How matching works
Every score is the sum of named factors.
Skill overlap, taxonomy fit and trust each add points you can check. No black box, no opaque machine learning. This page is the receipt.
0
Applications scored
Live
0
AI insights generated
Live
Try it yourself
Pick role and candidate inputs; the playground runs the live algorithm with the platform defaults. The inputs live in the URL, so you can share the result.
Open the playground
Six published weights
Every score comes from these six factors. Employers can tune them within the ranges shown; everyone else gets the defaults.
| Factor | Default | Range | What it does |
|---|---|---|---|
Skill overlapskill_overlap |
62 | 0–100 | Maximum points from candidate-vs-role skill match. Scaled by proportion of required+preferred skills covered. |
Industry matchindustry_bonus |
12 | 0–30 | Added when the candidate lists the role's industry as a sector. |
Sub-industry matchsub_industry_bonus |
10 | 0–30 | Added when the candidate's history covers the role's sub-industry. |
Department matchdepartment_bonus |
8 | 0–30 | Added when the candidate has worked in the role's department previously. |
Trust signal divisortrust_bonus_divisor |
5 | 1–10 | Divides the candidate's trust score to compute the trust component. Lower divisor = more impact per trust point. |
Trust signal captrust_bonus_cap |
18 | 0–30 | Hard maximum on the trust signal's contribution to the final score. |
How scores actually fall
The real spread of 0 scored applications, in ten bands. The highlighted band (80 and up) is what we call a strong match.
Scores by role category
The same scores, split by kind of role, so anyone can question how results vary. Groups under three applications are withheld: small numbers can identify people.
Nothing to split yet
The panel needs at least one scored application.
What this is, and isn't
What it is: a transparency tool. It shows how scores spread across kinds of role, so anyone can spot a pattern and ask us about it.
What it isn't: a fairness claim. A real fairness audit needs careful design and qualified review. Candidate demographics are deliberately out of scope here.
What's changed, and when
Every algorithm change, published. If you were scored before and after one of these dates, this explains the difference.
-
May 2026
Algorithm changelog moved to a model + staff editor
The published changelog is no longer hand-maintained in code. The content team edits entries directly at /ai/changelog/ (staff-only). Existing entries were migrated unchanged.
-
May 2026
Live weight tuner added to /ai/command-centre/
Sliders on the page let any visitor adjust each tunable weight and see the worked-example breakdown recompute in real time. Same algorithm as the live shortlist, run in the browser using the factor inputs already on the row.
-
May 2026
Live personalisation added to /ai/command-centre/
Logged-in candidates now see their average score + a playground prefilled with their actual profile. Logged-in employers see their tuned weights against the platform defaults, side by side.
Earlier changes 4
-
May 2026
Scoring playground added to /ai/command-centre/
Public can now run the scoring algorithm against arbitrary inputs and see the full factor breakdown. Same algorithm as the live shortlist — no demo data, no synthetic weights.
-
May 2026
Score-distribution histogram replaces the mean stat
Single-number 'mean Pilot match' replaced by a 10-bin histogram of real application scores. Honest distribution instead of a misleading central tendency.
-
May 2026
Department-match bonus added (default 8 points)
Existing role-vs-candidate match now also rewards prior department experience. Per-employer tunable; defaults are conservative.
-
Mar 2026
Per-employer MatchWeights tuning made GA
Employers can override the platform-default weights within audited bounds. Defaults stay the publishing contract — tuning is opt-in, not implicit.