A screening score looks like a verdict. It is not. It is a measure of how well a document matched some words you wrote. Reading it properly is mostly about remembering what it is not.
What the number actually is
It is a weighted total of how well the extracted data matched your rules. That is all.
It is not a prediction that the person will be good at the job. It is not a comparison against your best employees, unless you specifically switched that on. And it cannot be compared between different jobs, or between different versions of your rules.
Two things follow straight away. Do not report "average candidate quality" across jobs — the numbers do not mean the same thing. And do not compare this week's 82 with last month's 74 if you changed the rules in between, which you did.
False precision
Scores usually show as whole numbers, which suggests the tool can tell 81 from 79. It cannot. The gap between 81 and 79 is noise. The gap between 81 and 55 is real.
The fix is to use bands instead. Three or four groups: strong match, worth reading, weak, filtered out. Then work in bands.
This also stops a hiring manager asking why candidate six is above candidate seven — a question with no real answer that will still eat twenty minutes of your meeting.
Testing the explanation, not the score
Any tool worth using tells you why. Test that reason three ways:
- Is it a quote or a summary? A reason that quotes a line from the CV can be checked in seconds. "Strong background in distributed systems" is a summary, and it might be invented — see Module 1, Lesson 3.
- Does it change when you change the CV? Take a test CV, delete the line the explanation quotes, run it again. If the reason is unchanged, the reason was decoration added afterwards.
- Does it name one of your rules? Explanations should point at rules you wrote. If they mention things you never asked about, something else is driving the ranking.
That third test is how you spot a tool quietly using its own judgement underneath your rules.
Score patterns to be suspicious of
| What you see | Probable cause | Check |
|---|---|---|
| A long thin tail at the top | One unusual word is dominating | Remove that rule and run again |
| Two separate groups | A filter is acting as the score | Confirm your filters |
| Almost everyone above 70 | Rules too general for this pool | Add something that separates people |
| Scanned CVs all scoring low | Bad reading, not bad candidates | Screen scans as a separate batch |
| One employer's people all scoring high | The tool is adding something you did not ask for | Check what it uses beyond your rules |
The scanned-CV row is worth remembering. Any time score lines up with file format, you are looking at a data problem being read as a hiring signal — the same fault as Lesson 2, showing up one stage later.
Showing scores to other people
A hiring manager will read a number as a measurement of the person unless you frame it otherwise. Two habits prevent most of the damage:
- Show bands and evidence, not scores. "These six meet the bar on X and Y — here is the line in each CV that shows it."
- Never send a ranked list without the rules attached. A list with no rules is just an opinion with a number on it.
One sentence to remember
The score tells you how well a document matched your description. It tells you nothing about how well the person will do the job. That is what the rest of your process is for.