How we rank jobs for you
arkerah puts the jobs you can be shown in order by how close each advert is to what we know about you: what you can do, and what you are aiming at. The order is the ranking, so there is no match percentage. A few stated rules then leave some weaker jobs out of your list, each one says so, and those jobs stay in Every job here.
What does arkerah know about you when it ranks?
Everything we know about you is a list of small facts, such as a skill, a role you held or a city you would move to. They come from your CV, from the onboarding conversation, from a few fixed questions with real answer controls, and from the roles, sectors and places you confirm.
Each fact is either proposed, which means we read it from something you gave us, or confirmed, which means you said yes to it. Both count for ranking, because most of what anyone tells us arrives as a document rather than as a list of confirmations. A fact you rejected does not count.
How is a job's place in your list decided?
A language model turns each advert, and two slices of your facts, into lists of numbers that point in a direction. Texts that mean similar things point in similar directions. One slice is what you can do: your skills, your experience and your headline. The other is what you are aiming at: the roles, sectors and places you want, what drives you, and the jobs you said yes to in the ten-job step of onboarding.
A job's score is how closely its advert points the same way as those two, weighted 55 per cent for what you can do and 45 per cent for what you are aiming at. Those two weights are a guess, written down as one in the code: nothing has validated them yet, because that needs many people's verdicts on real jobs and we do not have enough of them.
The score is used for one thing only, putting jobs in order. The number in front of a job, such as #3, is its place in your list for the area you are looking at, and the dot beside it is the same place drawn as a colour, green at the top and red at the bottom. When what we know about you changes, your two directions are rebuilt in the background, so a fact you add moves your list shortly afterwards.
| Rule | What it does | Where the job goes |
|---|---|---|
| Closeness to your profile | Orders every job by its score | Nowhere: it only reorders |
| The advert does not state its level, and you stated yours | A small step down the list | Lower in your list, and its tile says the advert does not say which level |
| A kind of role outside the ones your profile points at | A small step down, and left out when it also scores below your own median in that country | Every job here, and Why not names the reason |
| The weakest tenth of your jobs in each country | Left out of your list | Every job here, and your list says how many |
| A kind of role measured to be almost never one of the kinds you confirmed, on adverts read by the model version that was measured | Left out of your list | Why not, where This is my kind of job brings it back |
Your own filters, such as the languages you read, your level and the companies you turned down, decide which jobs can be in your list at all before any of this runs. How arkerah decides what a corporate job is explains those in full.
Why is there no match percentage?
Because the score is not on a scale you could read as one. It is only meaningful next to your other scores. Measured across eight stored profiles on 12 September 2026, the best score any one profile reached anywhere in the corpus ran from 0.55 to 0.70, so a fixed bar would have shown hundreds of jobs to one person and none to another.
That is why every rule that leaves a job out is measured against your own scores, never against a bar shared between people, and why we show a place in a list rather than a number that looks like a grade. A band such as strong, fair or weak match was designed and deliberately not built, for the same reason.
What does Why this job show?
On a job's tile, Why this job opens the chat on that job and asks it for you. The answer starts with the sentence on the tile, then says which kinds of role you told us you look for. It also names any of your own filters that hold the job back.
Under the answer, a short list headed Why it is on your list says what we hold about the job, one line each: its place in your list, such as #3 of 120 in Amsterdam; the kind of role we file it under; its level next to the level you told us, or that the advert does not say which level; whether it is in a place you said you want, or inside the distance you said you would travel; and whether it is written in a language you read.
A line is shown only when we hold what it needs, and nothing is guessed to fill one. The kind of role is the label we already store for each advert, the one the family rules use. When that label was written by an older version of our reading, or with low confidence, the line says we are not certain of the kind instead of naming one.
Beside the list, the kinds of role you told us you look for are shown as chips, and the one the job is filed under is highlighted when it is one of yours. If you have not confirmed a set, the chips show the one we read from your roles and history, and say you have not confirmed it. Nothing on it is a percentage or a score, and it never says you would get the job.
We also tried a model that split each advert's work into shares, such as 70 per cent data and 20 per cent finance. Three versions of it were run on the same four adverts, and each version fixed one and broke another, so it was dropped rather than shown.
When does arkerah not rank at all?
Until we know you well enough. Three things have to be true together: what we know covers at least 65 of 100 points across ten areas such as your history, your skills and what you want; we know at least one place where you may legally work; and you have answered at least five of the onboarding questions in your own words. Until then, a job's tile says Not ranked for you.
What are the limits of this ranking?
- It answers whether a job is the kind of job you want, not whether you would get it. Nothing compares the years of experience an advert asks for with yours yet.
- The 55 and 45 per cent weights are unvalidated. They change only once enough people have given verdicts on real jobs to test them.
- The kinds of role behind the family rules were graded by an AI model reading each advert against a written rubric, not yet by a person.
Questions
Is the number on a job a score?
No. It is the job's place in your list for the area you are looking at. The score behind it only puts jobs in order and cannot be compared between two people.
Does Why it is on your list mean I would get the job?
No. Each line is something we hold about the job next to something you told us, such as its level and yours. It says why the job is in your list, not whether you would be hired.
Why did a job outside my field rank high?
Because its advert is close to the rest of your profile, such as your skills and experience. Ask Why this job on its tile: the answer lists the kind of role, the level, the place and the language we hold for it, next to what you told us.
How do I change my ranking?
Tell the chat what has changed, or confirm the kinds of role you want. Your list is rebuilt from what we know about you shortly afterwards.