A CI bill should reconcile to work that ran. That requires two facts to stay separate: how long a machine existed and how much capacity that machine represented.
Runnable records authoritative machine time, then applies the published size multiplier to produce normalized minutes. Included usage and overage are calculated from that ledger.
Measure the lifecycle once
The control plane owns the timestamps that start and stop billable execution. A runner cannot extend or rewrite its own usage after the fact, and a transient delivery failure does not erase the run record.
Cancellation closes the same lifecycle as success or failure. The conclusion changes; the accounting model does not.
Normalize capacity explicitly
A larger runner consumes more capacity per elapsed minute. Runnable applies a visible multiplier so teams can compare usage across machine sizes without pretending those minutes are identical.
The dashboard keeps elapsed execution, normalized usage, included minutes, and overage distinct. Each number answers a different operational question.
Spend controls belong beside usage
A plan defines included minutes and concurrency. Customer-controlled limits decide what happens before overage becomes an unexpected invoice.
The goal is not a clever calculator. It is a ledger an engineer and a finance team can both reconstruct from the same runs.