Skip to main content
Minimum result size protects individuals in small groups. You set a minimum number of people that must sit behind every value in a result. When a value would be computed over fewer people, Human Intelligence withholds it and shows it as suppressed instead. The row and its labels stay in the result, so you can tell a value was withheld rather than that the group is empty. It applies everywhere you get results: the app, MCP-connected AI tools and the SQL API. It is off by default: nothing is suppressed until a Data Admin sets a minimum, for the whole organization or for particular datasets.

What it does

Say your minimum is 5 and a user can see average salary and job title. Suppression works together with access control. Counting happens over the rows the person is allowed to see, so someone limited to one department is suppressed more often than someone who sees the whole organization, never less. A title with 10 people company-wide but 3 in their department is suppressed for them. In the app, a suppressed value appears as “Suppressed” with an explanation on hover. AI tools see the same label along with the minimum, so they can explain the blank. In the SQL API a suppressed value is a plain NULL, and you can select the matching __suppressed_ field for a metric to tell it apart from a missing value.

Where the minimum is set

You can set a minimum at three levels. Each one only ever adds protection: The highest minimum that applies wins. Restrictions accumulate, so setting a minimum in one more place can raise the bar but never lower it. Say your organization threshold is 5, one role is set to 10 and a survey dataset is set to 8. Someone with that role needs at least 10 people behind a value everywhere. Someone without it needs 8 on the survey dataset and 5 on every other dataset. Anyone with full data access is not subject to suppression: that means Human Intelligence staff accounts and your own Data Admins, who can already query all data. Everyone else, including Platform Admins who aren’t also Data Admins, follows the same rules. A role cannot exempt someone. To let a person see small groups, they need full data access. The minimum you can choose at any level is 2, 3, 4, 5, 6, 8, 10, 15 or 20. A minimum of 1 or 0 isn’t accepted, because it would hide nothing.

Setting the organization threshold

Anyone with permission to manage organization settings can set the organization minimum, in organization settings, in the Minimum Result Size card.
  • Threshold: No organization minimum by default. Choose a minimum of 2, 3, 4, 5, 6, 8, 10, 15 or 20. Setting no organization minimum does not mean nothing is suppressed — a role or a dataset can still set one, and the highest that applies wins.
  • Apply to: choose Metrics, including counts, or Metrics, excluding counts.
Apply to is one setting for the whole organization, and it covers every level: the organization threshold, role minimums and dataset minimums. You can change it even when there is no organization minimum, which matters when you only protect particular datasets. Use “including counts” when a count can itself be sensitive, such as the number of people rated a certain way within a team. With “excluding counts”, a count over a sensitive filter can still show a 1. That’s the trade this option makes. If you set a minimum on a dataset to protect small groups, choose “including counts”: with “excluding counts”, the dataset’s own counts still reveal how many people are in a group.

Setting a role’s minimum

Use a role minimum when one group of people should be held to a stricter standard than everyone else, for example analysts who only need broad trends.
  1. Go to Settings > Permissions and find the data role in the Roles section.
  2. Edit the role. A role’s minimum can be set when you edit an existing role, not when you first create it.
  3. In Minimum result size (override), choose a number, or No role minimum to set none of its own, then choose Save Role.
The role’s detail panel shows the result as Minimum result size. A role minimum at or below the organization threshold has no effect, because the highest minimum wins. A role minimum does apply when there is no organization minimum, so you can suppress small groups for one role only. When someone holds several data roles, the highest minimum among them applies.

Setting a dataset’s minimum

Use a dataset minimum when only some of your data needs protection. For example, you can leave the organization minimum unset and set a minimum of 10 on your survey datasets only, so survey results can’t single anyone out while headcount and recruiting numbers stay unsuppressed.
1

Open the dataset

Open Data Catalog from the sidebar, go to the Datasets tab and select the dataset. Its detail panel opens.
2

Set the minimum

Under Minimum result size, pick a number from the dropdown. It saves as soon as you pick.
3

Check the result

The dropdown shows the dataset’s minimum. The new minimum takes effect without any other change.
To clear a dataset’s minimum, pick No dataset minimum from the same dropdown. You need permission to manage organization settings to change it; without that permission the dropdown is visible but not editable. A dataset minimum can raise the minimum for that dataset but never lower it. If the organization threshold is 10 and you set a dataset to 5, the dataset still uses 10.

Datasets that can’t take a minimum

Some datasets don’t record which person each row is about, so there is no way to count the people behind a result. On those, the panel doesn’t show Minimum result size at all, because the setting can’t do anything there. Built-in metrics on them are never suppressed. The one exception is a dataset that already has a minimum stored, which can only happen if it was written outside the app: there the panel does show the dropdown, says the stored value protects nothing, and lets you pick No dataset minimum to clear it.
Setting a minimum on one of these datasets through the API, outside the app, protects nothing and also blanks your organization’s own custom metrics on that dataset. If you see a minimum stored on a dataset that can’t take one, pick No dataset minimum to clear it.

What counts as a person

The minimum counts distinct people, not rows. A person with six pay changes is still one person, and a candidate with several interviews is still one candidate. Each dataset counts the person its rows are about, such as the worker, the candidate or the person who was reviewed, rather than whoever wrote or entered the record. Some datasets don’t describe individual people, such as valuations or job openings. Their values aren’t suppressed. A small number of other datasets have rows that aren’t about a single person, so they can’t be suppressed yet. Their detail panel says so. While suppression is on, custom metrics your organization adds to a dataset that isn’t covered always show as suppressed, because there’s no way to count the people behind them. Setting a minimum on a dataset that can’t take one has the same effect, which is why the app doesn’t offer it. Metrics that look across a rolling window or across earlier results (multi-step calculations) aren’t suppressed.

Anonymous surveys

A survey that records neither an employee ID nor an email can’t say who answered. On those surveys, each response counts as one person, so suppression counts responses rather than people. One person who answers the same anonymous survey twice counts as two. Treat the minimum on an anonymous survey as a minimum number of responses.

Rollup datasets

A few datasets hold results that are already summarized, for example one row per manager, department or survey. For these, Human Intelligence adds up the number of people recorded on each row. A single row is counted exactly, but when a result groups several overlapping rows together, such as a director’s row and the rows of the managers beneath them, the same people can be counted more than once. That can let a value through that covers fewer people than your minimum.

What it does not do

  • It doesn’t hide extremes. The highest or lowest value in a group of five is still one person’s value.
  • It doesn’t stop comparison across queries. Asking for a group and then for the same group minus one person can reveal a difference. Each query is checked on its own.
  • It doesn’t control row-level reads. Whether someone can read individual pay or other detail is set by categories and org boundaries. Suppression covers only calculated values.
  • It doesn’t cover every dataset. See What counts as a person.
  • It doesn’t check each part of every ratio. A ratio built from other metrics is suppressed when either part covers fewer people than the minimum. Ratios and other derived metrics built directly from raw columns are checked against the whole group, not each part.