The key points
- Evidence quality is different from live availability and contractual assurance.
- Read original units and observation dates alongside normalized values.
- Unknown or conflicting information should remain visible.
An explanation of the beta, not an audit certificate
This note explains the current beta’s evidence concepts and selected public presentation rules, based on a review of the implementation on 27 September 2026. It is not a certification that every field in the catalogue has been independently verified, and it does not announce that every provider is integrated or partnered with Kovara. [1]
The distinction matters because a platform can preserve useful evidence without possessing a complete picture of the market. Our aim for this publication is to explain how to interpret what is present, what still needs confirmation, and where a buyer should avoid filling gaps with assumptions.
Start with the claim and its source
A source is the record supporting a statement. In practice, a provider pricing page, a technical specification and a response to a particular requirement support different kinds of claims. A source’s existence does not prove that every property of a related offering is covered by it. The relevant question is which fact the evidence supports. [1]
We recommend reading the original value and unit before the normalized reference. A per-GPU-hour calculation should not erase whether the purchasable configuration is a whole node, and a starting price should not become a specific quote. The current comparison surface retains qualifications concerning price floors, quantities and extra charges. [2]
Likewise, a dated source is not necessarily a fresh observation of stock. The time a price was checked, the date a provider supplied confirmation and the period for which that confirmation is valid answer different questions. [2]
A grade is not a warranty
Kovara uses the existing A–D confidence scheme: A for Highly Verified, B for Strong Evidence, C for Incomplete Evidence and D for Estimate / Unverified. These labels describe evidence confidence; they are not uptime promises, provider credit ratings or reservation guarantees. The scheme is documented in the project’s operating rules. [1]
A pending source-review flag means a review issue remains recorded. It should not be interpreted as proof that every value is wrong, but it should not be ignored when relying on the affected configuration. In the current comparison logic, pending source reviews and starting floors are excluded from the shared-basis lowest-price highlight. [2]
Our recommendation is to use the grade together with the original source, date and field-level gaps. A single letter cannot replace those details. Where an important condition is unresolved, ask for explicit confirmation rather than treating the grade as permission to assume the answer.
A map point also makes a claim
The Regions and Overview mapping work distinguishes city or metro references from broader references and offerings without a reliable map point. Unspecified and global placeholders should not be turned into invented facility locations. A regional catalogue association is not proof of a specific rack or data-center address. [3]
That is why an offering can contribute to a catalogue total without appearing as an exact-location pin. For diligence, keep provider region, physical facility and customer data-residency requirements separate until the supplier’s evidence connects them.
Unknown is a useful result
Workload Economics uses recorded performance observations when suitable evidence is available. The evidence-selection implementation uses comparison groups and withholds conflicting or stale profiles; it does not establish performance for an unrelated model merely because the GPU name matches. When suitable evidence is absent, performance-adjusted estimates should remain unavailable. [4]
Our position is that a visible unknown can improve a decision. It identifies the next question to ask. Replacing it with zero, a generic specification or an unqualified estimate may make a page look complete while making its conclusion less dependable.
Use Kovara to organize discovery and evidence, then carry the unresolved conditions into the sourcing conversation. A request, saved record or comparison remains a decision aid until the corresponding provider confirmation and commercial steps actually take place. [6]
Sources & editorial note
Reference documentation is listed below with its recorded check date. Technical statements are attributed; passages framed as our view or recommendation are editorial interpretation. Examples are hypothetical unless explicitly identified otherwise. No independent Kovara hardware testing is claimed.
- Kovara · Marketplace and recorded evidence ↗Kovara product surface · Checked 27 September 2026
- Kovara · Comparison qualifications ↗Kovara product surface · Checked 27 September 2026
- Kovara · Regional coverage ↗Kovara product surface · Checked 27 September 2026
- Kovara · Workload Economics performance evidence ↗Kovara product surface · Checked 27 September 2026
- Kovara · Candidate price index ↗Kovara product surface · Checked 27 September 2026
- Kovara · Request Compute ↗Kovara product surface · Checked 27 September 2026
Prepared with AI assistance. Publication approved by Tommaso Luci. Kovara Research is the publication label, not a claim of an independent laboratory or a named analyst team.
Implementation review basis: Kovara beta revision 6ec8ee0, 27 September 2026. Product-surface links provide context, not an independent audit.