Methodology

This page explains how we arrive at the numbers Dragon Eye shows. Your game's data comes straight from Steamworks and your social networks, market data from public Steam data. We always show estimates as a range, not a single number – Steam games differ enormously and the average is not your game.

Own games and games of interest

An own game has Steamworks data (wishlists, sales), so you can only connect it with a key that has access to it – we check this when you connect it. You mark a game of interest with the star in the Market comparison tab or in game search. We fetch its public data (reviews, price, discount, players online) every day into our database, so the comparison with your game loads instantly and also shows how many reviews it gained over the last 30 days.

Comparison at the same stage

We split a game's life into stages: store page without a demo, demo out, first month after launch and after launch. Your game is on day N of its stage, and for each game of interest we show where it was on day N of the same stage and how the stage turned out.

We find the start of each stage in public Steam data. For your game, the store page date comes from the game settings, otherwise from the first day with wishlists; for games of interest from the developer's first announcement (approximate, and only if it came before the demo or launch – otherwise the page date is unknown and the game is left out of the store page comparison). The demo stage starts on the demo's release date, or – if the demo is gone from Steam – on the first announcement mentioning it. If the first review came more than a month before the release date, the game launched in Early Access and we count the launch from the first review. We spot Next Fest from announcements that mention it.

Other games' wishlists, followers and page traffic aren't public. Before launch we therefore only compare how things went: how many days until the demo and the launch, Next Fest and the number of announcements. After launch we compare reviews by day N; for games of interest we work them out from monthly totals. We only draw a conclusion when at least three games of interest have data on the stage.

Sales estimates from reviews

We estimate units sold from the number of reviews: the review count times a multiplier. Today one review stands for fewer units than ten years ago, so we pick the multiplier by release year. For games released since 2022 we use 20× for the low, 30× for the mid and 55× for the high estimate. Early Access games get 25% added. We don't estimate free games. We show the range from the low to the high estimate.

We adjust the multiplier by rating: happy players review more often, so a game with 90%+ positive has fewer units per review than a game around 70%. We therefore scale the range by a factor of 0.86–1.71. Below 70% positive and under 10 reviews we don't adjust, because the adjustment isn't reliable there. Under 50 reviews the estimate is only rough and we mark it with an asterisk.

A review-based estimate is for orientation: for many games reality differs by more than a third. Treat it as an order of magnitude, not a fact. We don't estimate other games' revenue yet – the store price is before discounts, regional pricing, refunds, VAT and Steam's share.

Reviews: Steam categories and weighted rating

We compute categories the same way Steam does: none under 10 reviews; 80% positive or more = Positive (10–49 reviews) or Very Positive (50 or more), 95% with 500 reviews = Overwhelmingly Positive; 70–79% Mostly Positive; 40–69% Mixed; negative categories below. From that we tell you how many positive reviews you need for a better category and how many negative ones would drop you below 80%. The weighted rating pulls the positive share towards 50% the harder, the fewer reviews a game has: rating = s − (s − 0.5) · 2^(−log₁₀(n + 1)), where s is the positive share and n the review count. With 10 reviews the score moves about halfway to 50%, with 1,000 by an eighth. That lets us compare a game with 20 reviews and one with 5,000 fairly. We use reviews as the Steam store shows them.

Wishlists and the first-week estimate

Net wishlists = added − deleted − purchased − gifted. Velocity = net adds per day (7-day average), trend = slope of the line fitted to the last 28 days relative to their average, deletion rate = deleted / added over 28 days, launch conversion = wishlist purchases since launch / wishlists on launch day. For an unreleased game we estimate the first week with the median conversion of released games: 0.11× wishlists, 0.15× above 25,000 wishlists, 0.10× for games over $10; first year ≈ 2.6× the first week. Games differ from the median 10–20×, so it's orientation only. The numbers come from games released before Steam's personal calendar launched (23 Oct 2026), so we will recalculate them.

What brought wishlists

For each event and post we add up wishlist adds on the day it was published and the two days after (0–72 h) and compare them with the baseline: the median of daily adds over 14 days before the event, leaving out days inside other events' windows, adjusted for the day of the week (the day's ratio to the average of the last 8 weeks). A difference above twice the usual spread that is also at least 30% of the expectation is a clear effect; ± shows the usual spread. When several events' windows overlap, we evaluate them together and split the lift by reach (by views without reach, otherwise evenly). It's a correlation in time, not proof of cause. Steamworks UTM data is shown separately – we don't match it to individual posts yet.

Similar games

Per-tag lists of released games come from public Steam data (refreshed at most weekly). A candidate has at least the given number of filter tags; tags, reviews, release date and price are fetched in batches of 50 games (with more candidates than the limit we take an even sample). A similar games member is a candidate with tag similarity at or above the threshold: weighted Jaccard where a game's first 5 tags weigh 2 and the rest 1 (sum of the smaller weights of shared tags / sum of the larger weights of all tags). The default threshold is 30% – at 40% only about 5% of candidates would remain. Release years must match and the price must be within ±50% of your game's list price (no band for a game without a price). You can add or exclude games by hand; your games of interest are shown next to similar games.

