> ## Documentation Index
> Fetch the complete documentation index at: https://docs.correl8.ai/llms.txt
> Use this file to discover all available pages before exploring further.

# Concepts

> Observations, guidelines, research studies, and how Correl8 processes your data

## Observation

An **observation** is one product signal: **title**, **description**, and **user\_sentiment**. Create one or more per call via `post_observations` (MCP) or the [REST API](/essentials/overview) for custom integrations.

Each observation belongs to a **project** and is keyed by **`user_id`** and **`interaction_id`** — your strings for the end user and conversation thread, not Correl8 accounts.

## Issue

An **issue** groups related observations into a digest. Issue digest runs asynchronously after observations are created.

## Tags

**Tags** are matched server-side from observation content, asynchronously alongside issue digest. Tags appear on the observation after processing completes.

## Sentiment

Every observation requires `user_sentiment` — a five-level label from very negative to very positive. See [User sentiment](/essentials/using/observations#user-sentiment) for the scale.

## Guidelines

**Guidelines** are project-specific markdown your agent loads once per conversation via the **`get_correl8_guidelines(user_id=…)`** tool. Two types:

* **Observation guidelines** — when and how to call **`post_observations`**
* **Research guidelines** — how to conduct studies and call **`post_research_answer`**

Defaults ship with Correl8. Edit them in **Settings → Details**, or from the **Observations** and **Research** pages. Active studies (per-study **question**, **eligibility criteria**, and any logged **answer**) are loaded separately via **`get_active_studies(user_id=…)`**.

See [Guidelines](/essentials/using/guidelines) for how to tune them without changing your agent code.

## Free tier

If your workspace exceeds the free-tier limit (1,000 processed observations per month), Correl8 keeps ingesting but does not run tag matching or issue digest until the period resets or you upgrade.

## Research study

A **research study** is an open-ended question your agent pursues during normal conversation — not a survey. You define a **question**, **eligibility criteria**, and sampling limits in the app. When a study is **active**, users who fall in the sample see it via the **`get_active_studies(user_id=…)`** tool. **Eligibility criteria are evaluated by the agent, not the server** — the tool returns the criteria text and it is your agent's responsibility to check whether the current user/conversation matches before pursuing the study.

Your agent follows project **research guidelines** for conduct, gathers signal subtly, then posts one **answer** per user via **`post_research_answer`**. Each answer is a structured summary tied to **`user_id`** and **`interaction_id`**.

See [Research](/essentials/using/research) for creating studies and reviewing answers.