Scoring methodology
Astris HR rates each candidate-to-role match on eight clear factors and combines them into one score from 0 to 100. Likelihood of staying in the job is one of those factors, so long-term fit counts from the start. Two safeguards keep a poor fit from reading as a strong one. Every score can be traced back to the factors that produced it. The figures on this page come from internal testing on simulated data; results from real placements are still being collected.
The eight match factors
Every score is built from eight factors, each carrying a set share of the total. Skills matter most. Likelihood of staying is one of the eight factors in its own right, not an afterthought, so long-term fit is built into the score.
| Skills · most important | Compares the candidate's skills to the role. Credit is given both for an exact skill match and for closely related skills that mean the same thing under a different name, so transferable experience is not missed on wording alone. |
| Experience · high | Weighs relevant years of experience, with more credit for recent work in the same kind of role than for length of service alone. |
| Logistics · moderate | Combines commute distance, access to transit, and schedule and shift fit into a single practical-fit factor. |
| Retention · moderate | The estimated chance the candidate stays in the role for at least 90 days, carried straight into the score as its own factor. |
| Language · lighter | Checks the candidate's English level against what the role requires, with a small boost for candidates who speak more than one language. |
| Education · light | Compares education level to the role. Deliberately given a low share so a credential gap does not crowd out proven ability. |
| Culture · light | How well the candidate and the employer fit each other, measured across six areas, when signals from both sides are available. |
| Interview · lightest | A read on interview readiness. The lightest factor, used to break ties rather than to drive the score. |
Two safeguards
A candidate who has only 1 of 30 required skills should not score high just because the commute and language happen to line up — and an analyst should never surface as a strong match for a driver role. Two safeguards set an upper limit on the score before it is finalized.
Coverage is the share of the role's required skills the candidate actually has. The fewer of those skills the candidate has, the lower the ceiling on the score. A candidate with none of the required skills is held well below a strong match; the ceiling lifts steadily as coverage rises.
A candidate who has only a small fraction of the required skills cannot reach strong-match territory, no matter how well the other factors line up.
When the candidate has every required skill, the ceiling is removed and the score can reach its full value.
When the candidate's line of work is a different field from the role's — an analyst versus a driver, for example — the score is held down so a wrong-field match cannot read as a strong one.
The score, the skill-coverage limit, and the wrong-field limit are compared, and the lowest of the three is used. The result is kept within the 0–100 range.
Every match records the coverage figure, the skill-coverage limit, whether the role is in a different field, and the wrong-field limit, so the recruiter can see exactly which safeguard held the score down.
Recognizing related and cross-language skills
Skill matching goes beyond exact wording. Astris HR understands when two skills mean the same thing, even when they are described differently or named in another language, and it credits experience that carries over between related occupations.
| Exact match | When the candidate clearly has a required skill, it earns full credit. This is the reliable baseline of the skills factor. |
| Related-skill match | Astris HR also recognizes skills that mean the same thing under different wording. When a candidate's skill is close enough in meaning to a required one, it counts toward coverage, so synonyms and re-phrasings are picked up even without an exact word match. |
| Across languages | A skill named in one language can earn credit for its equivalent in another, because Astris HR compares skills by meaning rather than by the words used. This matters for candidates whose experience is documented in their first language. |
| Closer means more credit | The strength of each connection is learned from data, so a near-identical pair of skills earns more credit than a loosely related one. |
| Skill-gap read-out | Skills the candidate is missing are shown alongside an estimate of how many months of preparation would close the gap, so a gap reads as a plan rather than only a shortfall. |
How the skills factor is worked out
The skills factor is built to catch as much relevant experience as possible. It counts how many of the role's required skills the candidate has — whether through an exact match or a closely related one — and turns that into a coverage figure.
| Skills matched | The count of required skills the candidate has, either as an exact match or as a closely related skill that means the same thing. |
| Skills score | The share of required skills the candidate has, expressed on a 0–100 scale. |
| Transferable bonus | A small boost when the candidate brings at least two useful skills beyond those the role requires, up to the maximum score. |
| Feeds the safeguard | The same coverage figure drives the skill-coverage safeguard, so the skills score and the safeguard always agree. |
| No required skills listed | When a role lists no required skills, the candidate receives a neutral score for this factor rather than a penalty. |
How much weight a credential carries
A credential earned long ago, or issued somewhere its records are hard to check, carries less weight as evidence — but a genuine credential that simply cannot be verified never counts for nothing.
| Age of the credential | A credential's weight eases down gradually with the years since it was earned, so older records are discounted smoothly rather than in sudden steps. |
| How checkable it is | Where the credential was issued affects its starting weight. A source that cannot be identified or is unlisted starts at a cautious default. |
| A floor, never zero | The weight can never fall all the way to nothing, so a refugee with a genuine but unverifiable degree still carries a baseline of evidence. |
| Optional live check | When available, Astris HR can check whether the issuing body's records are actually reachable online, which informs how much weight the credential carries — it does not by itself confirm the credential is authentic. |
What if you trained them?
