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

# What you have to be able to produce

> Clause 4.4 requires written information on request. Here is what that means in practice and where each record sits in guardbook.

Clause 4.4 is the clause that gives all the others teeth. It requires firms to be able to provide, **on request**, written information about five things.

The standard says "on request" and does not limit who can ask. In practice that could be a client, RICS, or someone dealing with a complaint or claim. The obligation is the same either way, and "we do this, we just did not write it down" is not an answer to it.

> [!NOTE]
>
> RICS has not published an assessment checklist for this standard, and this page is not one. It is built from the five things clause 4.4 names, and the records the other clauses require you to hold.

**In guardbook:** Client Explanations, opened from Client Disclosures, puts all five together for each AI tool. Pick the client, and it only shows reliability decisions about that client's work. Record when you send it, and the evidence pack lists it.

## The five things

> [accordion] 1. The type of AI system used
>
> Which tools, supplied by whom, doing what at your firm.
>
> **Where it lives:** the [Tool Register](https://help.guardbook.ai/tools-register/overview.md).

> [accordion] 2. Its basic ways of working and limitations
>
> What the system does and where it falls down. Limitations, not just capabilities.
>
> **Where it lives:** the Purpose and Known limitations fields on each tool in the [Tool Register](https://help.guardbook.ai/tools-register/overview.md). The capability evaluation in the tool's [appropriateness assessment](https://help.guardbook.ai/tools-register/suitability-assessments.md) adds detail.

> [accordion] 3. The due diligence carried out before using it
>
> What you asked the supplier, what they said, what they would not say, and what testing you did.
>
> **Where it lives:** [supplier due diligence](https://help.guardbook.ai/tools-register/supplier-due-diligence.md) per tool.

> [accordion] 4. How relevant risks are identified and managed
>
> Not a list of risks. The process: how they are found, scored, mitigated, reviewed.
>
> **Where it lives:** the [Risk Register](https://help.guardbook.ai/risk-register/overview.md) and your AI policy together. The register shows the risks, the policy shows the process.

> [accordion] 5. The decisions made about the reliability of outputs
>
> The clause 4.2 written decisions.
>
> **Where it lives:** the [Oversight Log](https://help.guardbook.ai/oversight-log/overview.md).

## What else you would want to hand over

Beyond clause 4.4, the other clauses each leave an artefact. If someone is asking questions about your AI governance, these are what answer them:

| Record | Clause |
| - | - |
| Materiality determination and reasoning | 1.2 |
| Evidence of training, all three duties | Section 2, 3.1, 3.2 |
| Yearly data safeguards check | 3.1 |
| Written AI policy, adopted and signed | 3.2 |
| Appropriateness assessment per tool | 3.2 |
| Risk register, with evidence of quarterly review | 3.3 |
| Supplier due diligence per tool | 4.1 |
| Reliability decisions | 4.2 |
| Dip samples for automated or high-volume tools | 4.2 |
| Terms of engagement wording and client disclosures | 4.3 |
| Written explanations given to clients | 4.4 |

**In guardbook:** the [evidence pack](https://help.guardbook.ai/evidence-audit/evidence-pack.md) assembles what the product holds into one document.

## The two questions that are hardest to answer well

**"When did you last review this?"** A register or a policy created once and never touched is the most common weakness, and it is visible from the timestamps. Clause 3.3 says the register must be reviewed and updated at least quarterly, not just reviewed.

**"Show me a decision where you concluded the output could not be used."** A log in which every single output was reliable invites the question of whether anyone was really applying professional scepticism. Rejections are evidence that the process is real.


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