Cost to serve is a division

The board asks what a customer costs to serve. On a per-app cloud that's a week with the billing export, tags that were half-applied, and a model nobody trusts.

On a per-customer-container architecture it's two numbers. GET /api/v1/rentals/{id} gives the server's amount for the period. GET /api/v1/projects/{id}/stats gives container_count: 1,140 customers on that box this month. Divide. That's infrastructure cost per customer, and it's correct because there is nothing else on the bill: no per-container charge, no metered bandwidth, no storage tier, no request count.

Add the next hundred customers and the numerator doesn't move until the box fills: the runtime endpoint says when. Then you rent another server and the denominator jumps again.

Unit economics used to be modeled. Here they're read.

Cost to serve is a server price divided by a container count.