DEV Community: Indie Tooling

Your Steam wishlists may be worth half what they were

Indie Tooling — Wed, 14 Oct 2026 00:00:00 +0000

A Steam wishlist may buy you fewer launch sales than it used to. In a new analysis of his reader survey, Chris Zukowski of How To Market A Game found that games launched from 2023 to 2025 sold about half as many units in their first month, per launch wishlist, as games launched from 2020 to 2022. He adds that the range is wide, so treat “half” as a rough guide rather than a rule.

That matters because many of the rules of thumb solo developers use to plan a launch were set with older data. Zukowski’s own benchmarks page, built on a 2020 study attributed to Simon Carless, puts median first-week sales at 15% to 25% of launch wishlists, depending on how many wishlists a game has. In a 2020 GDC talk, publisher Mike Rose suggested multiplying wishlists by 20% to 30% to estimate first-week sales. If your forecast leans on numbers like these, it was built for a different market.

The data, briefly

Zukowski looked at 160 Steam games from his sales survey, where developers upload their own Steamworks reports. Of those, 34 launched between 2020 and 2022, and 126 between 2023 and 2025. He left out 2026 launches because they haven’t had as many months on sale yet.

The measure is simple: units sold in the first month after launch, divided by wishlists at launch.

There are three caveats. The data comes from developers who chose to submit it. The earlier group is small, at 34 games. And individual games vary a lot, so your result could sit well above or below the average.

Why it might be happening

Zukowski calls it wishlist inflation. Since about 2020, developers have chased wishlists harder through Steam Next Fest and third-party festivals. He argues that makes the average wishlist cheaper to get and less likely to turn into a sale. He points to other factors too:

Valve has made wishlisting easier, so for some players a wishlist works more like a bookmark.
Festivals spread wishlists across more games, and he suspects festival wishlists convert worse.
More developers now know wishlists matter, and more of them take part in Next Fest.

Steam is also more crowded than it was. SteamDB counts about 14,000 releases in 2023 and about 21,000 in 2025.

Not everyone sees the same decline. GameDiscoverCo’s study of 2024 and 2025 launches put median first-week sales at 0.17 times launch wishlists for games with more than 10,000 wishlists. It said it doesn’t think conversion is falling, just that more games are competing for the same players. The two studies measure different things: GameDiscoverCo uses its own estimates of first-week sales, while Zukowski uses developer-reported first-month sales. They agree on one point, though. Results vary hugely from game to game.

GameDiscoverCo’s study also hints at what separates the games that convert well from those that don’t. The worst converters had spent much longer on Steam before launch, 411 days on average against 214 for the best, and had lower early review scores, 67% positive against 91%. A store page that sits for years keeps collecting wishlists, but some of those players may have moved on by launch day.

Wishlists still predict sales

The good news in Zukowski’s data is that wishlists still rank games by sales almost as well as before. He measured this with R², a score from 0 to 1 for how closely one number tracks another, where 1 is a perfect match. On a log scale, total wishlists against sales scored 0.77 for games launched in 2020 to 2022, and 0.72 for 2023 to 2025. The two trend lines are nearly parallel. The newer one simply sits lower: each wishlist is worth less, but more wishlists still means more sales.

He also tested wishlist velocity, meaning how many wishlists a game added in the 30 days before launch. Across all 160 games, velocity alone scored 0.72, slightly below total wishlists at 0.73. Combining the two nudged it up to 0.75, a gap small enough that he puts it down to noise. His conclusion is that total wishlists are still the number to watch.

What to do with this

Re-run your launch forecast. If you planned with an older conversion rate, halve it and see whether your plan still works. Using illustrative numbers: if you expected 20,000 launch wishlists to turn into 4,000 first-month sales, plan for nearer 2,000, and check that still covers what you need.

Watch the total, not the daily graph. Recent velocity didn’t predict sales any better than the running total, so there’s no need to refresh your wishlist chart every morning.

Keep collecting. Zukowski’s advice is to keep gathering wishlists rather than give up on them, including the ones a Steam Next Fest demo brings in.

Be cautious with your launch date. If the revised forecast falls short of what you need, more time on your store page and another festival before launch may be worth it.

Add your data. The survey is still open, and every submission makes the numbers more reliable for everyone.

Should you care?

You are	Our verdict
Pre-launch, with a wishlist count	Re-forecast now with a lower conversion rate, but don’t overhaul your launch plan on one study alone.
Already launched	Compare your first-month sales per wishlist with these findings, and consider submitting your data.
Without a Steam page yet	Aim for more wishlists than the old targets suggested, and get your page up early.

Sources

AI can now decompile games. What that means for indies

Indie Tooling — Mon, 12 Oct 2026 00:00:00 +0000

AI coding agents can now turn a compiled game back into readable source code in weeks, a job that used to take hobbyist teams years. Two write-ups from the past two months show how far it has come, and publishers are already sending takedowns. Here’s what it means for small teams.

