> ## Documentation Index
> Fetch the complete documentation index at: https://docs.seekr.com/llms.txt
> Use this file to discover all available pages before exploring further.

# Core concepts

> How SeekrGuard's models, datasets, evaluators, examinations, and risk profiles fit together.

SeekrGuard measures how a model behaves on your data, using criteria you define. A model, a dataset, and an evaluator combine into an examination, and its scores roll up through risk categories into a profile that rates the model's risk. For the precise definition of any term, see the [glossary](/guard/glossary).

```mermaid theme={null}
flowchart LR
  Model --> Examination
  Dataset --> Examination
  Evaluator --> Examination
  Examination --> Scores["Stored scores"]
  Scores --> Category["Risk category"]
  Category --> Profile["Risk profile"]
  Profile --> Result["Risk score and level"]
  Model --> Spot["Comparison or chat"]
  Evaluator --> Spot
  Spot --> Immediate["Scores you read directly"]
```

## Provide inputs

An evaluation draws on three inputs:

* **[Model](/guard/glossary#model)** – the model under test, such as GPT-4 or Llama, registered from a provider. See [Register and manage models](/guard/app/models).
* **[Dataset](/guard/glossary#dataset)** – the inputs you run the model on, optionally with expected answers. See [Add and manage datasets](/guard/app/datasets).
* **[Evaluator](/guard/glossary#evaluator)** – a scorer that grades each output on one dimension, such as accuracy, toxicity, or relevance. See [Create evaluators](/guard/app/evaluators).

These inputs depend on one another. Each evaluator declares the dataset fields it needs, such as a user question, a context passage, or an expected answer, so it can run only on a dataset that supplies those fields and against outputs a model has produced.

## Run evaluations

You put the inputs to work in two ways:

* **[Examination](/guard/glossary#examination)** – a batch run that scores one or more models across all or part of a dataset and reports aggregated results. See [Run examinations](/guard/app/examinations).
* **[Comparison](/guard/glossary#comparison)** – a focused check that scores a single response with several evaluators and judge models at once. See [Run comparisons](/guard/app/comparisons).
* **[Chat](/guard/app/chat)** – talk to a model directly and score any response on the spot.

Use an examination for systematic vetting, a comparison for a quick side-by-side look, and chat for probing a model as you go. Only examination results are stored for risk scoring. Comparison and chat scores are read where they appear and go no further.

## Assess risks

Individual evaluator scores become useful once they combine into a single, weighted measure of risk:

* **[Risk category](/guard/glossary#risk-category)** – a named dimension, such as factual accuracy, backed by one or more evaluators paired with the datasets they run against. See [Create risk categories](/guard/app/risk-categories).
* **[Risk profile](/guard/glossary#risk-profile)** – a weighted combination of risk categories for a specific use case. See [Build risk profiles](/guard/app/risk-profiles).
* **[Risk score](/guard/glossary#risk-score)** and **[risk level](/guard/glossary#risk-level)** – scoring a profile against a model yields a score from 0 to 100, where higher means more risk, rated from low to critical.

For exactly how scores are normalized, weighted, and combined, see [how a risk score is calculated](/guard/app/risk-profiles#understand-how-a-risk-score-is-calculated).