For every match, Astris HR also asks which single missing skill would raise the candidate's score the most — and answers by actually testing each one, not by an estimate.
For each missing required skill, Astris HR takes the candidate as they are, adds that one skill, and re-scores the match from scratch to see how much the score actually rises. The missing skills are then ranked by the gain they produce.
Because each gain comes from a real re-score, it reflects the safeguards. A skill the candidate already covers through a closely related one may add almost nothing, while a skill that pushes coverage past a safeguard limit can add far more than a simple estimate would suggest.
Astris HR adds up an estimated preparation time for each missing skill, reduced by what the candidate already partly holds through related experience, to give a months-to-ready figure for the role.
This re-scoring approach and the months-to-ready estimate are documented methods in the Astris HR patent strategy.
How the likelihood-of-staying factor is estimated
The chance a candidate stays is not a single guess. Four independent methods each estimate the likelihood of staying at least 90 days, and their estimates are combined. How much each method counts is adjusted over time based on how well it predicted real placement outcomes. The accuracy figures below come from internal testing on simulated data; results from real placements are still being collected.
| Timing-of-departure method | Estimates not just whether but when a candidate is likely to leave, based on five factors that matter for people rebuilding after displacement — how quickly they are picking up the language, access to transit, distance from their own community, time since arrival, and whether their training matches the role. In internal testing on simulated data it ranked outcomes correctly about 73% of the time; results from real placements are still being collected. |
| Pattern-learning method | Learns patterns from many past examples to estimate the chance of staying, with a simple rule-based backup. In internal testing on simulated data it reached about 81% accuracy; results from real placements are still being collected. |
| Transparent statistical method | A straightforward, easy-to-read statistical estimate on the same factors, kept because its reasoning is easy to inspect. In internal testing on simulated data it reached about 83% accuracy; results from real placements are still being collected. |
| Neural-network method | A neural network trained on the same factors to estimate the chance of staying. In internal testing on simulated data it reached about 80% accuracy; results from real placements are still being collected. |
| How the four are combined | The four estimates are blended into one. As real placement outcomes come in, Astris HR gives more weight to the methods that predicted well and less to those that did not, and a challenger method is only promoted after it clearly proves itself. No single method is ever fully silenced. |
| Caseworker grouping (context only) | Astris HR can also group candidates with similar support needs to help caseworkers plan. This grouping is purely for context — it never contributes to the likelihood-of-staying estimate and never scores anyone for matching. |
Recommendation tiers
Fixed score ranges that sort every match into a tier. The ranges stay the same from one release to the next.
Top of the list. Recruiters review these first.
Worth interviewing. Most placements come from here.
Worth a conversation, but expect a real bar to clear.
Possible when the pipeline is empty and you're willing to invest in training. The skill-coverage safeguard usually holds low-coverage matches in this range.
Shown for the record only. We don't hide candidates; we tell you why they scored low.
Every gap report leads with skill coverage, so recruiters can see why a score is what it is.
What each match record keeps
Every match saves enough detail to rebuild the score exactly and to power the 'what if you trained them' projection.
| All eight factor scores | Each of the eight factor scores is saved, along with the share each one carried, so the full score can be rebuilt. |
| Final score and confidence | The final 0–100 score, a measure of how confident Astris HR is in it, and a flag for manual review when confidence is low. |
| Coverage and both safeguards | The coverage figure, the skill-coverage limit, whether the role is in a different field, and the wrong-field limit — showing exactly which safeguard held the score down. |
| Plain-language explanation | A plain-text summary built from the factors that mattered most to this score, so the headline reason is always the same for the same inputs. |
| Training scenarios and gaps | The ranked list of missing skills with the measured gain each would bring, plus the months-to-ready estimate, to drive the improvement plan. |
| Scoring version | The version of the scoring used is saved, so two runs can be compared whenever the scoring changes. |
See Astris HR in your workflow
Run it against your candidate base or a sample JD. We'll set up a tenant in under a day and show you the matches your existing tools are missing.
Request a demo