UNPRICED is not zero

There’s a small design decision inside TokenTriage that says almost everything about how it thinks. When it meets a model it has no price for, it does not render the cost as $0.00. It renders it as UNPRICED, and it counts the rows it had to exclude.

This sounds pedantic until you sit with it:

“We spent nothing on this model” and “we don’t know what we spent on this model” are opposite facts, and only one of them is good news.

A dashboard that collapses the second into the first is lying to you in the most plausible way possible — with a clean, confident zero. You look at your total, it looks complete, and a slice of your real spend is silently missing.

What TokenTriage does instead

An unknown model produces a null cost, flagged unpriced: true, and it propagates through every aggregate as an exclusion count, not a zero. So a total reads like:

$41.20 over 1,904 priced rows — 116 unpriced rows excluded

You can’t miss it, and you can’t mistake it for the whole picture. The excluded rows are a to-do (“go price these models”), not a rounding error.

The mirror image matters too: a model you’ve declared free — a local endpoint, say — gets an explicit $0 with unpriced: false and a provenance stamp. Real zero and unknown are kept apart on purpose, because conflating them is exactly the failure that makes cost dashboards untrustworthy.

Every dollar on every surface carries this lineage: the pricing snapshot it was computed against, whether it’s estimated, and where it came from. That’s not a feature we bolt on for the demo. It’s the invariant the whole product is built to hold.