Welcome back to Between the Lanes, a blog feature where members of our development team walk through some of the challenges, bugfixes, and occasional happy accidents we encounter while working on a game as unique as Dota.
During an event as massive as Crownfall, we got to try a lot of experiments — overworld maps, branching storylines, hero-themed tokens, minigames, comics, pixel art, and other ideas we’d been eager to explore for years.
What gave us the confidence to take these risks — and something we take constant advantage of at Valve — is the freedom to iterate: to gather real player data, let it challenge our assumptions, and even change our minds. We often say we don’t fully understand what we’re building until we see players interact with it, so we try to get to that point as quickly as possible.
That mindset also shapes when we make decisions — usually as late in development as possible, when we have the best (and sometimes only) data. At every step, we rely on actual player experiences to help us focus on what resonates, and steer away from what doesn't.
In this edition of Between the Lanes, we'll walk through the development of something that came online relatively late in Crownfall — the Act IV boss battle between our heroes and Queen Imperia, Nest of Thorns. We're also using it as a case study to show how playtesting, and the data it generates, shape our decision-making.
Getting to Imperia
The first (and only) minigame in Act I was a modest fishing minigame that had Tidehunter testing your angling skills in the hopes of catching him breakfast. A lot of Act I was like this — small ideas we'd scattered throughout the map to see what players would engage with. The encouraging player data from Act I gave us the confidence to get more adventurous in Act II, and by Act III we were shipping Sleet Fighter, a fighting game brawler at Tusk's favorite dive bar; Dragon Chess, a match-three battle of wits against Winter Wyvern; and Zaug's Lair, a shoot 'em up pitting an evil dragon against Vengeful Spirit through Icewrack's twisting cave system.
By the time we reached Act IV, we had a solid sense of what worked and what didn’t. What kind of data? Everything from the concrete — how many players replayed each minigame, and how often — to the more interpretive, like user-submitted feedback and discussions online. That clarity gave us the confidence to go all-in on a blow-the-doors-off finale: an epic boss battle against the evil Queen Imperia.
Of course, deciding what kind of game the boss battle should be wasn't easy. We knew it needed to deliver the high-stakes spectacle players were expecting. But it also had to offer enough variety and replayability to stay fresh after multiple runs. And because this was a mini-game, designed to be played while queuing for a match, it needed to be intense but brief. It should feel like a tasty snack, not a long meal.
To strike the right balance, we needed a genre that combined the high production values of Sleet Fighter with the replayability of Dragon Chess. After all, this would be the epic showdown between Vengeful Spirit, Skywrath Mage, and newcomer Kez (our "big three" from Crownfall's four-act story) storming Imperia's castle. We wanted the player to feel like they were up against impossible odds in their quest to reach the queen.
Hatching a Plan
One idea that briefly gained some traction was a turn-based JRPG-style battle. But it didn't take long to realize it wasn't the right fit. We needed something that would throw players into the action quickly — something short, intense, and fun enough to play between Dota queues.
That's where the survivors genre came in. A "reverse bullet hell" with endless waves of enemies, it was the perfect vessel for an epic-feeling, high-stakes boss battle. The player would hack their way through hordes of enemies, while using a little strategy to power up various abilities in preparation for a final showdown with Imperia. It would also be easy to map existing Dota items and abilities to the action, so players would already have a sense of how things worked.
Since we wanted the experience to be relatively brief, replayability was another big piece of the puzzle. The survivors genre gave us a lot of variability, since it's never the same experience twice. Players aren't guaranteed to get the items or events they're expecting in the exact order they want. The variability encourages you to keep trying, starting from scratch in a way that feels challenging without feeling insurmountable or frustrating. We didn't necessarily want you to win the first time... but we also wanted you to keep playing when you lost.
Early Prototypes
The earliest prototypes of Nest of Thorns started with the same core elements as the version we eventually shipped. A player picked one of three heroes, faced wave after wave of enemies, and eventually got to Imperia. What these early attempts didn't have was anything remotely fun. Sometimes a hero wandered a mostly empty map, bumping into the occasional skeleton. Other times the waves of enemies were so overwhelming it wasn't actually possible to win (or so we thought; more on that later). We knew there was a fun game in here somewhere. We just had to keep torturing our playtesters with bad builds until we found it.
Every Valve employee is a guinea pig, and putting someone in front of a build is how we figure out if something's working (or, just as often, not working). We can brainstorm forever, but until someone sits down and plays, we don't really know if the choices we've made are landing.
At first, playtesters struggled to survive the first 3-6 minutes. Usually, that's a sign that the game is either unbalanced or not explaining its rules well enough. We'd sit down with them and have them walk us through their experience — moments where they weren't sure what to do next, or where it got so overwhelming it felt impossible, or even where they just felt bored. Then we'd go back to our desks and make some changes, and put it in front of a new playtester. Slowly, we started to see progress.
