> ## 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. # Dip sampling > Check a random sample of what your automated or high-volume AI tools produce, and sign the result off. The RICS standard (clause 4.2) asks firms that use automated or high-volume AI to check a random sample of its outputs at regular intervals. You do not check every output. guardbook tells you how many to check, picks which ones at random, and keeps the signed record. Tools that produce work one piece at a time, where a surveyor reviews each output, are covered by the [Oversight Log](https://help.guardbook.ai/oversight-log/overview.md) instead. ## Which tools need it Only tools marked **Automated or high volume** on the Tool Register. That means tools that produce outputs automatically or in volume, such as automated valuations or bulk report drafting. To mark a tool, go to the **Tool Register**, choose the pencil icon next to it to edit it, set **Automated or high volume** to **Yes**, and pick how often to sample it: every month, 3 months, 6 months or 12 months. RICS sets no interval. Every 3 months is a sensible default. If no tool is marked, the **Dip Sampling** page says so, and the duty does not count towards your GUARD Score. ## Taking a sample **1. Count the outputs** Find how many outputs the tool produced since the last sample. Most tools show this in their job history, report log or usage page. Your own job list works too, for example "we ran 120 valuations through it since July". **2. Draw the sample** On **Dip Sampling**, choose **Start a sample** next to the tool. Enter the period and the number of outputs, then choose **Draw the sample**. guardbook works out how many to check using a recognised sampling standard (ISO 2859-1) and picks them at random, for example numbers 7, 31 and 88. 120 outputs means 20 checks. Random picks are what make a small sample count. **3. Check each one** Number your outputs in the order they were produced and find each picked one. For each, enter the report or job number, who checked it, and the result: **Acceptable** or **Issue found**. An issue needs a note saying what was wrong. The person checking should be suitably qualified to judge the output against their own professional judgement. **4. Sign it off** Once every output is checked, choose who is signing, write what the sample showed and choose **Sign off sample**. Once signed, the sample and its checks cannot be changed. ## When a sample finds an issue You must record the action the firm took before you can sign. guardbook also asks whether any client needs to be told. RICS 4.2 says that where an output cannot reasonably be relied on for its purpose, the client must be told in writing, with your reasoning or a summary of it. Choose one: * the affected client was told in writing (and the date) * the output was caught before it reached a client * the issue did not stop the output being reasonably relied on, so there was nothing to tell the client If an issue means an output could not be relied on, consider adding or updating an entry on the [Risk Register](https://help.guardbook.ai/risk-register/overview.md). ## When a sample is due Each flagged tool shows its status: **Next due** with a date, **Overdue since** a date, or **Never sampled, due now**. When a sample falls due, guardbook opens it for you and fills in the period. You only need to enter the count. If no sample is needed that time, for example because the tool was not used, choose **No sample needed this time?** and give the reason. The sample still shows as due until one is signed off. ## Discarding a sample If a sample was drawn by mistake, choose **Discard this sample** before it is signed. It stays on record as discarded, so anyone reviewing your records can see every draw, not only the ones you kept. ## In your evidence pack Signed samples, with their random draw, each check and the sign-off, go into your [evidence pack](https://help.guardbook.ai/evidence-audit/evidence-pack.md). This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.