“Where you stand” is a percentile: the share of similar games with a lower value (reviews, weighted rating, sales estimate, price). Active competition = games with at least 10 reviews; supply = releases in the last 12 months against the 12 months before; concentration = the 3 best-selling games' share of similar games's estimated units. Public Steam data doesn't include followers or peak players, so we don't compare them yet.

Store funnel

From the Steamworks traffic CSV we compute CTR = visits / impressions and page conversion = new wishlists in the period / visits, CTR also by traffic source. CTR under 2% points to the capsule and the name, conversion under 8% to the trailer, screenshots, description or a mismatch between expectation and page. We consider a CTR of 2–4% and a conversion of 8–12% typical; these are orientation ranges, not official Steam numbers.

From gross to net revenue

We show three levels: gross revenue, Steam net revenue (without VAT and refunds, as Steamworks reports it) and the developer's revenue after Steam's share: 30% up to $10M of a game's revenue, 25% from $10M to $50M, 20% above (marginal). US withholding tax (per the tax treaty and the W-8BEN form) is not deducted yet.

Engagement = (likes + comments + shares + saves) / reach; where a network doesn't return reach we divide by views, and where it has neither, by followers (a weaker metric, marked as such). Relative performance = a post's views / the median views of the last 20 posts on the same network; above 2× the post took off. Audience growth = net follower gain per week.

Personal Calendar

Since 23 Oct 2026 Steam shows players a personal calendar with wishlisted games that have an exact release date in the next 8 weeks. A game without an exact day (“Q1 2027”) isn't in the calendar, so we recommend announcing the exact date at least 8 weeks before launch. Nobody has measured yet how many wishlists the calendar adds or how it changes conversion – the conversion benchmarks come from games released before it.

What you keep

From Steam's net revenue (already without VAT and refunds) we subtract Steam's share by tier and the US withholding tax at the rate you enter – only from US revenue based on sales by country. We convert to Czech crowns at the Czech National Bank's daily rate.

Launch forecast

We calculate launch-day wishlists as today's count plus the average net gain of the last 7 days times the days to launch; then we add the gains of plan items before launch – festivals, the demo and Next Fest (see Game plan and “what if”). Then we simulate the conversion to first-week sales 10,000 times – around the median for your size and price, with the spread that really exists between games – and the first year as about 2.6× the first week. We show a pessimistic (10th percentile), middle and optimistic (90th percentile) scenario. After launch we compare where reality landed.

Game plan and “what if”

The Lab forecast comes from your game's plan: today's wishlists, last week's pace up to launch day and the gains of plan items that happen before launch. We estimate Next Fest from your wishlists before the fest (more when the demo is out at least two months earlier) and don't count it without a public demo. A demo adds wishlists from the number of players you expect and the usual demo player conversion. Post-launch discounts are already in the first-year estimate. What if shows the plan now and with the change side by side.

Festival or showcase: if you enter your own estimate, we use it. Otherwise we take the wishlist gains from your past festivals, and without any, a general range – selected games at free festivals gain hundreds to thousands of wishlists.

Price and number of buyers

We measure how price costs buyers on similar recent games with good ratings: we compare their estimated sales with price and account for the game's size (number of languages, early access), because pricier games tend to be bigger. When the data show that a higher price costs buyers, it shifts the forecast conversion; otherwise we only account for slightly lower conversion above 10 USD.

Competition calendar and launch date planner

We take upcoming games from Steam by your similar games's tags and keep only those similar to your game. Only some have an exact day – and only those are in players' Personal Calendars 8 weeks before launch. The planner rates weeks half a year ahead: a similar game in the same week counts three times, a game with an overlapping calendar window once. We skip weeks before your planned Next Fest ends and earlier than 8 weeks from now.

Players and peak players

Steam publicly shows only the current player count, so we measure it ourselves: your and interest games every hour during a fest and launch, newly released similar games every 3 hours, the rest once a day. The peak is the highest count we measured – the real one is usually higher. For games up to 3 months after launch we estimate first-week sales from it (usually 8–22× the peak).

Genre supply and demand

We take released similar games from the Market comparison filter, estimate sales from reviews as a range and revenue as units × list price (an upper bound – real revenue is lower due to discounts, VAT and Steam's share). The market level is our own rating: saturated when releases grow fast and few games pass 10,000 USD; hard when few games break through or hits take the sales; a gap when games break through and there are few quality ones; otherwise open. Gaps are tags with few games that get at least twice the genre's median reviews.

Content creators

We find channels by videos about similar games and rate them by the median views of the last 10 videos older than 7 days (one viral video does not skew it) and the time since the last video. How many wishlists a creator brings we show only from your own data: from the effect of creator videos on your game's wishlists.

Page lab and tests

We compare your page with the top quarter of similar games by reviews and with the rules of a good page. Every day we save a fingerprint of the public page and log the changes. An experiment compares the visit → wishlist conversion a week before and after the change and warns when something else changed in the meantime. Capsule tests measure the click rate of variants; a test is decided once the difference is statistically significant and every variant has at least 100 views.

Tester surveys

For ratings we compute the average and distribution of 1–5 answers, for choices the share of each option. The weakest question has the lowest average. The comparison with the previous survey pairs questions with the same text.