Instruments
Read the answer off the chart
Most of what is written about applied AI is a list of things you could use. Very little of it tells you what any of them costs. These are the working parts of this site: charts you can move, each holding one relation that decides whether an architecture is affordable.
The nomogram
The six governing relations of the stack, on one chart.
Drag two scales, read the third.
The window
The gap between the tokens you buy and the tokens the model reads.
Repartition the prompt, watch the two come apart.
Cache economics
When a prompt cache pays for itself, and when it never does.
Set the hit rate, compare four providers.