Platform map

EvalGate is one product with many surfaces. Use this page when you know the job but not which manual to open. Maturity and client gaps stay in Feature status; behavior detail stays in the current-behavior manuals.

Choose by job

You are trying to do this	Start here	Do not confuse it with
Understand the repository and current integration	Repository activation	Proof that application behavior was executed
Inspect what happened	Logs & Trace Explorer	A complete run when required evidence is absent
Preserve examples of expected behavior	Dataset Hub and feature manuals	Independently qualified truth merely because a row was saved
Define a reusable check	Scorer Studio or LLM judges	Proof that the check is accurate
Make a bounded semantic decision	Decision judges (Jev)	A generative explanation or deployment approval
Compare application changes	Playground and Experiments	An evaluator-only benchmark
Measure the evaluator	Compare evaluators and Evaluator qualification	The application’s business success rate
Compare score regimes	Calibration control plane	Task-specific qualification or action permission
Review failures and suggestions	Prepare review and Review Queue	Automatic approval from AI assistance
Govern a model call or side effect	Runtime controls	Universal interception of traffic outside EvalGate
Publish an artifact and verify its use	Deployable assets	An experiment’s pending publish proposal
Explain the full remediation	Unified Eval Loop	Proof that every deployment has completed all stages
Read inconclusive results	Resolve inconclusive results	Treating unknown as pass or fail
Connect a control to evidence	Connect controls to evidence and Risks & controls	A shipped universal audit registry
Inspect named production failure patterns	Failure patterns	Inventory-only topic queues without why-it-matters

Four distinctions

The application is not its evaluator. A better answer and a better judge are different experiments.
An observation is not an authoritative decision. Acceptance policy and terminal composition decide how an observation is interpreted.
Qualification is not permission. Strong measurement evidence can coexist with an absent action approval.
Source is not release proof. Code, a declared API, a local test, an installed package, and a verified live workflow are separate evidence levels.

Layer bridge

For how application behavior, evaluation, evaluator measurement, qualification, and authorization relate, start with How the evaluation system fits together.

Inventory

Every named product surface and its Shipped / Experimental status is listed in Feature status. Prefer that inventory over historical section stamps in older manuals. Canonical docs URL: evalgate.com/docs.