This update focused on improving the autoplay AI’s combat decisions, investigating its balance-test results, and polishing a couple of presentation details. It also brought up a broader design question: how much variety do the current level generators really provide?

Smarter Ranged Combat The autoplay bot needed to be more aware of the limitations of ranged attacks. It now assesses how inaccurate a long-range attack is likely to be and move closer when the shot isn’t reliable enough. It also recognizes when an attack cannot reach its target because an obstacle is in the way. Rather than repeatedly attempting an impossible shot, it repositions to find a better angle or get around the obstruction.

Reconsidering the Balance Results A 30-run balance check initially reported a 77% win rate, with no stuck runs and no errors. That looked considerably higher than the intended 50% target, but the sample was too small to justify changing the game’s balance.

A larger 300-run test puts the bot’s win rate at around 45%. With the statistical margin of error, 50% is still within the expected range. There’s no reason to retune the difficulty based on the original result.

I also investigated whether the bot handles merchants during automated tests. It does. However, it was getting stuck in the merchant area during actual auto-gameplay.

Victory Screen and 1-Bit Presentation The victory screen needed the same mobile treatment as the game-over screen, rather than remaining a desktop-oriented screen.

More Variety in Level Generation

Finally, I want to revisit the level generators. We need more variety in the environments and layouts players encounter, but before introducing new generators, the first step is to establish exactly which generators are currently in use.

That inventory will give us a clearer picture of what we already have, where the generators overlap, and which kinds of levels are still missing.