development and concurrency do not reduce the solo assessment.
Review totals use all languages and purchase types with Steam's off-topic filter enabled. We inspect up to 500 newest reviews by creation date for 30-day sentiment. A comparison with the previous 30 days requires full retrieved coverage and at least 20 reviews in each window. Limited samples and missing comparisons are labeled. A US store listing is not a worldwide purchase or offline-play guarantee. Store and review snapshots refresh daily and retain their original timestamps if a source fails.
Requested games join ongoing collection
Opening a validated game report enrolls it in a persistent collection queue. New games target daily collection; requests in at least three separate half-hour windows in a UTC day shorten the target to six hours. This is aggregate demand, not unique visitors. One worker processes up to five due games each minute, so backlog or source failures can delay collection. Failures retry with backoff; restarts resume overdue work. No personal visitor identity is stored. Sparse daily observations remain explicitly partial; enrollment does not fabricate historical coverage.