Cost coverage
What the report means by coverage, why every record carries two cost numbers, and how unexplained spend is shown.
What coverage means
Coverage is the honest answer to: how much of what we spend on AI, and of what it does, does this report actually see? Oabo states it three ways and never fills a gap with a zero.
- The grade counts N of M use cases: how many use cases in use hold a letter.
- Paid for but powering no use case in use: spend that nothing explains, named and totalled on its own line.
- Each connection's blind spot, on Connections beside its reach: what a feed cannot see is stated as prominently as what it can.
Note
A feed that goes quiet reads downstream as this system stopped costing money, which is the most dangerous wrong answer the product can give. That is why keys do not expire, why a silent feed shows as a gap, and why unknown spend reads Spend not entered rather than $0.
Two cost numbers
Every usage record is priced twice, and both numbers are reported:
| Figure | What it is |
|---|---|
| Actual | What this usage marginally cost. Zero for subscription and local billing, because a flat seat costs the same whether it is used or not. |
| Equivalent | What these tokens would cost at metered rates, against published list prices and the organization’s own rates on Model rates. Always priced. |
This is why sending your own cost figure loses information: subscription usage has a real actual cost of zero and a large equivalent cost, and collapsing them to one number destroys the comparison. A model absent from both rate tables gets a null equivalent and is reported as unpriced, not priced at zero.
Unexplained spend
On Costs, Total monthly AI spend is costs only; benefits are never added to it. Below it, Unexplained spend lists the tools and agents that power no use case in use this month. The same figure closes the grade sentence and opens What's missing on the report, with the instruction Explain this spend or cut it.
A cost reaches a use case only through Powered by. A tool that powers several use cases divides across them; How shared costs divide shows the split. Nothing is allocated by guess.
Gaps
| Gap | Where it shows |
|---|---|
| A use case missing an input | Not graded yet: followed by the input, on the use case, in the report’s Not graded yet list, and as the one next step. |
| Usage nobody attributed | The unattributed pool, counted per source on Connections. Usage evidence, not spend, until someone assigns it. |
| A machine without the collector | Nowhere, which is the point of the Developer machines blind spot. Install it, or use a fleet door. |
| A model with no rate | unpriced_models in usage summaries; the equivalent cost is null. |
| A feed that stopped | Connections: last event time stops moving. GET /usage/gateway/coverage lists which surfaces are covered. |
The connection doors and what each misses: Reach and blind spots. The pricing endpoints: GET /usage/summary and GET /usage/summary/by-model.