What an ATS score actually means

The arithmetic

A weighted sum of 14 dimensions, with weights that add to 100 so no rescaling is needed.

Each check scores your document out of 100 on its own question, and that score is multiplied by the check's published weight. The weights sum to exactly 100, which is deliberate and slightly awkward to maintain: it means a reader can add the published table up by hand, and it means any change to a weight is visibly a redistribution rather than a quiet inflation. An earlier version of the checker had checks summing to 105, which made its score out of 100 not actually out of 100.

Model version 4.5. Content is the largest group, which is the direction several model versions of measurement have pointed in: what the CV says is where the recoverable points are.

The full check list with individual weights is on the ATS-friendly CV page and in the checker's own published table. The heaviest single check is Reading order (14 points); the lightest are Personal details and File hygiene at 3 each.

The two gates, and why they override the arithmetic

Two conditions cap the total regardless of what the weighted sum produces.

No text layer: capped at 25

A file with no extractable text cannot be assessed. Every other check would be scoring the absence of input rather than the quality of a CV, and thirteen meaningless numbers averaged together produce a meaningless fourteenth. The report is capped and says the one thing that matters. The text layer guide covers how a CV ends up in this state, usually by accident.

Critically scrambled reading order: capped at 69

A CV whose reading order is critically broken cannot be ready to send however good its bullets are, because the bullets were measured on text that arrives interleaved. Without the cap, a two-column test file scored 91, with 86 of those points describing a document that does not survive parsing. 69 is the top of the readable-but-leaking band, which is the literal situation - the words arrive, the order does not.

The two gates behave differently on purpose. The text-layer gate suppresses detail, because there is one finding and printing thirteen more would be noise. The reading-order gate stays generous: every check still runs and reports, and the columns finding recomputes its own "points if fixed" to include the lifted cap, so repairing the layout visibly pays back everything it cost.

The bands

Four bands, fixed, and the same definition drives the ring, the breakdown and every example on the site.

There is no band called "looks solid". It described no band and existed only to be reassuring, so it was removed.

A band is arithmetic; a critical finding is a fact about the file. When a report contains a critical finding, the state line says so rather than reporting the band, because a reassuring word above a critical problem is the exact failure this model is meant to avoid.

Checks that cannot be measured

A missing input is reported as unavailable, never scored as zero.

2 of the 14 checks - Reading order and Format traps - need per-glyph coordinates. A DOCX has no page geometry at all, so on a DOCX both are reported as unavailable and their combined 20 points are redistributed across the checks that could be measured.

The alternative - scoring them zero - would tell every DOCX user their layout is broken, which is wrong and which would undermine the credibility of the other twelve rows at the same time. The format guide sets out what each format makes measurable.

The findings matter more than the number

The report shows the 7 most severe findings and keeps the rest behind a control.

A wall of twenty findings is not a report, it is a reason to close the tab. Findings are ranked critical, then major, then minor, and each one carries what it costs, the evidence line from your own document, and what to do about it. The evidence line is the part worth your attention: it is a quotation from the extracted text, which is to say from the version of your CV that a system stores.

Two habits make the report useful rather than decorative. Read the extracted text before the score, because a surprise there explains half the findings. And treat the findings as a queue ordered by severity rather than by how easy each one is - the cheap fixes are satisfying and the critical one is the application.

What the number cannot tell you

Three things, stated plainly, because the category is not in the habit of stating them.

• It is not a prediction of an interview. It measures whether a document survives being read by software. Whether the content earns a conversation is a separate question and no parser has access to it.
• It is not a simulation of a named employer's system. Appliora reports how its own parser reads your file and flags failures common across systems. Nobody outside a given vendor can tell you what that vendor's extractor does, and a checker claiming to replicate one is guessing.
• It is not comparable to another tool's number. Different dimensions, different weights, and in most cases weights that are not published at all. Comparing two scores across two products is comparing two unrelated measurements that happen to share a scale.

What the number is good for is comparison against itself. The model is versioned - 4.5 at the time of writing - and the version is printed in the report, so the same file scored twice gives the same number and a change between two runs is a change in your document. The version is bumped on any scoring change, including a change to the parser, because what a document scores is the only definition of a scoring change that matters to the person reading the number.

Questions people ask

What is a good ATS score?

In Appliora's bands, 85 and above is ready to send, 70 to 84 means a few fixable things, 50 to 69 means the document is readable but leaking information, and below 50 means it needs work. There is no band called good, because a score is a readability measurement of one document rather than a grade.

Why did my score drop when I only changed the layout?

Because the content checks are measured on extracted text. A layout change alters what the parser recovers and in what order, so bullet quality, concrete skills and recognised sections can all move without a word being rewritten. That is the model working rather than failing.

Two checkers gave me very different scores. Which is right?

Neither is right in the sense you want. Each measures its own dimensions with its own weights, and most do not publish either. The question worth asking a checker is whether it will give the same file the same number twice, and whether it tells you which version of its model produced that number.

Does a high score mean I will get an interview?

No, and any tool claiming otherwise is selling something. A clean score means a category of mechanical failure has been removed from your application. Whether the content earns an interview is a separate question that no parser can answer.

Why does my DOCX report show fewer checks?

Two checks need per-glyph coordinates that a DOCX does not have, so they are reported as unavailable rather than scored as zero, and their weight is redistributed across what could be measured. Scoring a missing input as a failure would tell every DOCX user their layout is broken.