How CandidRanker scores candidates: the published 8-dimension methodology

The two scores, and why they are separate

CandidRanker produces two independent 0–100 numbers, and confusing them is the most common misreading of any screening tool. The match score answers: how well does this candidate fit this job description? The document quality score answers a different question entirely: does this file contain enough reliable, extractable information to support an evaluation at all?

They are separate because a strong candidate can submit a poor-quality resume, and a beautifully formatted resume can describe a poor fit. Collapsing the two into one number is how screening tools quietly reward formatting over substance. MinMaxHR keeps them apart so the recruiter can see which problem they are looking at.

The eight dimensions of the match score

• Skills: required and relevant skills identified from the job description and evidenced in the candidate's history.
• Tools: the technologies, platforms and equipment named in the role and found in the candidate's work.
• Experience: evidence of relevant work history and its depth against what the role asks for.
• Education: degree and education alignment, including supported degree variants and branches.
• Certifications: role-relevant certifications where present in the source documents.
• Projects: project evidence that supports fit for the specific role.
• Title similarity: how closely the candidate's role titles align with the target job.
• Semantic similarity: meaning-level overlap that catches relevant equivalences literal keyword matching misses.

Every dimension is visible on the candidate card in CandidRanker with the evidence behind it. A recruiter can always answer "why did this candidate score what they scored?" by reading the dimension, not by trusting the total.

Determinism: the property that makes rankings comparable

CandidRanker scoring is deterministic. The same job description and the same candidate documents produce the same score and the same ordering on every run. This is testable in a few minutes on the free plan: rank a role, note the order, re-run the ranking, and compare.

Determinism matters more than it sounds. Without it, a ranking cannot be compared across weeks, two recruiters cannot reproduce each other's results, and no audit can reconstruct what the system said at the time of a decision. If a screening tool cannot promise this, its scores are impressions rather than measurements.

Mandatory versus optional skills

MinMaxHR lets the recruiter distinguish must-have requirements from useful extras rather than treating every term in a job description as equally important. Mandatory skill coverage is weighted more heavily in the match score than optional coverage.

Crucially, a missing mandatory skill is surfaced to the recruiter as a visible gap. It is never used as a silent automatic filter. The candidate stays in the ranked list with the gap shown, and a human decides whether it is disqualifying. This is the difference between a decision-support system and a knockout filter.

Skill depth, aliases and organisational vocabulary

Counting keyword mentions rewards resume padding. Instead, CandidRanker weights each matched skill by skill depth: how long and how recently it was used, derived from the work-history timeline. Six years of current, daily use outranks the same word listed once against a course from 2019.

Boolean skill groups and alias matching let equivalent terms satisfy the same requirement, so a candidate is not penalised for writing "K8s" when the job description says "Kubernetes". Beyond generic equivalences, a workspace terminology catalog lets an organisation define its own vocabulary once (the internal system names, the industry shorthand, the process acronyms) and have it applied consistently to every future ranking. Suggested terms are mined from the organisation's own documents and require human approval before they affect any score.

Education is matched the same way: degree-aware matching interprets supported degree variants and branches rather than comparing exact strings, so an equivalent qualification expressed differently is not silently missed.

Why zeros are shown, and how gaps are treated

When a dimension scores zero, CandidRanker displays it with the reasoning rather than hiding it. A hidden zero is indistinguishable from a dimension that was never evaluated, and that ambiguity is exactly what makes a shortlist hard to defend later.

Employment gaps are handled by an explicit rule: they are surfaced as neutral facts for discussion and are never deducted from the match score. A career break is a conversation to have in an interview, not a penalty applied by software. Fair-evaluation reviewers ask about this specifically; in the MinMaxHR methodology the answer is a design decision, not a configuration option.

Where the human sits in the workflow

• The recruiter defines the role criteria the whole evaluation runs against.
• The recruiter reviews anything CandidRanker could not extract or classify confidently, and corrections are kept in history.
• The recruiter reads the ranked list and the per-dimension evidence.
• The recruiter records every decision (shortlist, select, reject, or clear the current decision) with a written reason, under their own name. There is no 'pending' state and no automated rejection.
• Changes to scoring weights, filters, skills, aliases and terminology are logged with the person, the time, the before and after values, and a reason.

What this methodology does not claim

A ranking is an early-stage prioritisation of recruiter attention. It does not validate technical capability, replace interviews, references or assessments, or predict job performance. It amplifies the criteria it is given, which means unclear criteria produce confident-looking noise. The fix for that is a better job description, not a different score.

Two capabilities are on the roadmap and stated in the future tense on purpose: a document quality gate that will route weak inputs to human review against a configurable threshold, and one-click data-subject export. