Every prompt sent to Claude while building this game, in order, with what it cost.
How to read the numbers
Subscription share: how much of the plan's weekly limit went up while Claude worked, converted
to a share of one month (a month holds about 4.35 weekly allowances). The limits are account-wide, so time
between prompts never counts, and jumps that a prompt's own tokens can't explain are left out as concurrent
use. Readings have 1% resolution.
Tokens: text the model read and wrote, with the model that did it. Every tool call re-reads the conversation, so "read" is
large but mostly served from a cheap cache.
API list price: what the same tokens would cost on Anthropic's pay-as-you-go API. Shown only for
reference.
Time: from sending a prompt to Claude's last action, including checking the game in a headless
browser.