Frequently Asked Questions
The questions the community asks most, in one place.
Why aren't all heroes available in the draft tool?
The draft AI is trained on the most-played heroes in ranked RTA, currently around 120 of them, which together make up the large majority of real matches. Keeping the pool focused is what keeps the predictions sharp; very rarely-picked heroes are left out because there isn't enough data to model them reliably. You can still browse every hero's numbers on the Hero Stats page, even ones that aren't draftable.
What if my opponent picks a hero that isn't in the pool?
Right now the tool can only reason about heroes in its trained pool, so if an opponent locks in an off-meta pick you won't be able to enter that exact hero. This is a known limitation. The closest workaround is to pick a similar unit to gauge the matchup. As more data comes in, the pool grows.
What does the win rate percentage mean, and how are the suggestions ranked?
The percentage is the model's predicted chance that your side wins the current matchup, and it updates as picks and bans are locked in. The suggested heroes aren't shown with individual percentages; they're ranked best-first. The brightly glowing gold hero is the model's top pick for that slot, and the dimmer ones are alternatives it rates nearly as strong. So the highlight tells you how good a pick is relative to the best option, not a separate win rate per hero.
Does the AI account for gear, sets, artifacts, or speed?
The model doesn't have access to your exact in-game stats like speed or gear. It learns from real draft outcomes at scale, so it captures the typical build and role of each hero implicitly, but it can't see how your specific unit is geared. Adding a speed signal is something we're exploring.
Why does the AI rarely recommend some units that top players like?
The recommendations reflect large-scale win and pick data across many players, not any single pro's opinion. A unit that top players value highly can still be under-recommended if the broader data doesn't back it as strongly. Cases like this are exactly what we look at when improving the model.
How often is the model updated?
The model is retrained after balance patches and new hero releases. Recompiling the dataset and retraining takes a few days, so there's usually a short gap right after a patch. New models are announced in the Updates section and on Discord.
Is it free? What does supporting add?
The core tools, live draft, quick analysis, and hero stats, are free. An optional supporter tier unlocks a few extra perks and helps cover development and server costs. Nothing essential is paywalled.
Will you add team or lineup suggestions?
Placement suggestions are live. After the post-ban phase, Live Draft suggests where to position each hero on your team, learned from how players actually place each comp: spots the model is confident about pulse and are marked Recommended, and the rest are Flexible so you can arrange them how you like. Suggesting a whole team composition from scratch is a different problem, and that isn't something the tool does today.
Is this affiliated with Epic Seven or Smilegate?
No. E7 RTA Bot is an independent, community-made fan project. Epic Seven is developed by Super Creative and published by Smilegate Megaport, and all game names and assets belong to their owners.