Each output is a signed weighted sum of input contributions, with optional non-monotonic peaks: some outcomes rise then fall as their driving input grows (acidity peaks at moderate temperature and time; clarity declines as anything that drives extraction goes up). Coefficients and peaks are listed in the DATA block at the top of the source.

This is teaching-grade, not lab-grade — it's meant to make the relationships visible, not predict your next cup. Real extraction has hundreds of variables, non-linearities at every level, and bean-specific behaviour. The playground is to coffee what a sketched ƒ(x,y) is to fluid dynamics: useful, abstract, wrong in the details.

The point of presets: they're the named brewing methods from the dimensional map, projected back into input space. Click 'Espresso', then 'Cold brew' — watch the outputs flip. Then try high pressure with coarse grind: see why that combination doesn't show up in the field (low everything, watery body, no acidity). The playground lets you break the bundles — every covariation in the map shows up here as 'why moving one slider alone gives a worse cup than moving the bundle together'.