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

# Coaching

**AI Coaching** is the Assistant workspace for investigating poor outcomes and turning them into reviewed changes.

Open **AI → Coaching**.

Coaching combines a grounded AI conversation with a review queue of poorly rated Assistant tickets.

## Ask questions about Assistant behavior

You can use Coaching as an investigation workspace to ask questions such as:

* Why did RAUM choose this procedure?
* Which source supported this answer?
* What policy is missing for this case?
* How should the existing procedure be rewritten?
* What should RAUM remember for similar tickets?

Coaching can retrieve relevant RAUM sources and show those sources alongside its answer.

## Pin sources to the coaching context

When a source is relevant to the investigation, pin it to keep it in context for subsequent coaching messages.

This is useful when you are comparing a ticket against one particular procedure or guideline rather than asking the coach to rediscover the same source on every turn.

## Review poorly rated tickets

Assistant feedback with a negative rating can enter the Coaching review queue.

A review item can include the ticket context, RAUM's recorded answer, and an agent note/comment associated with the negative feedback.

Load the item into Coaching to discuss what should have happened instead.

## Proposed edits are reviewed before they are applied

When Coaching recommends changing a knowledge item, it can produce an **edit draft** rather than modifying the source silently.

The review card shows the current and proposed content side by side.

You can:

* inspect the diff;
* copy or expand the proposed content;
* **Apply** the change; or
* dismiss it.

> [!WARNING]
>
> Coaching suggestions are not automatically policy. The user still decides whether to apply a procedure/guideline edit.

## AI memory notes

Coaching can also propose adding information to AI memory.

A memory proposal is shown for review with a **Remember** action. Use this for durable context that belongs in memory rather than rewriting an entire procedure.

Do not use memory as a hiding place for important policy that should be explicit and reviewable as a procedure or guideline.

## Closing the review loop

When an applied coaching action is linked to a ticket from the negative-feedback queue, RAUM can mark that review item as reviewed so the queue reflects work that has actually been handled.

You can also mark a review item reviewed without applying an edit when the ticket does not require a knowledge change.

## Open Coaching from ticket activity

Assistant activity can deep-link a past RAUM-answered ticket into Coaching. This loads the recorded ticket context so you can investigate the run instead of copying details manually.

## A practical workflow

**1. Open a poorly rated ticket**

Load the ticket from the Coaching review queue or from Ticket Activity.

**2. Identify the failure**

Decide whether the problem was bad source selection, missing knowledge, an unclear procedure, a communication rule, a tool result, or model behavior.

**3. Pin the important evidence**

Keep the relevant procedure, guideline, or source in the Coaching context.

**4. Request a concrete fix**

Ask for a procedure/guideline edit or an appropriate memory note rather than a vague promise to “do better.”

**5. Review and apply**

Inspect the diff or memory proposal, then apply only if it reflects the policy you actually want.

**6. Retest**

Run a representative ticket in Copilot Mode or Simulation and verify that the change solved the original failure without creating a new one.

> **[Ticket Activity](https://docs.raum.am/assistant/activity.md)**
>
> Find the run you want to investigate and open it in Coaching.

> **[Quality & Testing](https://docs.raum.am/assistant/quality-testing.md)**
>
> Validate the fix against outcomes, gaps, and Simulation.


This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.