How the numbers are made

1. Where the data comes from

Every night at 03:00 UTC a job reads Valve's public store APIs for every game on Steam: name, release date, list price in 13 regions, tags with their weights, developer, publisher, platform flags, review count and score. Nothing is scraped from store pages, no third-party dataset is used, and no figure on this site comes from anywhere but these calls plus the arithmetic below. Every page shows its observation date.

2. Copies: reviews × 30

Steam does not publish sales. We multiply a game's total review count by 30 and call the result est. copies. One multiplier for every game, disclosed here, never tuned per game or per genre. The true ratio varies by game, between 20 and 60, and it drifts over time as Steam changes how it prompts for reviews. That puts a typical error of ±30% on any single game's figure. Percentiles across dozens of games cancel most of it; a single game's row does not, and the "est." in front of every figure is the reminder. We check the multiplier against a small private reference set twice a year and print the date of the last check here: 17 Sep 2026.

3. Revenue: copies × list price

Est. revenue is est. copies multiplied by the game's US list price today, before discounts, regional pricing, refunds, Steam's cut, taxes and bundles. It is gross, not what the developer received; the developer's share is usually 50 to 70% of it after Valve's cut, VAT and refunds. Free-to-play games have a list price of zero, so their revenue is zero here and they are excluded from every money percentile; the page says how many free games a niche has. Games listed above $70 are excluded too, because those are almost always bundles or catalogue errors. Price changes over a game's life are ignored: a game that launched at $19.99 and now lists at $9.99 is counted at $9.99.

4. One entry per developer

Percentiles are computed over developers, not games. Every developer's games inside a niche and window collapse to that developer's median before the percentiles are taken. A studio that shipped 40 asset flips at $5 counts once, at the median of the 40. This is the single biggest difference between this site and a chart of game revenues, and it is why the numbers read "of 189 developers" rather than "of 300 games". It also means a developer with 1 game at $200k and 3 under $500 appears at the median of the 4, far below the $200k.

5. Which games count

A game is in the population when Steam types it as a game (not DLC, software, video or demo), it has a Steam release date in the past, and its list price is $70 or under. Free games count as released games and as announced games; they sit outside every money figure (item 3), and each niche page says how many it has. A game is in a niche when it carries every one of the niche's tags, at any tag weight. Early Access counts as released from its Early Access start. Delisted games stay in the data with their last observed figures. Announced means the game has a store page with coming-soon status and carries the niche's tags today.

6. Windows and cohorts

A window is a rolling period ending at the observation date: the last month, 3 months, 6 months, year, 2 years, 4 years, or all time, by first Steam release. The default on every page is 2 years. It is the span the verdict rules were set on, and it is short enough that the market it describes still exists. When fewer than 5 developers shipped in 2 years, the page widens to 4 years, then to all time, and says so in the plate caption.

A cohort is a fixed period by release quarter, half-year or year. Cohort charts mark the newest cohorts as "still earning". A game released 3 months ago has 3 months of reviews; one released 3 years ago has 3 years.

7. Momentum and the market baseline

Momentum compares two adjacent periods of equal length, for instance the last 12 months against the 12 before. Supply is the log2 ratio of developer counts between the two periods. Revenue momentum is the log2 ratio of a percentile (bottom 30%, median, top 25%, top 5%) in the later period over the earlier one, divided by the same ratio for the whole market. Dividing by the market cancels the age effect (younger games have fewer reviews) and the market-wide tide. So "up 30% on the market" means the niche gained 30% more than the market did. Momentum needs at least 3 developers in each period. Below that the page says "too few recent games to read momentum".

The market baseline is a different number. It is the per-developer median of every released game in the same effort tier (1-2, 3-4, 5-6, 7-8, 9-10) over the last 2 years. Verdicts compare a niche's median to it.

8. Verdicts

A verdict is a pure function of the numbers above, with the rules in one file and the three deciding numbers printed under every label. Ratio is the niche median over its tier baseline. Floor is the bottom-30% figure. Flood is announced games per developer who shipped in the last year. Tests run in the order Unrated, Dead, Strong odds, Fair, Poor odds, Long shot, Poor odds; the first that matches wins. Unrated: fewer than 10 released games, or no baseline. Dead: fewer than 5 developers shipped in the last year, or the median under $500 over both 2 and 4 years. Strong odds: ratio at least 1.5, floor at least $1,000, revenue momentum no worse than down 20% on the market, flood at most 1.5. Fair: ratio at least 0.7 without passing every Strong odds test, momentum better than down 30%, flood at most 2. Long shot: ratio under 0.7, momentum better than down 30%, flood at most 2, and the top 5% at 20 times the baseline or more. Poor odds: everything else, which is ratio under 0.7 with momentum at or below down 30% or flood above 2; ratio under 0.7 with a top 5% under 20 times the baseline; or ratio at or above 0.7 with momentum at or below down 30% or flood above 2. The rules are the same for every niche and never edited by hand for a page.

