Ranking parameters
Disclosure under Art. 5 of Regulation (EU) 2019/1150 (P2B)
Aptora puts results in order in two directions: talent search orders candidate profiles for a job listing, job matching orders job listings for a profile. This page describes which parameters decide that order, and why they weigh against each other the way they do.
Both directions are built from the same parts, so every section here applies to both. What follows describes the parameters and their relative weight, not the algorithm itself — Art. 5 (6) expressly does not ask for that, and publishing it would mostly help whoever wants to game the results.
The main parameters
In order of weight:
1. Skill coverage
The main parameter and the starting value of every score: how many of the required skills a profile evidences, weighted by how central the listing makes each one. A skill the profile states explicitly counts most; one evidenced only in the profile text counts less; one reached through a curated edge in the skill graph — a broader category or a more specific variant — counts least. Only reviewed edges count.
2. Depth of coverage
How many skills are evidenced in absolute terms, not only as a share. A listing with three requirements that a profile meets in full says less than one with twelve requirements of which ten are met. Depth lifts the score, up to a saturation point.
3. Job title proximity
How close the listing's title is to the title in the profile: identical — synonyms included — or related through a curated relation in our title taxonomy. The title can close no more than part of the distance to a full score that skill coverage left open, currently half of it. It cannot stand in for weak skill coverage.
4. Seniority
Junior, mid-level, senior, lead and so on. It adjusts the title's contribution only, never skill coverage. If either side does not state a seniority, that costs nothing.
5. Refinement you ask for
Happens only when it is requested: refining a search or a match with your own free text lets a second pass judge the hits against it. That contribution is capped — currently about a third — and it can reorder the list, but it cannot remove a hit from it.
Equal scores are settled by title proximity and skill score, then by a fixed key that is the same for everyone. There is no random component: the same search returns the same order. A result is stored and reused until its inputs or our weights change.
Why this weighting
Skills are the most reliable statement about whether somebody can do a job. Job titles are not — the same work is named differently in two companies, and the same title means two different jobs. That is why skill coverage is the starting value and everything else is an addition to it.
Every further parameter can only lift a score, never lower it. Something left unstated — no title, no seniority — therefore costs nothing; it simply earns nothing. The one exception is the refinement you ask for, whose whole purpose is to reorder.
The exact weights are operating parameters and are adjusted when a criterion turns out to weigh too much or too little. Their roles do not change with them: what is the starting value, what is an addition, and what does not rank at all. If more than a number changes, this page changes with it.
What selects, but does not sort
Some criteria only decide whether a profile or a listing appears in the result set at all. They have no effect on the position within it:
- How recent a listing is and how long it runs: published within the last months, not expired, publicly visible.
- The candidate's consent to being discoverable. Without it, a profile is in no search at all.
- A minimum completeness of the profile, a minimum number of skills among it — a half-filled profile is not offered as a hit.
- The filters the searcher sets: country, share of remote work, day rate, languages, availability, how recently the profile was updated.
- The searcher's own team, whose profiles do not appear in its own search.
A refinement you ask for may read these details. They then reach the order indirectly, through that pass's capped contribution — because that is what was asked for.
No remuneration influences placement
There are no paid placements on Aptora: no sponsored hits, no ads between the results, no way to move a profile or a job listing up in exchange for money, neither directly nor indirectly. The statement required by Art. 5 (3) is therefore, for both directions: no.
Plans decide access and quota: whether a workspace may use talent search, and how often it may ask for a refinement. They never decide the order. The ranking code knows neither plan nor subscription, and a workspace without quota gets the same order as one with it.
Reviews do not influence ranking
Reputation data is not one of the inputs — neither the scores nor the number of reviews. An automated test holds us to that on every change: it fails the build the moment a search or matching query reads so much as one reputation table. The long version is in our reviews policy.
No personalisation, no case-by-case placement
We do not personalise results by your past behaviour, and we do not test variants of the order against each other. The only personalisation is the one you ask for yourself.
We curate the skill and title graph editorially: which skill sits under which, which titles are related. That work applies to all results equally; no single profile or listing is moved up or down in it. A hand-made order exists only in a team's shortlist — its own working list after the search, not in the search results.
Questions about this page
Use the contact address in our imprint.