And then something happened.
After walking us through their first run, a playtester who'd just lost asked, "Actually, can I try again?"
That was new, and encouraging. Nest of Thorns was still nowhere in the neighborhood of good yet — it usually took at least three runs before someone figured out how to beat our prototype — but it was at least nice that, for once, we weren't making coworkers play our crappy prototype — they were actually asking to play our crappy prototype a second time. It was a teeny tiny flicker of interest: "I don't absolutely hate this." We were happy to take it.
But the real eureka moment was still to come.
Hero Builds
One key aspect of the survivors genre our early versions didn't have was constant pressure. Initially, we'd just spawn a wave of a hundred skeletons into the level and let the hero hack away at them. Once the hero'd hacked them down to zero, we'd spawn in the next wave of baddies at a harder difficulty. Playtesters got bored of this quickly. Once they figured out the only goal was endlessly hacking at bad guys, it started to feel like a chore. The only reward for killing them more efficiently was that you got to fight the next, harder wave sooner. (Yay?)
Eventually we figured out that when we spawned a hundred skeletons in and the hero started hacking away, we needed to put that wave on a timer and keep topping the skeletons back up no matter how many the hero killed. Once the timer went to zero, we'd spawn in a harder wave of enemies on a new timer. Now the pressure never let up, and the player stayed engaged.
But more importantly, their goals changed. When our hero was just slaughtering a dwindling number of skeletons, the goal was "Make number of skeletons go to zero." But if the number of skeletons never goes down no matter how many you kill, then it doesn't take long for players to realize killing them isn't actually the goal.
Instead, they turned their focus to what they were getting from killing all those skeletons — in this case, abilities and upgrades. Suddenly the goal wasn't "Kill all skeletons," it was "How many of these things can I kill as fast as possible to get as powerful as possible?" Instead of feeling like a chore, harvesting waves of enemies started to feel like chasing a jackpot — with every kill bringing you closer to the perfect build.
Because progressing to the next wave was no longer tied to killing everything from the last one, we'd also gained something new: a shared clock. This meant players were now sharing the same rhythm — a common language for when things started to get hard. So when one player said "I can't stand that wave of spiders at minute four," everyone else instantly knew what they meant (those spiders are the worst).
We'd finally found the fun. This small shift in focus helped clarify what players were enjoying and why, and kicked off a cascade of new ideas — more powerful upgrades, crazier abilities, and build possibilities we hadn't even considered. This meant playtesters could make radically different hero builds from each other in a single playthrough. And that meant they were more engaged during playtests, giving us better feedback and suggestions to feed into the next build.
Well, if we asked more than once. After a playtest, we'd sometimes find that coworkers no longer wanted to talk to us about their highs and lows. They wanted to talk to each other about their builds, and argue about which one was better.
As build variety kept growing, we started wondering if we'd missed a big piece of the puzzle — what if the heroes themselves felt just as unique as the abilities? In early prototypes, other than their starting ability, the player experience was largely identical no matter which hero you picked. Now we were tweaking them into truly unique choices, making one stronger, another faster, another better at magic damage — which opened up the variability of the builds even more. Picking one hero over another also meant opening up specific branches of abilities you might not otherwise get to build towards. (And because there's always a bit of luck involved in which upgrades you get, every choice felt like a high-stakes gamble.)
The First Six Minutes
Meanwhile, while you're distracted figuring out which upgrade branch is going to get you to Imperia, we're slowly ratcheting up the difficulty. At some point, without you even noticing, the waves of enemies have gotten deadly, the pace is relentless, and you can barely keep your head above water.
At least, that was the goal. With the upgrades coming along, we moved onto the pace of the gameplay. We wanted the experience to be intense, but not overstay its welcome. (We knew most players would be playing while they were queued for a match.) On the flip side, though, we wanted to leave enough breathing room for players to enjoy watching their upgrade choices stack.
Our back-of-the-napkin math gave us about twelve minutes of total gameplay — not counting the three-minute Imperia fight at the end. So we got to work on pacing the first six minutes. How often do we give the player an upgrade choice? How early can they start stacking abilities? How much damage should each ability do where it feels satisfying but not overpowered?
Because we were tuning the gameplay in sections, focusing on the first six minutes meant the final six minutes were essentially unplayable at this point. We'd playtest people on the first six minutes and then make them stop. Technically they could keep playing, but everything after minute six was placeholder. You weren't meant to play past that. (We actually begged our coworkers not to.)
One funny side effect of making the abilities and upgrades as fun as they were turning into was that playtesters often didn't care that half the game was still unplayable. We'd sit at a coworker's desk, watching them play through the first half of the game. Once they hit the six-minute mark:
"Great, thank you," we'd say. "Can you start the game over?"
A pause.
"Actually, hold on, I think I can beat it."