What happened

In August, developer Chris Lewis finished decompiling the Nintendo 64 game Snowboard Kids. That meant all 2,145 functions, rebuilt as C that compiles back to the original machine code. It took 84 days, with AI agents working in four parallel copies of the project. The same job on the sequel, Snowboard Kids 2, had taken him 596 days.

The same developer, two decompilation projects: 596 days for Snowboard Kids 2, 84 days for Snowboard Kids with AI agents. Figures: Chris Lewis

On 9 October, a developer known as momo5502 published a write-up of a bigger project: rebuilding a PC first-person shooter from its executable. The post doesn’t name the game, but TechSpot reports it’s Call of Duty: Modern Warfare 2 (2009). Over about three months, up to 16 agents used an estimated 600 to 700 billion tokens. The result has 99% of the game’s functions, 83% of them byte-for-byte identical to the original, and the game runs. The author is keeping the code private.

TechSpot’s roundup lists more. The community decompilation of Super Smash Bros. Melee reached 100% with AI help on the final stretch. GoldenEye 007’s reached 100% through human work, and AI has since helped turn it into a native PC port. A Halo: Combat Evolved project says it matches 99.5% of the original. Activision had footage of the Modern Warfare 2 work taken down. Separately, ZeniMax sent a cease-and-desist within days to an AI-built browser game set in the Fallout universe. That one was a recreation rather than a decompilation, but it shows how quickly publishers are moving.

Why it works now

This kind of project is called matching decompilation. You write source code, compile it, and check whether the output matches the original binary byte for byte. That check is automatic, which makes it a good job for AI agents. They can try, fail and retry for hours without a human marking their work.

Matching decompilation is a loop with an automatic referee: write C, compile it, compare the bytes, try again.

The agents still need supervision. momo5502 says they invented logic, used the wrong data layouts, and even tried to cheat by editing the script that checked their work. Lewis estimates his project might have stalled at around 90% without three human experts who knew the N64 compiler’s quirks. So it’s faster and cheaper, but you still need skilled people in the loop.

Is your game at risk?

For many indie games, the code was never hidden in the first place. Unity games built with the Mono backend ship .NET bytecode that free tools like ILSpy turn back into readable C# in minutes. Godot exports can be unpacked, and their GDScript recovered, with community tools such as GDRE Tools. If you’ve shipped on either, anyone curious could already read your code.

How much effort it takes to read a shipped build depends on how it was built. AI agents mostly change the right-hand end.

The change is at the harder end. Native code, such as C++ or Unity’s IL2CPP builds, used to take a lot of skilled time to reverse. AI agents make that much cheaper. They don’t make it free yet: momo5502’s run used two premium subscriptions and three months of close management.

Remember what decompiled code doesn’t give someone. It doesn’t come with your art, music, writing or name, and those are protected separately. Matching projects usually ship without assets and ask players to supply their own copy of the game. Cloning your gameplay never needed your source code anyway.

Will publishers switch to streaming?

Our take: probably not as a way to stop decompilation. If a game only runs on a server, nobody can decompile the client. But every streamed player needs a GPU in a data centre, and Google closed Stadia in January 2023. Handhelds like the Steam Deck and Switch 2 also need games that run on the device.

More likely, publishers will keep valuable logic on their servers, as online games already do, and keep using anti-tamper tools. Denuvo, for example, is reportedly seeking subpoenas to unmask a cracker. Expect more lawyers too. Hybrid set-ups, where your machine draws the game and a server runs the parts worth protecting, are a more realistic middle ground than full streaming.

The law, briefly

Decompiling a game doesn’t make the result yours. In the US, courts have allowed reverse engineering for compatibility (Sega v. Accolade, Sony v. Connectix), but that doesn’t extend to sharing someone else’s program. Breaking copy protection raises separate problems under the DMCA. The EU allows decompilation only when it’s needed for interoperability, and not to build a substantially similar program. So far, publishers have mostly responded with takedowns, as Activision and ZeniMax just did. This isn’t legal advice. Get some if you’re planning anything public.

What to do about it

Don’t hide secrets in your client. API keys, store credentials and admin passwords in a shipped build should be treated as public. Move them to a server.
Keep the logic that matters on a server. For multiplayer, leaderboards or in-game economies, let the server have the final say on results.
Know what obfuscation buys you. IL2CPP builds and obfuscators raise the cost of reading your code. They won’t stop a determined, well-funded team.
Protect what’s yours. Keep dated source history and art files, so a DMCA takedown is easy to back up if someone ships a copy.
Use the tools yourself. If you’ve lost the source to an old game, the same methods could help you recover it and port it to new platforms. Back up your source so you never need to.

Sources