V6 - Finding Our Way Back

It's been a busy few months for us, and honestly, that's because we've been doing a lot of thinking. Not about features or roadmaps, but about what Game Oracle is actually for.

When we first started building this, the goal was simple: help developers research a market, find a space where their idea could breathe, and walk away with a solid concept. Three steps. Somewhere along the way, particularly with Game Gap, we drifted. We got obsessed with metrics like undersaturation, profitability, and demand signals to the point where the tool started implicitly telling developers what they should make rather than helping them figure out where their idea fits. That's not what we want to be. A game made purely because a spreadsheet said it would sell is rarely a game worth making.

So v6 is, more than anything else, a course correction.

The Big Stuff: Game Gap Gets a Rethink

The centrepiece of this release is a full redesign of Game Gap. The old version was built on a fairly blunt clustering system, and it showed. The new version is powered by a Market Segments algorithm using a custom built recursive graph clustering method. With this method we've generated around 17,000 overlapping segments that much better reflect how the Steam catalogue actually organises itself. Games don't belong to one neat genre box, and neither do market opportunities.

Each segment now gets its own dedicated page with detailed overviews, player sentiment breakdowns, developer fit assessments, and a launch efficiency section that shows wishlist-to-sales conversion data with honest sample-size reporting. If there isn't enough data to say something meaningful, it says so. We'd rather admit uncertainty than give you a confident number that means nothing.

We've also added a proper warning system. Things like Goliath's Playground (a segment dominated by huge studios), Stampede (too many releases too fast), and Old But Gold (segments where the classics still rule). These aren't flags telling you to avoid a market, they're context for you to weigh up yourself.

And crucially, you can now actually bring your own constraints. Art style preferences, genre affinities, or simply sales goals. Our goal with Game Gap is to work with what you care about, not crowbar your creativity into whatever trend is hot right now.

Image Search

This one has been on the wishlist for a long time. You can now upload screenshots or concept art and search Steam for visually similar games. It sounds simple, but it required building out a whole serverless model warmup pipeline to keep response times reasonable. With the current demand for GPU compute this was no easy task. Probably the toughest part was reducing our model size without compromising accuracy.

Luckily for us the tooling around vision transformers and AI is better than ever. There are also reliable serverless solutions for this exact use case now and we're capitalising on it to offer the best possible experience. With these new tools, not only can you search the Steam market using a game concept in plain english, but you can also provide a screenshot of a game that inspires you or add some concept art, and this will be incorporated to steer your search results.

It's available on our paid tier and we're genuinely excited to see how people use it. We suspect it'll be particularly useful for those who have a visual direction in mind before they have a concept fully formed.

The Flow Problem

One thing we noticed was that people would do great research in Data Explorer or Steam Map, then open Game Gap from scratch with no context. That friction was killing the experience. In v6, you can carry your research directly through your market research pipeline. Within Data Explorer we have algorithms working in the background, analysing your search results, and identifying market opportunities within Game Gap for you. 

You can then take the opportunities that fit your expertise and carry them straight into Concept Compass, solidifying your ideas and surfacing a detailed picture of where your game fits in the market and how to set realistic goals around its performance. It's the kind of thing that sounds obvious in hindsight, but it has taken a lot of iterations for us to get here.

Under the Hood

The PR diff for v6 is intimidating. We rewrote a lot of the service layer (consolidating 20+ scattered fetch calls into a centralised client), migrated fully to Svelte 5 runes (cleaning up a lot of old code that desperately needed attention), and knocked out over 1,000 unit tests, including new Playwright end-to-end coverage. We also reviewed our new glass panel design system to bring more visual consistency across the tools, which has been long overdue.

After all this work, we also needed to do a full documentation audit. Our internal architecture docs had drifted badly out of sync with reality, a problem that compounds quietly until it doesn't. That's fixed now. A task like this would usually take weeks and make any developer want to give up, but thankfully this is one area coding agents really shine, and I've been really impressed with the quality of work that can be achieved with Codex and Claude. Don't get me wrong, there is still a lot of hand holding and frustrated prompting back-and-forth, but I can't deny how much it has helped deliver this release.

One more thing: the Steam Map has been refreshed with improved data cleaning and a new generation of embeddings, so search results should feel noticeably more accurate. The updates have also been propagated through to the three new trend map layers we introduced in v5.2 — Supply, Demand, and Opportunity — so you can get a higher-level read on the market before drilling down with the new Game Gap tool.

What We Didn't Ship (And Why)

We want to be upfront about a few things that didn't make it:

• Community data to help identify abandoned games: still something we care about deeply, but Steam's community API rate limits make this genuinely hard to do at scale without it being unreliable. We'd rather release it properly in a smaller dedicated update than bolt it on and have it half-work.
• Capsule Art Analysis: this is part of a longer post-concept phase we're planning. It'll come, just not yet.
• Trend Watch: honestly, this dropped down the priority list. If we see renewed interest we'll revisit it.
• Improved sales & wishlist estimates: we're still hunting for data good enough to validate a custom model here. We're not going to publish estimates we don't trust.

What's Next?

The v6 work has surfaced a few things we want to build on. The Market Segments system is powerful but we've only scratched the surface of what it can unearth. We want to make gap discovery feel more like an intelligent collaborator and less like a filtered table. The flow between tools is better but not perfect.

We're also going to be taking some time to work on our data infrastructure, increasing the depth of data we have, improving reliability, and seeing how we can expand on community data and better estimates. For now though, v6 is out. We hope it feels like a step back towards what Game Oracle was always meant to be. As always, if you have thoughts, feature requests, or you just want to tell us something doesn't work, drop us an e-mail or pop it on the Discord if you have an account.

Thanks for sticking with us. 🙏