Hey everyone, I’m working on a farming sim in Godot and struggling with balancing my AI characters. I want their decisions to feel realistic but not too predictable. I’ve set up basic probability curves for their daily tasks, but playtesters say the patterns become obvious after a few in-game weeks. Has anyone successfully used community feedback to tweak NPC behavior trees? Specifically looking for methods that:
- Capture unexpected player observations (“The blacksmith never opens shop during rain even though she has a roof?”)
- Translate qualitative feedback into numerical adjustments
- Avoid overcorrecting based on outlier opinions
I’m curious if others have tried A/B testing different probability sets with small player groups. Any Godot-specific tips for implementing this kind of iterative system would be awesome!
I struggled with similar issues in my village sim! What helped was creating a simple feedback form asking players to rate NPC ‘surprise vs predictability’ on a scale after each play session. I used those ratings to tweak my probability weights weekly. Maybe start small with 2-3 key NPCs first?
Welcome to the community! The Developer Resources section has a great template for categorizing playtest feedback into ‘behavior bugs’ vs ‘immersion breaks’ that might help prioritize what to adjust first. Have you tried logging NPC decision data to visualize patterns?
Implement a CSV import system for probability curves. Use Godot’s FileServer to hot-swap different configurations during playtests without rebuilding. Document version differences clearly for testers.
Dude make the blacksmith randomly take ‘mental health days’ - players will think it’s lore and stop complaining about patterns. Add some empty wine bottles model near her anvil as visual storytelling lol
Don’t overengineer early. Use a weighted random system where ‘surprise’ behaviors have 5-10% base chance, then adjust based on playtest upvotes. Pro tip: Separate your core behavior logic from probability tables so you can iterate faster.