Ch.11: RTS Game Design: Why Perfect Balance Is Impossible

0:00 Patch day in an RTS. One line in the notes, three players reading it. The first begged for this nerf for months — pure relief. The second just watched a favorite opening die. And the 3rd one says nothing, because she's already in a custom lobby poking at the change, hunting for whatever just became the new strongest thing. Three reactions to one sentence: fix one, break two. So why does every patch do that — fix the thing you hated, break something you liked? Because "just balance the game" assumes balance is one dial you can set correctly.

0:35 And it isn't — once you swap in the real model, patch notes stop reading like developer malice. That sets up today's whole job: replace the dial with the actual picture. So let's start with the hidden model behind "just balance it" — the one I carried for years, and honestly the one most people assume: every faction sits near 50 percent win rate, so the game is balanced, so we're done. If the numbers are even, what's left to fix? Take a fake game — three factions, X, Y, and Z, hypothetical on purpose.

1:09 X sits at exactly 50 percent overall. Sounds balanced to me. Except X stomps Y and folds to Z, one map is nearly a free win, and past a certain league the faction just stops working. The average is perfect. The experience is miserable. So the 50 didn't measure balance. It sort of averaged the misery away. That's what aggregates do. Balance isn't one score — it's a surface: faction against faction, across skill levels, maps, strategies, even phases of a single match. Which means a patch never moves one number — it reshapes a graph of matchups.

1:46 Pull one edge toward fair, and every edge touching it moves too. That surface hides one more axis — the one tier lists never mention: skill. The same unit can be two completely different units depending on who's holding it. Meaning what — the same stats play differently? Same stats, different game. StarCraft 2's balance council hit this with the Disruptor — the Protoss unit built around one big, slow, dodgeable shot. Their own public-test notes — linked below — lay out the tension. At the very top, pros had learned to play around it so well that, after earlier nerfs, it barely earned its cost anymore.

2:22 Below that level, one connected hit was still, Ruining somebody's whole afternoon. So the same unit ends up... ...useless up top, and a menace down the ladder. Two broken halves making a whole. So who gets the fix — the pro who deserves an expressive tool, or the ladder player eating the worst-case hit? The council's proposal refused to pick: restore some reliability for high-level control, and soften the catastrophic worst case for everyone else. Whether it landed is a live-game question — the notes only promise the intent.

2:56 Either way the graph just grew an axis. So "is this unit fair" isn't even a question yet — not until you say for whom. Same graph, narrower target — sometimes the patch is aimed at one strategy, not a matchup. For example, way back in an early StarCraft 2 patch, the team delayed a single research timer to hit one specific Blink all-in while leaving every other use of Blink alone. That was the entire stated goal — kill the rush, break nothing else. A surgical cut instead of a rework. And compensation is the other half of the craft: when a change would gut a unit, the notes hand strength back somewhere healthier.

3:35 Like one heavy unit getting cheaper but attacking slower — easier to field, less damage over time. Or a long-range unit losing maximum reach but responding faster — more control, more exposure. Doesn't that read like the fix failing halfway, though — nerf the thing, then hand the power right back? I mean, it reads like the design finishing the job: keep the unit's promise, remove its worst expression. Narrow the change, pay back the loss — that shrinks the blast radius instead of pretending it's zero.

4:05 Every lever touches other matchups and other phases. You can aim a change; you can't aim its whole shadow. Now the next patch story is the brutally honest one. The council had redesigned the Cyclone — and by their own stated goal, it worked. The unit went from barely seen to showing up in every matchup. Mission accomplished, right? Hold on. If it hit the goal, this wouldn't be a story — something else must have bent. What was it? Protoss early game. The follow-up test notes said the new Cyclone was squeezing out early build variety and creating defensive trouble in that matchup.

4:41 So the unit was finally fine — and everything around it got worse. Worse enough that the council proposed reverting the entire redesign. Oh, that's the twist: the fix worked, and they still want it gone — because it was removing strategy from the strategy game. That's the sentence to keep — and the bigger one behind it: a fix can hit its target and still damage the game around it. The result is a harder bar for every patch: watch the whole graph, not just the target. Same idea from the opposite direction in Age of Empires 4: a unit that looked fine on any scoreboard — imagine the forum post: "win rate's fine, what's the problem?" — and still bent the game.

5:24 The Springald, the bolt-thrower that had quietly become a do-everything answer. The fix wasn't a number. The team rewrote its job description: you are the anti-siege specialist now. Great at that, worse at everything else. Right down to the job description — and I think that's the part people miss about fairness. Make everything identical and you delete the reason faction choice matters at all. Nobody's picking a faction for its character then. And the same studio said it straight out later: civilization-unique units should feel like exciting perks, not liabilities.

5:58 Asymmetry isn't the bug — unpriced asymmetry is. So a healthy identity is a loud strength with a legible bill: one clear job, a real cost, a known weakness, a counter your opponent can actually find. That gives "fair" a better definition than "identical." Those bent edges leave one question: how do teams even see one? My guess would've been win rates — which, after this chapter, worries me. Better than that. StarCraft 2's old balance lead once listed it: skill-adjusted ladder data, tournament results, feedback from pros, feedback from everyone else.

6:32 Four different lenses, no single scoreboard. Not even one per faction — it was split by league, by region, by matchup, by map. And they tolerated small map-specific imbalances on purpose, when those made strategies more interesting — map vetoes as the pressure valve. Think about what that admits: the report's own matchup rates wobbled week to week and differed by league. And any engineer who's chased a dashboard in production knows the failure mode — optimize the one visible metric, and you can damage everything it doesn't measure.

7:04 So metrics locate pressure; they don't choose the game. Now we can see the full loop: evidence in, judgment out. And once you see that loop, a patch note reads completely differently. It's a hypothesis: here's the problem we see, here's the bounded change, here's what should stay viable — now let's test it. The test bites back. Bites back how? That same public-test cycle put the proposals in front of real games and player feedback, and the council cut or rewrote several of its own changes — the Cyclone revert included — before the rollout to the live game.

7:39 Which is where trust actually comes from, I think. Not from getting it right the first time — nobody does. From stating intent you can check, and reversing in public — rather than quietly — when the graph says you missed. So take that whole shape back to patch day and our three players. The next balance note you read — any game — deserves four questions instead of one verdict. Alright, give me the four. First: name the target problem. Second: identify the affected players — which skill level, which matchup?

8:10 3rd: find the counterplay that's supposed to survive. 4th: spot the new tradeoff entering the graph. Let's say you're reading the Springald note. The problem: one unit answering everything. For whom: anyone stuck fighting siege wars. Surviving counterplay: siege still dies — to a dedicated hunter. New tradeoff: you carry a specialist now, not a crutch. Four answers, no mystery — a patch you can argue with instead of just yelling at. That's the whole tool. And keep the picture: a balance patch doesn't move one number — it redraws the matchup graph.

8:43 Uneven on purpose, legible by design. Perfect balance was never the goal; a game full of real choices is. Thanks for listening to Learning Podcasts.