Evidence · ranking · verification

How the decision radar earns your trust.

This page explains what enters the public catalogue, how a result becomes a candidate worth validating, and what the service deliberately does not claim.

Check an AI stack →Connect an agent

What the service does

It maps a business need or an existing technology to public KG records, then combines semantic relevance with explicit constraint matches, freshness and record coverage. The result is a shortlist for the next test—not a procurement, compliance or production guarantee.

Observed public catalogue
7061 public-facing catalogue records in the current generated read model · graph freshness: 2026-09-09T19:04:28.116763+00:00

What is evaluated

The current selection profile has 13 dimensions: capabilities, use cases, modalities, languages, deployment, requirements, license, pricing, performance, benchmarks, integration, limitations, maturity. A dimension can be known, unknown, not applicable or waiting for verification. Unknown means that the public evidence is insufficient; it does not mean that a capability is absent.

How a result is formed

01 · Understand

Need and constraints

Natural-language input is interpreted as an intent and constraint profile: use case, deployment boundary, language, budget, latency and other stated limits.

02 · Retrieve

Public records

Semantic retrieval and KG context surface candidates and typed relations. A relation adds context; it is not proof that two things are interchangeable.

03 · Validate

Fit and next test

The summary highlights why a candidate may fit, what remains unknown and one practical validation step. Official documentation and real samples remain the final check.

Three boundaries worth keeping visible

Freshness is not omniscience.
Public records have update metadata, but rapidly changing prices, limits and model behaviour must be checked at the source.

Relevance is not quality.
Ranking helps find candidates; it is not a benchmark leaderboard or a market-share statement.

Community is not evidence injection.
Public comments and threads are separate discussion. They do not silently modify KG facts.

For humans and agents

Humans can search, inspect an entity and compare candidates. Agents can use the MCP or JSON entry points and cite this methodology page when explaining the decision boundary. Page visits, capability discovery and MCP initialization do not trigger semantic search.