ABOUT
How the BANG Score is made
The value desk for Steam.
Steam BANGERS scores every game on Steam by whether the price matches what you get, and prints the answer as one number from 0 to 100, the BANG Score.
Games are art, and worth is subjective. Length is a design choice; a game built to run six hours is not failing at being twenty. The score reads value-fit: the price against how a game reviews and how long it was built to be. That is how a six-hour game people love can outscore a sixty-hour one people find dull.
The score is absolute, so a game's number does not change because some other game launched.
Numbers in the film as of Nov 6, 2026.
What goes in
- Value: the price against how long the game is, from HowLongToBeat where the game can be matched, floored by what its genre is supposed to be, so a short game built to be short is not punished for it. With no match, the genre's typical length stands in where that helps the game and a neutral value where it does not, and the score takes a dock either way until a real length arrives.
- Reviews: the Steam review score, the strongest signal of whether a game delivers. Reviews set the ceiling; price and length only move you within it.
- Engagement: how far players actually get. The finish rate is the global unlock rate of a game's ending achievement, from Valve's public achievement stats. Valve does not say who is counted. It only gates the very top of the scale, and a poor finish rate costs a game nothing.
- Monetization: in-app purchases drag the score down, graded by how aggressive they are. Cosmetic is a light touch; loot boxes and pay-to-win are heavy. With no PCGamingWiki entry, Steam's own In-App Purchases flag takes a light dock. A game documented as having none gets a small credit, unless Steam lists in-app purchases for it. Source: PCGamingWiki and the Steam store page.
- Players: multiplayer and roguelike games have no ending to divide the price by, so their value reads the player base instead: SteamSpy's estimate of how many own the game, and SteamSpy's count of how many were playing it at once. A big game nobody plays anymore takes a dock. Free games have no price to divide, so theirs reads the owner count the same way. Source: SteamSpy.
- Recent reviews: when at least 50 reviews from the last 30 days run well above or below the lifetime score, the score moves a few points with them. Source: Steam.
- DRM: Denuvo, any other DRM notice, and a required third-party account are small docks, capped so they can never sink a good game. Source: the Steam store page.
- Early access: an unfinished game scores a little lower, because it may never be finished.
- AI content: a game whose developer disclosed AI-generated content on Steam takes a heavy dock. It reads the disclosure Valve requires from every developer, so a studio that uses AI and does not say so is not caught. Source: Valve.
- Refunds: the share of a game's Steam-purchase reviews whose writer refunded it, from Steam's own Product refunded label. An unusual share is a small dock. Steam keeps a refunded buyer's review and counts it in the review score; this site checked on 2026-08-27 (Gollum's 41% of 829 reproduces only with its 21 refunded reviews counted). The site also shows Steam's own summary of the reviews written inside the two-hour refund window. Source: Steam.
We weigh these into one number. We never publish the weights. The recipe is the one part that stays ours; everything else, sources included, is on the table.
The tiers
How to read a verdict
Every game page opens with one command, derived from the tier, the game's own BANGER price and its own price record. It means exactly what it says:
- BUY IT: A BANGER or worth it at today's price with no seasonal sale to wait for, or fair or better on a discount at the lowest Steam price we have tracked or at its typical cut in a running seasonal sale.
- PLAY IT: A free game that scores a BANGER or worth it.
- WAIT: A better buy is on record: a price that lifts its tier and that it has sold at before, or a cut in its own seasonal record with the next Valve sale opening within two weeks, or a regular cut of 50% or more within six. The page names the price or the sale.
- NO RUSH: Fair value at today's price. Buy it if you want it.
- SKIP IT: Under the fair line at today's price, and no price it has sold at lifts it.
- CHECK PRICE: The last price we read was a sale that may have ended. The page dates that reading.
Below the verdict, each input wears a grade from S to D, read from the buyer's side: an S in Reviews means the reviews push the score up hard, a D in Engagement means a low share of players reaches the ending achievement. A ▲ marks a lift, a ▼ a drag or a dock, and a ? an input we do not have. The grades explain the score; they never change it.
The honest parts
- Value is estimated from the current price, so if you bought it on sale three years ago the score knows nothing about that. Dollars-per-hour divides that price by the game's length; your own playtime never enters it.
- A great game at a steep price-per-hour can still land at worth it, and that part is on purpose.
- A game needs enough reviews to be scored at all. Better a gap than a guess.
- Three things come from SteamSpy, and SteamSpy has not updated its per-game numbers since about mid-2025: the owner estimates, the count of players at once that the Players input reads, and the user tags that sort games into genres. They are an old snapshot, so no page puts a date on them. The players online number on a game page is Steam's own, taken when we last checked and printed with that time. We are checking whether Steam's count should replace SteamSpy's in the score; until it does, the score reads SteamSpy's.
