How to make a good game, starting from one verb

A verb is the thing the player does most, written as one word. Jump, shoot, dash, place, steer, match, talk. The best small games often have exactly one and spend every level finding new uses for it. Celeste has jump, dash and climb, and every screen is a puzzle built from those three. Its dash moves at 240 pixels per second for 0.15 seconds with a 0.2 second cooldown, values you can read in the source its developers published.

The loop is the verb plus a reason to repeat it. Write it as a short chain. Dash to cross a gap, cross the gap to reach a collectible, collect it to open a harder route. If you cannot write the loop in one line, the game does not have one yet, and more content will not add it.

Test the verb alone first. In an empty grey room with no goal, is pressing the button still fun for a minute? If yes, you have something to build on. If no, change the verb before you build anything else.

A vertical slice is a short, finished piece of the game with a start, the loop, a challenge, an ending and feedback close to final. Ten minutes is long enough to show whether the loop holds up and short enough to rebuild in a week if it does not.

Resist the pull toward content. More levels, enemies and story all multiply the cost of every later change. A slice that is fun makes content easy to justify. A slice that is not fun turns content into a way to avoid the problem.

• One level from start to finish, with a way to win and a way to fail
• The real core loop, repeated at least three times with small variations
• Placeholder art is fine, placeholder feedback is not, so sounds and hit reactions go in now
• A title screen, restart and quit, so testers never need your help

Jakob Nielsen's usability research found that about five users uncover roughly 85% of the problems in a design, and that three rounds of five teach more than one round of fifteen. Games are not websites, but the lesson carries over. Small, repeated tests beat one big one.

Strangers matter. Friends are kind, and anyone who watched you build the game already knows the answers. Hand over the controls, say nothing, and note where players stop, what they try that does not work and where they smile. What they do is more reliable than what they say afterwards. If three of five miss the same jump, the jump is the problem.

When you are ready for strangers you cannot sit next to, Steam Playtest is a free, separate app attached to your store page. Players request access from the page, and the playtest does not touch your reviews or wishlists.

Make the first game small enough to finish in about a month, with one mechanic, one setting and three to five levels or one loop that repeats with variation. Finishing teaches the parts prototypes skip, such as menus, save data, builds, store pages and bugs that only appear on someone else's computer.

Small does not mean unambitious. Celeste Classic was made by two developers in four days in PICO-8, and it grew into the full Celeste, released on Steam in January 2018. The small version proved the idea first.

A quick test is to write the whole game on one page. If the page needs a second sheet, cut features until it does not.

Game feel is the difference between a jump that feels sticky and one that feels right, and it lives in a handful of values. The Celeste developers published their Player class, so you can read their values instead of guessing. The table above lists the key ones.

Coyote time lets a jump register for a moment after the player walks off a ledge, and Celeste uses 0.1 seconds. Jump buffering does the reverse, remembering a jump pressed just before landing and firing it on touchdown. A buffer of about 0.1 seconds is a good starting point. Players never notice either one, they only notice when it is missing.

Feedback is the other half. Every input should answer with a sound, an animation and a change in the world. Add them in layers and play after each one, the way Jan Willem Nijman does in his talk The Art of Screenshake, where a flat shooter becomes a satisfying one without any change to its rules.

• Coyote time around 0.1 s
• Jump buffer around 0.1 s
• Variable jump height, so a tap gives a short hop and a hold gives a full jump
• Falling that caps at a readable speed, with a faster cap when the player holds down
• A short sound and a visual reaction on every hit

Steam requires a Coming Soon page to be public for at least two weeks before a new product releases, and Valve recommends posting it as soon as you are ready to talk about the game publicly. Its documentation says there is no strong downside to a long Coming Soon period and that wishlists collected far ahead of launch can convert as well as later ones.

Valve's caveat is useful too. If the art direction or core features may still change a lot, wait, because early wishlisters may be confused by a very different game. That is one more reason to finish the vertical slice first. When the slice looks like the game, it is ready to become the store page.

Publishing on Steam costs a 100 USD Steam Direct fee per game, recouped after that game earns 1,000 USD. The money guide covers the full list of fees, store shares and waiting periods.