We'd remind them the last six minutes was a punishing unbalanced wasteland of copy-pasted enemies and was currently unbeatable. They would say they understood. We would interview them for their highs and lows. They'd give us lots of great feedback on their playthrough of the first six minutes. And then we would go back to our desks, and they would watch us go, and then unpause the game and try to beat it anyway.
Some of our coworkers took it as a personal affront that they couldn't get through the last ferociously unfair, literally impossible-to-beat six minutes. So while we were designing the first half of the game, they were pooling their resources slamming their heads against the unbeatable second half.
And here's the thing: They beat it.
Eventually they figured out the precise combination of overpowered abilities that we hadn't fine-tuned yet, and beat the part of the game that wasn't meant to be beaten. (Which tells you a lot about game design — but maybe even more about the kind of people who get into game design.)
Controlled Chaos
By the end of development, the core game was built. Abilities, mini-boss encounters, and the final showdown with Queen Imperia were all in place. At this point, we were no longer looking for game-breaking issues, but ways to adjust, balance and polish to a shine what we already knew was working. We kept playtesting, but had dialed back the exhaustive post-mortem interviews and were just letting people play for fun (and, of course, to generate more data).
While the quantitative data that came from every run was important, the qualitative data coming from playtest games was no less important. In one memorable example, one playtest ended epically with the Wailing Mass — internally named 'Meatball' — taking over half the screen and devouring the player. At this point the playtester had an audience of coworkers watching behind him, all laughing at his misfortune. Of course, he immediately pressed the "play again" button.
We were confident we had something worth shipping, and had opened up playtesting from people who had volunteered to playtest, to basically anyone we could corner at the company and rope into a game. In our earliest prototypes, we'd get treated to an 11PM Slack message saying "The game is too difficult." By this stage, we'd still get those 11PM slacks, but then the 3AM follow-up of "Never mind, I beat it."
Even with a mostly finished game, we were still working on Nest of Thorns right up until we shipped Act IV of Crownfall. Way back at the beginning of this post, we said that we can make better decisions when we have good data to support them. Well, now we had some: With internal playtesters as excited as they were to play (and replay) the game, we had a ton of data to know which parts of what we'd built were resonating. And nothing gets a room full of game devs excited like the moment when a game starts being fun. We wanted to do more of the things we now knew were working, as much as we could possibly fit in. New items, abilities, lore and even waves — every time we thought we'd nailed the game down, someone would inevitably come up with one more 'just one more' idea. And because it was usually a pretty great idea, we couldn't resist trying it.
It wasn't just that feature ideas kept coming in: they kept getting better the closer we got to release. In the early days, when things were more amorphous, it was hard to distinguish between a potentially high-value feature and an actually low-value feature. By the time we were a week out, the best places to put our resources and time to make the game better were obvious.
Leaving the Nest
As we put the finishing touches on Nest of Thorns, we were able to step back and look at it with fresh eyes. The challenge had been to use things we'd learned from prior acts to build a suitably epic boss battle to close out a massive event. We were pretty happy with what we'd pulled off, especially given the time constraints. We'd found the fun.
Unfortunately, the community hadn't found it yet. What we hadn't considered was that, since Nest of Thorns was a boss battle, logically it had to live at... well, where the boss lives, at the end of the game. This meant at least a day or two of quiet torture as we waited for Dota's power players to grind their way through Act IV at inhuman speeds, waiting to see when players would finally get a chance to play it.
It turns out we didn't have long to wait. We thought we'd have at least a day or two before anyone reached Queen Imperia's throne room. Turns out we'd severely underestimated how fast Dota players can haul ass through content. Just sixteen hours after launch — yes, hours — the first power players had already fought their way through the entire last act and were getting their first taste of Nest of Thorns.
And eventually everyone else caught up too. As of this writing, Nest of Thorns has been played nearly a hundred million times. (If you haven't caught up with it yourself, or just want to try your hand at it one more time, it's still playable in the Crownfall Archive in-client.)
Embracing Uncertainty
We hope this behind-the-scenes look at the Dota team getting a minigame up on its feet gave you a better idea of our process — and didn't just make you stare at this blog post in horror with the revelation that we're making it all up as we go along. But really, we hope it proves that game design isn't some mystical art by game-making geniuses steepling their fingers in dark rooms, but an iterative process of trial and error driven by data and anxiety and playtesting and failure and playtesting and more failure and more playtesting.
Game development thrives on iteration — what starts as "no idea" slowly turns into "some idea" by looking at past ideas that worked. Then it turns into a bad prototype that nobody wants to play but we make them anyway. Then we watch them not having fun playing it, and go off and fix all the not having fun parts, until eventually we're watching people have fun playing it. Then we polish it and polish it and look at the ship calendar and weep. Valve's ethos has always been to embrace uncertainty and allow data to shape development. It means we don't always ship things on time. But we think you're going to like them when they're ready.
