The whole arithmetic behind an ATS score, printed on the page

A résumé checker that gives you 68 out of 100 and a list of tips has told you almost nothing. You cannot tell whether 68 is a formatting problem or an evidence problem, whether the missing 32 points sit in one rule or thirty, or whether the number would move at all if you did the work. The number is doing the job of a verdict while carrying none of a verdict's obligations.

So the ATS résumé checker here prints its own sum. We ran it, on an invented candidate, and photographed what came back.

The first screen asks for one thing

We pasted a 162-word résumé for a backend engineer: two roles, seven achievement lines, one degree, nine skills. It took four minutes to write and it is not remarkable. That matters for reading the number that follows.

Three dimensions, three fixed weights

The engine scores three things and weights them the same way on every document it sees.

Completeness is not a checklist of eight equal boxes. The points are identity 10, email 10, phone 5, experience 30, education 15, skills 10, dates 15 and summary 5. Experience written only as projects earns 18 of its 30, because an ATS application form asks for an employer, a job title and dates for each job, and the report says so. The dates points are earned in proportion to the career entries that carry their own date: a period written once on a section heading is not attached to the entries beneath it by any parser we can see.

Content is graded, not passed or failed

The content dimension is four rules, each graded in proportion rather than passed at a threshold. Achievement quality, 35 points: every line is graded from 0 to 1 (a result verb 0.5, a measurable figure 0.35, enough words to say what changed 0.15), and the rule pays the mean. A line that opens with a duty (worked, helped, supported, assisted, responsible for) earns no verb credit, so a page of duties no longer scores like a page of results. Fewer than three lines earn a share of it. Measurable evidence, 20: lines with a figure, against a target of 35% of the lines (at least three). Evidence depth, 25: the number of lines, full at twelve. Readable volume, 20: distinct words, full at 500; a sentence repeated on the page counts once.

These weights were checked on 25 September against the one public set we found where people scored résumés: 302 résumés, each rated by two domain experts who agreed with each other at a rank correlation of 0.65. The previous rules agreed with the experts at 0.20 within a field; these reach 0.26. That is a modest number and we print it for that reason: no résumé checker we have seen publishes one.

Then the sum, printed

Our candidate reads 100 × 30% + 100 × 30% + 71 × 40% = 88, and the report shows the multiplication rather than the conclusion. Seven lines are all results, but seven lines of twelve and 162 words of 500 are graded as the thin page they are. (On 12 September, under the rules of that day, the same page read 97.) The three dimension scores and the published score come back from the scoring engine with every check, and the card prints the sum from those numbers, so the arithmetic on the page and the code that produced the number come from the same place.

And a job description

The ATS review scores the document alone, and that is deliberate: structure and evidence are properties of your résumé, not of any posting. Reading a résumé against a particular posting happens in the cover letter builder, which counts the requirements it recognised in the advert, builds the letter from the lines of your résumé that answer them, and names the ones your résumé cannot support.

The ceilings, and why they exist

After the weighted sum, four ceilings apply. Under 25 readable words the score is capped at 15. If no experience, education or skills can be found, 25. If experience and education are both missing, 40. If there is no experience and the content dimension is under 30, 55. A blank page with a beautiful header can otherwise score respectably on parseability and completeness, which would be a measurement telling a flattering lie.

When a ceiling applies, the report says so and shows both figures. That is the whole difference between a capped score and a low one, and it changes what you should do next.

What the checker read, in counts

Beside the score the report shows the extracted text itself, line by line, with its count under it: 23 lines, 162 words for our candidate. That is the text the score was computed from, and anything a parser could not read is not in it. If your page is two dense sides and that line says 90 words, the scoring is a distraction. The extraction is the problem, and the format guide is the fix.

Every check says what it earned

Our candidate passed fifteen of seventeen checks. The two marked Review say exactly what they are worth: evidence depth 15 of 25 points, seven achievement lines where twelve earn the full score, and readable volume 6 of 20, 162 words where 500 do. A checker that lists eleven suggestions in a uniform typeface has quietly made every suggestion equally urgent, and the reader's afternoon disappears into the cheapest one. Printing the points is the difference between a list and advice.

Eighty-eight is not a compliment

The candidate does not exist and the résumé took four minutes. That is what a high readiness score means: the document is readable and evidenced, and nothing about it is broken. It says nothing about whether the work behind it is any good, and the report says so on its face, above the score. A score you can check is worth having. A score that flatters you is worth exactly what you paid for it, which here is nothing.