140.6

140.6: Artist's Statement

I'm training for an Ironman: a 2.4-mile swim, a 112-mile bike and a 26.2-mile run, with 17 hours to finish all of it. 140.6 is that whole day compressed into about two and a half minutes. What it's about

In endurance racing, it's rarely your legs that end your day. It's your fueling. You can feel great at mile 60 of the bike and be finished at mile 15 of the run because of choices you made hours earlier. Push harder and you go faster but drain the tank. Play it safe and the cutoffs close on you. I wanted a game where that tension is the whole game. When you bonk, the screen tells you: "The race is decided hours before you feel it."

Decisions I'd defend:

Effort is a zone, not a gas pedal. You press up or down to set Z1 through Z5, the heart rate zones endurance athletes train by, and the setting stays until you change it. Holding a key to go fast would have made it a reflex game. A zone makes it about commitment: you pick a pace and live with what it costs.

One item per aid station. Every station forces a choice between gel and water, never both. The afternoon heat drains water faster, so the right answer shifts as the day goes on. Real fail states. You lose by bonking or by missing a real cutoff: 2:20 for the swim, 5:30 PM for the bike, midnight for the finish. I chose "tense and frantic" over a slower, emotional feel because race day doesn't feel cinematic from the inside. It feels like math is under stress. Where the collaboration was real, and where it wasn't

Claude added a drafting rule I never asked for. Sit in the box behind another rider and you save energy, but stay more than a second and a half and you eat a five-minute penalty, like the real rule. It's my favorite mechanic because it hides a second pace-versus-energy temptation inside the first one. Where it wasn't: Claude's brainstorm pitched twelve concepts pulled from my resume, mostly finance ideas like a leveraged buyout played as Lunar Lander, and it recommended two of them. I rejected all twelve and picked the thing I'm actually training for. What I authored

I chose the subject and made the four calls that define the game: what the struggle is, how much of the race to include, how you lose and how it should feel. Claude built everything else, including the difficulty. It ran simulated races at every effort zone until Z1 missed the bike cutoff, Z2 finished slow and Z4 survived only with perfect fueling. That balance is why the game

works, and I didn't do it. What I'd claim is the premise. A model can write a triathlon game in seconds, but it couldn't have known this was the one I'd make, specific details about the ironman I wanted included in the game, or that the bonk was what it should be about.

Prompts that shaped the game

Prompt 1, fixing small details:

"Make the swim, bike, and run look more realistic. Add animations to the water and the aid stations and fix the water and gel bar to be more accurate"

Prompt 2, my design spec (answered as four choices):

Core struggle: pace vs. energy. Structure: all three legs in one round.

Fail state: bonk or miss a cutoff. Feel: tense and frantic.

AI disclosure

Tools: Claude code. The assignment called for Claude Code; which builds the single HTML file. Iteration: one brainstorm round (12 concepts, all rejected), two rounds of design questions, and two build rounds in which Claude wrote the game and ran three rounds of its own simulated playtests to tune fuel burn, water drain and how crowded the bike course got. I did not edit the code by hand.

Corrected or abandoned:

All 12 brainstorm concepts; an early balance where every effort zone finished easily; and a bike course so crowded that drafting penalties were unavoidable.