VERIFY · IN DESIGN
Compute by initiative
What each initiative spent on inference — token counts and request ids, never money.
Every task an agent runs spends inference, and compute/1 is the record of how much, attributed to the initiative that spent it. Attribution rides an initiativeId the call carries through the estate’s inference seam (@flashyos/llm-gateway); the ops console already folds its own call log into one of these histories and renders it. What is not built yet is a served fragment — a URL a stranger can read — so this page shows the shape and says it is waiting rather than drawing a number it cannot fetch.
The record carries token counts and request ids and no money: the price is derived at read from a model registry this page does not hold, which is why a per-initiative history is safe to publish here at all. And it never shows the measured compute without the count of initiatives that ran uncaptured beside it — a total you can grow is worth publishing only next to the gap that would expose you growing it.
The compute source has not answered. This is not an empty history — it is a surface waiting for its data. no compute/1 source is configured for this deployment yet. A served compute/1 fragment renders here the moment one is reachable; until then the page reports the silence rather than drawing zeros that would claim the network did no thinking.
How it fills in
A property tags the inference it spends with the initiative that spent it, folds its call log into a compute/1 fragment, and serves it. This page reads served fragments and draws them here beside the settlement explorer and the tally — the same record, measured a different way. Until a fragment is served, an untagged call is a gap to instrument, not zero compute, and the page says so rather than implying the network did no work.