What is an ATS score?

There is no universal ATS score

Applicant tracking systems do not share a common scale. Each company configures its own filters, its own searches, its own display criteria. There is no authority grading resumes, and no grade that follows you from one employer to the next.

So any tool telling you your ATS score is 78 is describing its own measurement, not an external truth. The only question worth asking then is: what does that measurement cover, and can it be verified?

What a score can honestly measure

One thing is measurable without inventing anything: extraction fidelity. You compare what your resume contains against what a parser pulls out of it. Whatever is missing at the far end is lost, as a matter of fact, and that loss can be counted.

It is an objective measurement because it judges neither your career nor your fit for a role. It does not say you are a good candidate. It says of the eight roles on your resume, six were reconstituted and two disappeared. That is checkable line by line.

A machine-legibility score is not a substitute for fit. They are two separate questions, and conflating them is the main source of confusion on this subject.

Our scale, in full

The Passmatch fidelity score runs from 0 to 100. It is a weighted average of five components, and here are their exact weights:

• Roles recovered — 40%. A role only counts if the title, the employer and the start date are all three reconstituted. It carries the heaviest weight because it is the core of a resume.
• Identity and contact details — 20%. Name, email, phone. A perfectly extracted resume nobody can attribute is unusable.
• Education — 15%. Degree and institution.
• Skills — 15%. Hard skills, the ones that get searched for.
• Section order preserved — 10%. A document whose blocks come back shuffled reads badly, even when nothing is missing.

The product holds itself to thresholds on that score: a Master resume is only validated above 95, a targeted resume above 90. Below that, the document at fault is ours rather than yours, and it goes back through.

Why we keep the worst parser, never the average

Your document is re-read by several independent extractors, all deliberately crude. One reconstructs the layout; the other reads the raw text with no reading-order reconstruction at all — that is, in the dumbest way possible.

The score we keep is the more pessimistic of the two, and the text we display is its own. Averaging would give a flattering, less useful number: you will not meet the average of all ATS, you will meet one, and nothing guarantees it is the clever one.

It is also why the re-read text is shown to you raw rather than summarised. A number you cannot hold up against reality is a number nobody checks.

What this score is not

We would rather say it ourselves than let the doubt do the work:

• It is not a simulation of any particular commercial recruitment system. We operate none of them. We use deliberately crude parsers standing in for the least favourable case.
• It is not a prediction of being hired, or even shortlisted. That decision belongs to the recruiter and to criteria we do not know.
• It is not a measure of fit with a job ad. That question is handled separately, by a compatibility score between your resume and the ad.
• It is not a language model's opinion. Fidelity is computed by comparison, deterministically: the same documents always give the same number.

Two scores answer two different questions

The commonest confusion is expecting a single number to answer everything. Two distinct measurements coexist, and neither replaces the other.

• Fidelity answers does my document survive a machine reading? It depends on your resume alone. It can, and should, reach a very high level — it is a question of form, and form is entirely fixable.
• Compatibility with a job ad answers does my background match this role? It depends on the ad as much as on you, changes with every application, and cannot reach 100 everywhere — that would be the sign it measures nothing.

The order matters: compatibility computed on a badly extracted document is meaningless, since it would be measuring an amputated career. Fidelity first, compatibility second.

What a low number means, and what it does not

Low fidelity is not a judgement on your background. It is a formatting diagnosis, and that is rather good news: it is the part of the problem that can be fixed completely, in one pass, without changing anything you have actually done.

Watch out for one reading trap, though: the score is weighted, so two resumes sitting at 70 can be in trouble for opposite reasons. Losing the identity costs 20 points at once when only three lines are missing; losing half the roles also costs 20, but the document is far more damaged. Always look at the per-component breakdown before concluding.

Finally, a high score guarantees nothing about the outcome of an application. It guarantees one thing, and that is already a great deal: that you will not be set aside for a reason that has nothing to do with your work.

How to read your report

A score on its own is useless. What is useful is the list of what was lost, and the actually extracted text beside it.

• Look at the identity first: a missing name or email almost always means they sit in a header or inside an image.
• Then look at the missing roles: they are usually split apart, inside a table, or in a second column.
• Finally, read the raw text. That is your document as the machine sees it. If you do not recognise yourself in it, no amount of keyword tuning will compensate.

The steps for fixing all three cases are set out, one by one, in the practical guide.