What if a job doesn't post a salary?
That sub-score is dropped, not guessed.
When pay is unknown on either side, comp is excluded from the average entirely. The same happens when a job description lists no domains, or when its requirements could not be parsed cleanly. The score is then built from whatever is genuinely known.
This sounds like a detail and is actually the whole design philosophy in miniature. A missing input reduces how much the score can tell you. It does not reduce the score. Filling the gap with an assumption would produce a more confident-looking number that means less, which is the same failure the Calibration Engine exists to prevent one layer up.
Some jobs come back with no score at all
Those cards carry a pill reading "Not scored". It is doing something, rather than failing at something.
Most match percentages in this market are a similarity measurement between two blocks of text. A measurement like that is defined for any two pieces of text you hand it, which means it can never come back empty. Give it a posting with nothing useful in it and you still get a confident looking number.
This one can come back empty. When a posting carries no recognised skills there is nothing to weigh your experience against, so the card says that, and says what fixes it: open the job and the full description gets read properly instead. When your Vault is the thing that is still empty, it says that instead, because that is what is actually in the way.
The same restraint has a quieter form. A score built only from a posting's tags is capped at 88 and cannot reach the top of the scale at all, because that kind of estimate has never read a requirement in the job description and is in no position to claim a perfect match. Only a score that has read the posting properly is allowed to.
There is a real cost to this. Now and then a card here shows no number where another tool would show you 91%. A number that can be absent is worth more than one that is always present.
Nothing about your score depends on who else applied
Job boards compute a different thing, and the difference is worth knowing before you read one of their badges.
A board's match signal generally places you inside the pool of people who applied to that posting, and recomputes as that pool changes. It can fall this afternoon because three strong candidates arrived this morning, with nothing about you having changed. That is genuinely useful information about a marketplace. It is not a statement about whether the job suits you.
Fit reads your Vault against this job's requirements and reads nothing else. It does not know how many people applied, and it does not move when more of them do. Run it twice a month apart on an untouched Vault and an untouched posting and you get the same number both times.
It also has nothing riding on your answer. A 1.4 that talks you out of spending an evening on a job is a good outcome, and the score is built to be able to say it.
Is the score generated by AI?
No. Fit is computed, not generated, and that is deliberate.
It is ordinary deterministic code: requirement coverage, domain overlap, seniority and pay, combined by fixed weights. The same job against the same Vault always returns the same number, and every factor behind it is listed.
The reason matters. A decision aid you cannot audit is one you cannot sensibly disagree with, and disagreeing with this one is often correct. You know things the score does not: that you have wanted to work at this company for years, that the hiring manager is a former colleague, that the missing requirement is something you could learn in a fortnight. Fit is there to tell you what the evidence says, clearly enough that you can knowingly overrule it.
Does it change if I'm switching careers?
The framing changes. The number does not.
JobShifu tracks which of your target roles are pivots rather than continuations, and a pivot changes how the results are presented: gaps are described differently, and existing skills that carry across are surfaced as transferable rather than counted as misses.
What it will not do is inflate the score because the job matters to you. A pivot genuinely is harder, and a tool that hid that would be flattering you rather than helping. The honest version tells you the fit is a stretch and shows you which of your existing skills to lead with, which is the combination you can actually act on.