> ## 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.

# Oversight Log: your firm's written reliability decisions

> What the Oversight Log is for, when RICS clause 4.2 requires an entry, and what each entry records.

The Oversight Log is where your firm records its written decisions about whether AI outputs are reliable enough to use.

It exists because of clause 4.2 of the RICS standard *Responsible use of artificial intelligence in surveying practice*, which takes effect on 09/03/2026. That clause requires a surveyor to apply professional judgement to the reliability of an AI output, and to write that decision down.

## What an entry is, and is not

An entry is a record of professional judgement. It is not a record of having used AI.

The distinction matters because it changes what you write. A log of AI use answers "what did we run this through?". A reliability decision answers "what did I assume, what concerned me, and can this output be used?". Only the second satisfies clause 4.2.

Each entry captures:

* the **surveyor** who accepts responsibility for using the output
* the **tool** that produced it, selected from your Tool Register
* **what the AI produced**, as the output itself, a summary, or a reference to it
* the **decision**: Used as-is, Modified or Rejected
* the **five reliability fields** clause 4.2 requires, covered on [the five reliability fields](https://help.guardbook.ai/oversight-log/five-required-fields.md)
* the **decision date**, and an optional client or job reference

## When you need an entry

The duty applies to outputs that have a **material impact** on the delivery of the surveying service. Not every use of AI meets that test. The standard's own examples of material outputs include a summary of a document you then rely on when writing a report, text composing a significant part of an opinion, and a recommendation about where to investigate a fault.

> [!NOTE]
>
> Where AI automates outputs or produces them in high volume, the standard accepts that writing a decision for every one is generally neither necessary nor proportionate. It requires randomised dip samples at regular intervals instead. Firms remain accountable for each output either way.

Record the entry before the work is issued where you can. The standard does not set a deadline, but a decision written after a client has already relied on the output is a weaker record, and a reviewer will notice.

## What the log gives you

Entries are dated, attributed to a named surveyor and linked to a specific tool. That last link is what makes the log useful rather than merely compliant: if a tool is later found to be unreliable, you can identify which pieces of work relied on it.

Every entry, and every deletion, is written to your firm's audit trail with the user and the time.

> [!WARNING]
>
> Entries can be deleted from the log. Deleting marks the record as removed and writes that to the audit trail rather than destroying it, but the entry no longer appears.
>
> Treat deletion as a way to correct a mistake, and make sure the people using the log understand that is what it is for.

## Where to go next

> [card] [Record a Decision](https://help.guardbook.ai/oversight-log/recording-a-decision.md)
>
> A step-by-step walk through the form, field by field.

> [card] [Five Required Fields](https://help.guardbook.ai/oversight-log/five-required-fields.md)
>
> What clause 4.2 asks for in each of the five reliability fields, with a worked example.

> [card] [The Named Surveyor](https://help.guardbook.ai/oversight-log/sign-off.md)
>
> Who can be named, how supervision works, and what guardbook does not check.


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