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

# Write criteria that score well

> How the scorer reads your criteria, and how to write ones that give you useful scores.

Every candidate is scored against your criteria, one criterion at a time. The scorer reads what it knows about the candidate and asks, for each sentence you wrote: is there evidence for this? So the quality of your scores depends almost entirely on how you phrase the criteria.

This guide shows what works, with examples. For the mechanics, see [How match scores work](/companies/match-scores).

## Write things you'd find evidence for

The scorer reads the candidate's profile, their messages with you, their answers in your process steps and your team's notes. A criterion is only useful if one of those can show it.

| Hard to score            | Easier to score                                          |
| ------------------------ | -------------------------------------------------------- |
| Great leader             | Has managed a team of five or more engineers             |
| Strong communicator      | Has written customer-facing documentation or given talks |
| Passionate about fintech | Has worked at a payments or banking company              |
| Senior                   | Has at least six years in backend roles                  |

Qualities like curiosity or grit rarely show up in a profile, so they tend to come out as **Missing evidence** for everyone. If they matter, find out in the process instead: a questionnaire question or an AI interview produces answers the scorer reads at the next rescore.

## One thing per criterion

Each criterion gets one score. "Has used React and TypeScript in production" scores the same whether someone has one or both, so you can't tell which is missing. Split it into two criteria and the **Match** tab shows you exactly where the gap is.

## Use Required for what you'd reject without

Required criteria count double compared with preferred ones. If everything is required, nothing stands out, and a candidate who is excellent on your two real must-haves looks the same as one who is average across the board.

A useful test: if a candidate didn't meet this, would you still talk to them? If yes, it's **Preferred**.

## Put everything that matters to the score in the criteria

The score only reads the criteria. The **Details** fields, such as **Location**, **Work mode** and **Languages**, shape the search and the listing, but they don't change anyone's score. If the location really decides fit, write it as a criterion: "Lives in or can relocate to Oslo".

Staffer has no separate seniority field either. Write seniority as a criterion.

## Avoid near-duplicates

Two criteria that say almost the same thing waste weight and can confuse the scorer. If a rescore fails with a message that the scorer only covered part of the brief, look for criteria that read very similarly and reword or merge them.

## Keep the list focused

A brief can have up to 30 criteria, but more criteria make each one count for less. Most roles are well described by five to ten. Each criterion can be up to 400 characters, one idea per row.

## Don't select on protected characteristics

Don't write criteria about age, gender, ethnicity, religion, health, family status or anything similar. Staffer's own rules tell the agent not to write them. If your organization turns on **Live policy enforcement**, the agent's changes are also checked against your policies before they're saved (see [AI policies](/admin/ai-policies)).

The check covers what the agent writes. It doesn't check criteria you type into the brief yourself, so those are your responsibility.

## After you change the criteria

Editing criteria doesn't rescore anyone. To see the effect on people you've already found:

* On the **Scouting** tab, open a search's **Search actions** menu and click **Re-score**.
* For candidates in the pipeline, use **Rescore** in each candidate's **Candidate actions** menu.

See [When scores change](/companies/editing-and-rescoring).

## A worked example

A first draft for a senior frontend role:

* Required: "Great React developer"
* Required: "Team player"
* Required: "Knows design systems and accessibility"

A better version:

* Required: "Has built production web apps in React for at least five years"
* Required: "Has built or maintained a component library or design system"
* Preferred: "Has shipped work that meets accessibility standards such as WCAG"
* Preferred: "Has worked in a product company rather than an agency"

The second version gives each candidate four separate answers you can read in a minute, and it tells you exactly what to ask about in the interview.