Reverse Engineering the Balance Model

Before I could fix anything, I needed to understand what was already there.

I started with a sample: two cards of each major type — effect-focused cards, attack cards, and defense cards. For attack and defense I chose the simplest, most direct examples in the game. From those I derived a baseline: how much "gold value" one point of attack was worth, and how much one point of defense was worth.

From there I expanded outward. Effect cards were more complex — an effect that hit one card was worth less per target than the same effect spread across multiple cards. A debuff had a different value than direct damage. I worked through each effect type with the simpler cards first, building up a rate per effect unit. When I encountered cards that fell outside the emerging curve, I flagged them and moved on — I needed the model before I could evaluate the outliers.

Once the base rates were stable, I validated progression: I leveled cards up, tracked how stats scaled with level and gold investment, and confirmed the curve was consistent. It was roughly consistent. That was enough to build on.

The final output was a balance spreadsheet that automated the model. You selected modular values — attack, defense, effect type, effect magnitude, number of targets — and it returned the expected gold cost of the card. Not a perfect system, but a defensible one. Before it existed, balance decisions were made by feel and revised after player complaints. After it existed, every new card was costed against the model before it went into the game. The process became predictable.

The model had limits. Cards with multiple overlapping effects or highly contextual power — the kind that felt weak in isolation but broke the game in specific combinations — couldn't be fully captured by a linear cost formula. For those, we made a deliberate choice: where a card needed to feel powerful, we accepted the imbalance and pushed the stronger effects to higher card levels, gating the problem behind progression rather than pretending the model solved it. It was an honest workaround, and it worked — as a tradeoff, holding itself within the constraints of soft launch.

With the model in place, we applied it to the full card roster in a single pass. The rebalancing identified significant outliers — cards undercosted or overcosted by meaningful margins. Correcting them reduced the volume of player complaints about broken cards. More importantly, it reduced iteration time — balance changes could be evaluated before reaching players rather than after. New cards designed from that point forward started from the spreadsheet, not from intuition.