quack / water
loading the cups

Every little drop.

Quack’s usage, poured into a cup. Real request counts, estimated water.

The global cup

—

Reading recorded usage…

—model requests
—full 250 mL cups
scaled against all-time usage
every request
makes a ripple.

A cup for every server

pick a cup to see who’s filling it.

Gathering the glasses…

Who’s filling it?

everywhere

Counting the drops…

How is the water estimated?

Change the assumption and every total updates. This is a scenario, not a water meter or a measured footprint.

Longer work now counts more than small talk. Uncached input, output and cache-writing tokens count at full weight; cache reads count at 10% as a rough compute proxy. The starting 0.4 mL per 1,000 weighted tokens is adjustable because model size, hardware, cooling and electricity can change the real amount substantially.

Each recorded model call counts, including tool follow-ups. Requests are not Discord replies. Completed 250 mL cups are counted separately. Glass levels share an all-time scale, so a shorter period can’t look fuller than all time. Rankings follow weighted tokens.

Global includes all recorded departments. “Unassigned & other” covers DMs, background work, former servers and activity that can’t be mapped to a current server. Tracking may omit calls that weren’t logged. Dates use America/Chicago.