Why a percentage is not a decision

Most tools that grade a CV against a posting are doing a version of the same thing: extracting terms from the posting, looking for those terms in the document, and expressing the overlap as a proportion. Some weight the terms. Some group synonyms. The shape is the same.

That is a real measurement. It is not a measurement of whether you can do the job, or of whether a human screener will advance you. It is a measurement of how much of the posting's vocabulary your document reuses.

Which is worth knowing, because there is a legitimate use for it — some employers do run automated keyword screens, and lexical overlap is a fair proxy for passing one. The problem is the proxy being presented as a verdict.

Accurate about what? That is not a dodge; it is the whole difficulty. A number is accurate relative to something, and these numbers are rarely told what they are predicting.

Three properties should make you cautious. The scale is uncalibrated: nothing tells you what separates a good reading from a poor one, so the threshold you are aiming at was chosen by whoever built the tool. It is unanchored to outcomes, because the tool cannot see who else applied or who was hired. And it is unstable across postings, since the same CV measured against two descriptions of the same job will move with the employer's wording rather than with anything about you.

That last one is the practical test. If a number swings when the job has not changed, it is describing the posting, not you.

The reason this matters beyond accuracy is that a visible number changes behaviour. Give someone a figure and a way to raise it, and they will raise it.

Raising it means importing the posting's phrasing. Sometimes that is a genuine improvement — you did the work and the posting's word for it is better than yours. Often it is not: the wording gets closer to the posting while the evidence underneath stays exactly where it was, and the document drifts towards the vocabulary of a job you have not done.

The cost lands later. A CV written to a score reads fluently and dies in the first interview, where somebody asks for the story behind a line that was optimised rather than lived.

The useful question is not how well you score. It is which requirements you can evidence, which you cannot, and which lines of the posting produced that reading.

Judge a tool on whether it can show you its working. Three things are reasonable to expect:

• The posting's own words. If it says a requirement is unmet, it should point at the line, so you can see whether it keyed on an accountability or on boilerplate.
• A verdict you can overrule. You know things about your history that no document contains, and the tool cannot.
• Its mistakes left visible. A tool that hides what it ranked low is asking you to trust a judgement it will not show you.

None of that requires a number, and a number cannot substitute for any of it. How to check your CV against a job description is the same comparison done by hand, which is worth doing once so you know what a tool is doing for you.

Huntly does not produce a match score, and it is not an omission we intend to fix. Internally the matcher orders postings; what it publishes is a verdict with the lines of the posting that produced it, because that is the part you can argue with.

The honest trade is that a verdict is harder to skim than a figure. A percentage is a very efficient way of telling you something that is not quite true, and we would rather be slower and checkable. What a match verdict can and cannot tell you sets out the limits of ours.