Game Feel

Swink’s 2008 book formalized game feel as three things: real-time control, simulated space, and polish. (“Real-time” here means the loop between your input and the game’s answer stays under about 100 milliseconds.) The juice talks of 2012–2013 taught a generation the third leg and let them mistake it for the whole body. It’s worth reading the triad again, because polish is the only part you can see in a GIF. The other two are the parts that take a career.

When I work a feel problem with a team, I cut it into three buckets you can actually assign owners to: stick feel, audio, animation. Does the avatar answer the hand? Does the sound tell the truth about the impact? Does the animation commit to weight, or honor the next input? Most feel bugs I run into live in a disagreement between those three, not a deficiency in any one of them.

Then try running the same triad up the timescale. The lock-in sound in champ select is stick feel for a menu, the rank-up splash is animation for a season, and the chest-opening reveal is audio for a wallet. I don’t mean any of that as metaphor: it’s the same correction loop between what the player did and what the game says back, just stretched over longer intervals.

Swink was writing about avatars, but I’d push it further than he did: it’s also the first, last, and most common thing in your queue, your storefront, and your ban phase, at every timescale you ship.

“Juice” started as prototyping economics. Kyle Gabler and his crew put it this way in 2005: “A juicy game element will bounce and wiggle and squirt and make a little noise when you touch it.” Then “Juice It or Lose It” (2012) and “The Art of Screenshake” (2013) turned a trick into an ideology, and for a decade the path to good feel was thirty toggles deep. The talks were great! The lesson people took from them was the wrong one.

The data eventually showed up. David Kao’s 2020 study put 3,018 players into an action RPG across juice levels and found that zero juice and extreme juice both significantly cut play time, motivation, and performance. That’s an inverted U. The backlash had been saying this since 2014 (Folmer Kelly at GDC Europe: “while adding polish makes a game feel more alive, we’re actually losing a level of immersion”), but in my experience designers respect a curve more than a complaint.

The reason is attention. Attention is currency, the player’s and the designer’s both. Every screenshake is a withdrawal from a finite account, and when everything is emphasized, nothing is. The designer’s account is just as finite: a team spending its whole feel budget on particles is not doing the invisible work, the buffering and window-widening Maddy Thorson described on Celeste, where “everything is fudged a tiny bit in the player’s favor.” That forgiveness layer never shows up in a screenshot, but I’d argue it carries more feel than anything that does.

One more thing juice can’t fix: cadence. How often your core loop pays out (a fight every forty seconds, or every four minutes) is a feel decision, and it usually gets made by accident, by level design or economy tuning. Try to decide it on purpose. The interval between payoffs sets the player’s pulse more reliably than anything you do inside the payoff itself.

I’m one of the few designers who shipped inside both TTK religions, back to back. At Treyarch in 2018, Black Ops 4 raised base health to 150, a deliberate, public stretching of CoD’s famously instant deaths toward more reaction time and more skill expression. A few months later I was at Respawn, where Apex carried the longest effective time-to-kill in the genre: armor and a heal economy that give every fight a beginning, a middle, and an end.

Both schools are coherent. CoD’s sub-300-millisecond kill says death is a verdict on positioning: you lost that fight before it started, go think about the minimap. Apex’s long fight says death is the end of a negotiation, where you had time to react, reposition, and get out-played, so the kill means something when it lands. They’re different fantasies, but both are honest. TTK tells players what death means, and what death means tells them what mastery is.

There’s no sin in picking a school. The sin is moving the number without restating the philosophy. When Apex’s Season 24 damage increases pushed the game toward CoD-style lethality (2025, after my time), the community split loudly enough to force partial reverts. That anger wasn’t really about a spreadsheet cell; players were defending what death had meant for six years. Touch TTK and you are renegotiating your death philosophy in public, whether you meant to or not. One practical footnote: short TTK also amplifies every netcode sin you have, because players start dying before the world state agrees they were visible.

The combo system in Street Fighter II was a bug. Designer Akira Nishitani kept it: “It didn’t seem to cause any bugs, so we decided it could be a feature to expand the gameplay.” Thirty-five years later that’s still the cleanest statement of how great feel actually happens: it gets discovered, then blessed. The corollary is harder: once the community discovers something, they begin to own it, and your design authority over it starts depreciating.

Tap-strafing is the canonical case, and I had a front-row seat. Apex inherits Quake-lineage air control through Titanfall (it’s called “lurch”), and players found that binding forward to scroll wheel let them take right-angle turns mid-air. In August 2021, Respawn announced its removal: “It’s inaccessible, lacks readability/counterplay, and is exacerbated by movement abilities.” Every word of that is defensible design reasoning. The community revolted anyway, and the removal was walked back. In 2024 a nerf attempt was reverted within days, with the studio conceding that “feedback and feel around movement systems are critical, and there’s legitimate skill expression that we intend to preserve.”

I was leading Experience Design at Respawn through both rounds. The movement calls weren’t mine, and I’m not here to relitigate them. What I took from watching it happen twice is that by 2021, tap-strafing wasn’t a technique anymore, it was identity: thousands of hours, a montage culture, a vocabulary. The change was correct by every rubric we teach, and it was unaffordable on the only ledger that mattered. Meanwhile, bunny-hop healing left the game with minor grumbling, because nobody had built themselves out of it. The rule here isn’t “never touch movement.” It’s that the political cost of a feel change scales with the identity accreted onto it, and you should price that before the announcement, not after.

The same lesson played out in those same years on a different channel. Respawn published a netcode deep dive in 2021 defending its 20Hz servers with honest math: “for triple the bandwidth and CPU costs, you can save two frames worth of latency in the best-case scenario.” The math was right, and it convinced almost no one, because players don’t experience math. s1mple, the best CS player alive, reviewed Valve’s subtick servers in one sentence (“It feels… like 16”) and that sentence outran every benchmark published. Telemetry sets your floor and wins your internal arguments, but it doesn’t win the public ones. (Chapter 02 makes the full floor-versus-target argument; the narrower point here is that data doesn’t win public arguments.) Players can be wrong about causes and still right about the feeling, and both halves of that sentence matter.

Rotational aim assist is zero-latency target tracking. No human wrist matches it at close range, which is why pro Apex players converted to controller (not out of nostalgia, just arithmetic). In August 2024, in the final weeks of my time at Respawn, designer Eric Canavese said in public what years of patch notes had danced around: “Simply put, aim assist is just too strong in these PC lobbies,” and PC values were cut by about 25 percent. A year later, Treyarch published close-range win-rate data ahead of Black Ops 7 and made rotational assist require active stick input. I respect both moves, but neither one ends the war.

It can’t be ended this way, because the premise is broken: you cannot tune one number to be fair across two physically different input devices. A thumbstick and a mouse are just different instruments. Any single assist value picks a loser (set it high and mouse players are fighting software; set it low and stick players are fighting physiology), and every tuning pass just changes who’s losing while calling itself balance.

The honest fix is to stop pretending the number can be fair and make the pools fair instead: match inputs with inputs wherever your population allows it. Where it doesn’t (small regions, crossplay parties, off-hours), say so out loud, in the product, and publish what assist is doing. Players will forgive a stated tradeoff almost indefinitely; what they won’t forgive is a number that pretends to be neutral. Notice this is the netcode lesson again. The pattern repeats at every timescale, which is really the point of this site.