Eligibility
Exact GPU, commercial basis, geography, freshness, confidence grade, public visibility and comparable normalized price are evaluated explicitly.
Kovara
Loading your page…
You can continue using the navigation while this page loads.
You can continue using the navigation while this page loads.
Kovara separates market evidence, eligibility, provider contributions, benchmark calculation and publication. A fixing records the exact methodology version used and is never silently rewritten.
Raw listings do not vote equally. Kovara first creates one provider contribution, then calculates the reference level across independent provider contributions.
Exact GPU, commercial basis, geography, freshness, confidence grade, public visibility and comparable normalized price are evaluated explicitly.
Kovara selects each provider’s latest eligible timestamp. If multiple comparable observations share that timestamp, their median becomes that provider’s contribution.
The benchmark median, quartiles, range and dispersion are calculated across provider contributions, preventing listing-heavy providers from dominating.
Produce transparent candidate reference levels for exact-GPU public on-demand compute pricing using one provider contribution per independent provider.
Kovara does not infer capacity from a listing, and it does not convert unknown availability into unavailable supply.
Available, Limited and Unavailable are counted only when explicitly present on a tracked offering. Unknown status remains a separate bucket.
Known capacity includes only explicit available_gpu_count values attached to offerings currently marked Available or Limited.
Depth accumulates explicit known capacity whose latest eligible comparable offering price is at or below each price ceiling. It is market evidence, not reserved inventory.
Kovara publishes term observations only when commercial-duration evidence is comparable and coverage clears the same institutional thresholds used by the underlying series.
The zero-term anchor is the published KCI fixing when available, otherwise the live methodology-compliant candidate.
Committed observations are grouped by duration and reduced to one latest provider contribution before cross-provider aggregation.
Observed committed pricing may be shown as evidence coverage, but no reference term point is published until provider, region and observation thresholds pass.
Provider count, region breadth, observation count and freshness requirements must all be met. Below-threshold series remain visibly insufficient.
If a fixing requires correction, Kovara publishes a new revision linked to the earlier fixing rather than changing the historical row in place.
Experimental, Candidate, Reference, Suspended and Discontinued are explicit series states. Candidate series must not be represented as official reference rates.