Predicted scores, for unreleased games
This site works out a PREDICTED BANG Score for some unreleased games, with an 80 percent band and the odds of landing above the typical released game in its genre. It is worked out from the track record behind the game: the studio's own scored games and its publisher's games from other studios, each weighed by how many there are, so a studio with one game leans on a publisher with forty. Those games are read two ways and mixed: their reviews run through this formula at the announced price, and the scores they actually got here. A developer's AI disclosure docks it the way it will dock the real score. It is a forecast, and no page shows one while the home page's upcoming rail is off. Checked September 24, 2026, on the 3,874 games released in the year before that drew 5 or more reviews and whose record would have earned a forecast. Each forecast used the studio's or publisher's earlier games and the game's price as they stand today, not as they stood at launch: 8 in 10 landed inside their range, and the forecast missed by 9.9 points on average, against 12.1 for quoting the genre's typical score. The kinds of record that get a forecast were picked on this same check, so these numbers flatter it a little. The ranges are wide: 32 points on a typical card, where the middle 80 percent of real scores in a genre spans 41. A card shows a forecast only when the studio itself has a real record here: 3 or more scored games, or games that typically drew 25 or more reviews. Then its kind of record, sorted by how many games and how many reviews they drew, has to have beaten quoting the genre's typical score by at least 10 percent over the past three years of releases, by more than chance (a 95 percent interval, with each studio counted once), in every one of those years, with a range narrower than the genre's own. 5 of 16 kinds cleared it. A studio's own record cleared it, except with 1 or 2 games that typically drew 1,000 or more reviews, or 3 or more games that typically drew 1,000 or more reviews. A publisher's record never earns a forecast on its own; it only counts beside the studio's. Every other unreleased game gets no forecast and no number in its place.- Refunded reviews count reviewers, and a refund inside two hours rarely leaves a review at all, so the share usually runs under what really happened. Use it to rank games against each other; it is too thin to read as a game's refund rate.
- The scores rebuild every six hours from whatever the crawlers have read since. The crawlers run every hour, but each run reads only a slice of the catalog, so a game's reviews can be weeks old; its price is re-read more often. Each game page dates both under the numbers.
What it does not do
- Not affiliated with Valve or Steam.
- Steam never gives us your password, email, purchase history or payment details. Sign in with Steam is opt-in. Signing in with Steam stores your public SteamID, your persona name and avatar, whether your profile is public, your owned games with playtime, the games you played in the last two weeks, and your wishlist. Each sync replaces the last one. Delete my data on the For You page erases all of it.
From the heart
Games are subjective, and one number can't hold what a game means to you. Some of my favorite games of all time aren't BANGERS, because they are too short for the price, or a niche reviews never rewarded, or a slow burn the math doesn't catch. I love them anyway, and a low score doesn't make me wrong about that.
So don't take a low score personally. The BANG Score answers one narrow question for one specific person: the budget-conscious gamer working out where a limited dollar should go. That was me, counting out twenty dollars and needing it to count. A "skip" only means the reviews, length and price do not add up right now, for someone on a budget. If a game looks like your thing, buy it. The right game is always worth the money.
Common questions
What is the BANG Score?
The BANG Score rates every game on Steam from 0 to 100 for value: whether the price matches what you get. Whether the game is your kind of thing is a question it leaves to you.
How is the BANG Score calculated?
The Steam review score sets the ceiling. Inside it, the score weighs the current price against how long the game is, using HowLongToBeat's length where the game is matched. With no match it uses the genre's typical length where that helps the game and a neutral value where it does not, and docks the score for the guess either way. Free, multiplayer and roguelike games have no price or no ending to divide by, so how many people own and play them stands in. A developer's disclosure of AI-generated content on Steam takes a heavy dock. Smaller adjustments cover how aggressive the in-app purchases are, an unusual share of Steam-purchase reviews from buyers who refunded, DRM, early access, a recent swing in reviews, and a top-end bonus for how far players get by public achievement stats. The exact weights stay private; every input and data source is public.
Is the BANG Score the same as a review score?
No. A review score measures whether a game is good. The BANG Score measures whether it is worth the money, so a great game at a steep price per hour can still land at worth it.
Where does the data come from?
Valve and Steam, Steam's Web API for the players online number, HowLongToBeat for length, SteamSpy for ownership estimates and user tags (an old snapshot: SteamSpy has not updated its per-game numbers since about mid-2025), IsThereAnyDeal for price history, PCGamingWiki for monetization mechanics, Metacritic for critic scores, and Frankfurter (European Central Bank rates) for currency display. The seasonal sale dates and the discount rules on the game pages come from Valve's Steamworks documentation.
How often is it updated?
The scores rebuild every six hours from whatever the crawlers have read since. The crawlers run every hour, but each run reads only a slice of the catalog, so a game's reviews can be weeks old; its price is re-read more often. Each game page dates both under the numbers.
Is Steam BANGERS free?
Yes. Every score and page is free to read. There is also a free API and an MCP server so developers and AI assistants can query a game's value.
Who makes this
Steam BANGERS is run by one person, Jordan Allen Lewis. I decide what it measures, how the score is built, and which sources it trusts. AI coding tools write the code under my direction. The score itself is plain arithmetic on public data with no AI in it, so when a score moves, it is because one of its inputs changed or the formula did.
The guides and collection write-ups are made the same way and go out under my name, so I answer for every claim in them. The claims come with the numbers that back them, and where the data is missing, the page says so instead of guessing.
Why I built it, and what I got wrong first: the longer version. How the number is actually computed: the method, including where it is weak.
Contact
Found a bug, a wrong score, or have an idea? Send me a message on LinkedIn.