HOW IT WORKS
How GameBro decides whether it is a good time to buy
GameBro reads a Steam game's price against its own history: how low it has gone, how often it goes on sale and how deep. From that it gives a verdict and estimates when the next sale may come, always saying how much confidence that estimate deserves.
Where the data comes from
- Today's price comes from Steam's US store, in dollars. It was checked against Steam on eight games: the difference was zero cents.
- The history comes from IsThereAnyDeal, which records every Steam price change with its time. It is taken from 12 October 2024, the day the history of games-popularity.com, the previous source, began; the forecast was built and measured on that span. The forecast starts there for every game, even ones released long before: with the full history it was no more accurate and its discount range grew twice as wide. When IsThereAnyDeal lacks the game or does not answer, the history comes from games-popularity.com.
- The lowest price does use IsThereAnyDeal's whole record. It is an all-time low only when that record reaches the game's release; when it starts later, GameBro says since which year ("lowest since 2012"). The lowest since October 2024 stays in the game page's detail.
- Steam changes a price instantly and the history may take a while to record it (with games-popularity.com, up to about 12 hours). In that gap the chart ends at Steam's current price and says so.
- Prices in other currencies are not converted at an exchange rate: Valve sets regional prices by hand and discounts are global percentages, so those percentages are applied to the local base price.
The verdict
Each game gets one, with its reason in a sentence:
- Good time to buy: it is at its lowest price on record (or within 2% of it), or at a discount of at least 90% of the best it has offered, and nothing better is expected soon.
- Decent time to buy: today's discount is at least 60% of its best discount.
- Better to wait: a sale at least five points deeper is expected within the next six weeks.
- Bad time to buy: it is at full price and the game has been discounted before.
- No reference yet: it has never been discounted in the recorded period, so there is nothing to compare with.
The next-sale forecast
The forecast is a window ("in 3 to 9 weeks"), an expected discount ("around −80%") and a confidence level.
- The window comes from the game's own rhythm: how often it has been discounted and how long it has been since the last time.
- Steam's sale calendar (summer, autumn, winter, spring) can only bring the window forward or widen it, never push it later. Anchoring on the calendar made the forecast worse: it missed the sales a game runs on its own before the event.
- The expected discount comes from its latest sales. The top of the range can go past its record: five points more if its last sale was its deepest, if the range comes from a seasonal sale, or if it has fewer than five sales, because that is where games tend to go deeper than ever.
- Confidence is high when the game has twelve or more sales, is overdue against its rhythm, or the window matches a seasonal sale it usually joins. With two to four sales, or no measurable rhythm, it is low.
- News and announcements about the game are shown as context, but move no number.
How often it is right
It was tested with a backtest: the history is cut at a point in the past, a forecast is made knowing only what was known then, and it is compared with what happened. In October 2026, with 1,075 predictions over 116 games:
| Confidence | Predictions | Right | By chance | Date error | Discount in range |
|---|---|---|---|---|---|
| High | 195 | 80% | 18% | 9 days | 91% |
| Medium | 703 | 69% | 18% | 11 days | 87% |
| Low | 177 | 53% | 19% | 14 days | 90% |
| All | 1,075 | 69% | 18% | 11 days | 88% |
- The percentage is more reliable than the date. The median discount error was 0 points; the median date error, 11 days. That is why the page shows the percentage plainly and the date as a wide window.
- Each prediction is scored against a window of the same width placed at random, so a wide window cannot be "right" by luck. A sale that never came counts as a miss.
- The result holds outside the games the model was built on: it does about as well on the most played games and on recent releases.
- Besides the backtest, every forecast the page shows is stored and checked against what Steam does next.
Reviews
Steam's score accumulates from launch, so a game that launched broken and was fixed keeps dragging its old score. GameBro looks at how many reviews come in each fortnight and what share of those is positive. The score counts reviews in every language.
What GameBro is not
- It sells no games and uses no affiliate links. The Steam button goes to Steam's store.
- It is not affiliated with Valve or Steam.
- It is free and runs on donations. It is made in Costa Rica.
- A verdict holds for the day it is read: it changes with the price.