Scoop Count

Artist’s Statement: Scoop Count

Scoop Count is a game which allows you to enjoy one of my precise daily rituals:

preparing a pre-workout drink in a time crunch before heading out to the gym. Behind the simple mechanics of filling the shaker bottle is a controlled and measured daily optimization. I am very interested in how humans take the time to quantify details of their routine and learn to make it more efficient when necessary. This game displays just that with the high-stakes, 60 second race where the player must balance speed and precision.

To bring this game to life, I defended two critical design choices during the development.

The first was that I insisted on implementing non-linear and exact accelerating pour physics as opposed to a static fill rate. This creates a psychological tension between precision and the restraint necessary. If you go too fast, you miss out on points for precision or worse, risk spilling all together and losing the game. Secondly, we refined the UI to make the chemical details and shaker more intuitive at a glance. This was critical in later levels when new mechanics brought in new issues where it required deciding how to best increase the pressure for the player and what mechanics of the game to alter which would do so.

The development process was much back-and-forth collaboration with Claude Code but

the prompts were particularly direct, especially the initial ones since the more foundational infrastructure for the game came about very quickly after the ideation process. Claude’s performance exceeded my expectations in its execution of the visual UI and overall initial layout of the game. It rendered a clean and functioning layout which matched my mental image pretty

exactly and instantly. It also accounted for various subtle gaps that I had not even considered such as outer edges to the goal lines on the shaker. However, we did hit a difficulty when it came to identifying and interpreting a core challenge for the game. Claude’s initial version was very flat and predictable with no dynamic variables or difficulty parameters which made the game feel empty and trivia. After we went back and forth, the tighter physics implemented allowed for more of a direct difficulty round by round.

Ultimately, I would consider Scoop Count to be my work made using Claude Code as a

tool. While Claude generated the underlying syntax, the model could not have conceived this specific game on its own. An LLM can go and generate infinite loops of arcade games but they would most of the time lack personal intent and aim at a lived human experience. The concepts used, priorities, core mechanics, and corrections were driven by my preferences. One prompt which helped specifically was, “fix the clerical layout of the sliders and gauges on screen to make ingredient limits readable.”, as well as “Each ingredient in the shaker has a distinct target volume with a green "optimal tolerance zone" marked on a visual fill gauge. Overfilling past the upper red line ruins the mix.”.

AI Disclosure:​

- I used Claude Code in VS Studio as the terminal interface and Claude Desktop for brainstorming. I also used Gemini earlier in the process to figure out how to effectively use Claude Code in VS Studio since it was not something I was initially familiar with. It took between 8 and 10 rounds of iteration which was less than I had initially thought it would need but I was impressed by how much of the game actually was built out from 1 or 2 initial prompts. I overruled a lot of the initial difficulty ratings for the game as well as built out how a level style

should work since the actual depth of the games was rather barebones whereas the UI and development was strong.