> ## 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. # The five reliability fields RICS clause 4.2 requires > What a written reliability decision must contain under RICS clause 4.2, and what to write in each of the five fields in the guardbook Oversight Log. Clause 4.2 of the RICS standard *Responsible use of artificial intelligence in surveying practice* requires a surveyor to make a written decision about the reliability of an AI output, and sets out five things that decision must detail. In guardbook these five sit together in the Oversight Log form, under the heading **RICS 4.2 Reliability assessment**. All five are required. The entry will not save with any of them blank, and the server checks again before the record is written. > [!WARNING] > > These five fields are about **reliability**, not about describing the task. > > The most common mistake is to record the tool, what it was used for, and a note that someone checked it. That is a useful audit record, but it is not what clause 4.2 asks for. Clause 4.2 asks what you assumed, what worried you, why, what could be done about it, and whether the output can be used. ## The five fields | # | Field in guardbook | What clause 4.2 asks for | | - | - | - | | 1 | **Assumptions** | Any relevant assumptions you made when judging the output. | | 2 | **Concerns about reliability** | The key areas of concern about reliability, including the reliability of any underlying datasets. | | 3 | **Reasons for each concern** | Why each concern you raised is a concern. | | 4 | **What could reduce the concerns** | Whether anything could be done to lessen each concern. | | 5 | **Overall reliability conclusion** | The effect of those concerns on overall reliability, including a stated conclusion on whether the output can reasonably be used for its intended purpose. | Field 5 carries the most weight. A conclusion is required, not a summary. "The output has some limitations" is not a conclusion. "I am satisfied this can be used for X, subject to Y" is. ## The named surveyor Clause 4.2 also requires the written decision to be prepared by, or under the supervision of, an appropriately qualified and named surveyor who accepts responsibility for the output's use. That is the **Surveyor** field at the top of the form. It is a picker rather than free text, and it draws on the people recorded against your firm. If someone is missing from the list, add them to your [People record](https://help.guardbook.ai/account/people.md) first. ## A worked example A valuer uses an AI tool to draft a yield analysis for an industrial unit. > [accordion] Assumptions > > I assumed the comparables the tool drew on were current to Q1 2026, and that the unit's stated floor area of 1,840 sq ft was correct as given in the instruction. > [accordion] Concerns about reliability > > The tool does not disclose which comparables it used or how recent they are. Industrial yields in this location moved noticeably in the first quarter, and the output may not reflect that. > [accordion] Reasons for each concern > > Two in-house comparables agreed in the last six months show a net initial yield roughly 0.25% below the figure the tool produced. Without visibility of the underlying dataset I cannot tell whether the difference is a data lag or a genuine difference in view. > [accordion] What could reduce the concerns > > Cross-referencing against Land Registry data for the past six months and the two in-house comparables. I did this before finalising, and adjusted the yield down by 0.25%. > [accordion] Overall reliability conclusion > > With that adjustment made and the reasoning recorded, I am satisfied the output can reasonably be used to support the valuation. The unadjusted figure could not have been. Note what makes this a strong record. It names the assumption rather than implying it. It says what could not be seen, not only what was checked. It states an outcome. ## If the answer is no Where the conclusion is that the output **cannot** reasonably be used for its intended purpose, clause 4.2 requires the firm to communicate that conclusion to the client in writing, together with the reasoning or a summary of it. > [!WARNING] > > guardbook records your decision. It does not send that communication for you. > > Selecting **Rejected** and writing a conclusion satisfies the internal record. It does not discharge the duty to tell the client. That letter or email is yours to send, and it is worth keeping a copy against the job file. ## When a written decision is needed The duty bites on outputs that have a **material impact** on the delivery of the surveying service. That is a judgement on the facts of the use, not a category of tool. An output that summarises a document you then rely on when writing a report is the standard's own example of a material one. Where AI automates an output or a series of outputs, or produces them in high volume, the standard accepts that a written decision for every output is generally neither necessary nor proportionate. Firms remain accountable for each output, and must instead take randomised dip samples at regular intervals. > [!NOTE] > > Dip sampling is a separate mechanism from an Oversight Log entry, and guardbook does not yet schedule or record dip samples. If your firm relies on the dip sampling route for a high-volume tool, keep a record of the samples you take somewhere durable in the meantime. This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.