9. Effort and AI-buildability

Effort is a hand-written rubric per tag, 1 to 10, describing what a typical shipped game carrying that tag needs; a niche takes the highest effort of its tags. AI-buildability is a second hand-written rubric per tag, high, medium or low, describing how far current tooling gets on the systems that tag implies; a niche takes the lowest of its tags. Both rubrics are on their own pages with every placement visible (the effort scale, AI-buildability), and both are opinions, stated once, dated, and open to argument on the submit page.

10. Players and hours

Not collected yet. When the player sampler and reviewer playtime ship, this section describes them and the changelog (item 13) records the date.

11. What the numbers are not

They are not sales figures; Valve has those and does not publish them. They are not net income to the developer. They are not a forecast for any one game, and nothing here predicts what a particular game will make. They are not a judgment of quality: a $200 game can be excellent and a $2M game can be dull. They do not see wishlists, marketing spend, publisher deals, launch discounts, bundles, keys sold off-Steam, console ports, or refunds. They do not know who the reader is. A verdict is a summary of what happened to a population, printed with the three numbers that decided it, and nothing more.

12. Errors we know about

The review multiplier is the largest: ±30% on a single game, less on percentiles. Tags are set by developers and voted by players, so a game can carry a tag it barely fits, and a niche with a vague tag (Adventure, Indie) is wider than its name. Developer names are strings; two studios with the same name collapse into one, and one studio with two spellings splits into two. Regional list prices are collected but not converted; only the US price feeds revenue. Delisted games keep their last figures. Steam's own review count occasionally drops after a review purge, which lowers a game's est. copies overnight. The nightly job refuses to publish when the game count moves more than 3% in a day, when fewer than 95% of games have tags, when fewer than 12 of 13 regions are priced, or when any tag that had games yesterday has none today. On those nights the site shows the previous day and its date.

13. Changes to the method

Every change to the multiplier, the filters, the windows, the verdict thresholds or the rubrics is listed here with its date and the reason, newest first.

No changes yet.

Old figures are not restated; the observation date on each page says which rules produced it.

14. Glossary

niche

A set of 1 to 3 Steam tags a game must carry to count; the unit every page describes.

tag

A Steam store tag, as set by the developer and voted by players; we use a fixed list of 355.

mechanic, setting, mood

The three kinds of tag: what you do (M), where and what it is about (T), how it looks and feels (D).

effort

A 1-to-10 hand-written rating of what a typical shipped game with this tag needs to build; a niche takes its highest tag.

AI-buildability

A hand-written high, medium or low rating of how far today’s tooling gets on the systems a tag implies; a niche takes its lowest tag.

window

A rolling period ending at the observation date (last month to 4 years, or ever) by first Steam release; default 2 years.

cohort

Games grouped by release quarter, half-year or year; the newest cohorts are still earning.

momentum

How a niche’s revenue percentiles and developer count moved between two equal, adjacent periods, divided by the market’s own move.

on the market

Momentum after dividing by the whole market’s move; up 30% on the market means the niche gained 30% more than the market did.

announced

Games with a coming-soon Steam page today that carry all the niche’s tags.

devs, developers

Developers who released a game in the niche inside the window; each counts once, at the median of their games.

floor

The bottom-30% figure: 3 in 10 developers made less than this.

typical

The median: the middle developer’s est. revenue, half made more, half made less.

top 25%

The 75th percentile: 1 in 4 developers cleared this.

hit

The top-5% figure: 1 in 20 developers cleared this.

share

A niche’s summed daily peak players over Steam’s total peak the same day, as a percentage.

players now

The sum of the latest sampled concurrent players across the niche’s games.

hours played

Area under the sampled concurrent-players curve, per day, summed over the niche’s games.

est. copies

Review count × 30; an estimate, not a sales figure.

est. revenue

Est. copies × today’s US list price; gross, before Steam’s cut, taxes and refunds.

baseline

The per-developer median est. revenue of every released game in the same effort tier over the last 2 years.

ratio

The niche median divided by its baseline; ×2.1 means the middle developer here made 2.1 times the tier’s middle developer.

flood

Announced games divided by developers who shipped in the last year; above 2 is flooded.

verdict

One of Strong odds, Fair, Long shot, Poor odds, Dead or Unrated, computed by fixed rules from ratio, floor, momentum and flood.

comparable games

Released games carrying all the niche’s tags, ranked by est. revenue inside the window.

observed

The date the nightly run that produced this page finished.

specimen, plate

The field-notes names for a niche page and for a block on it; decoration, not data.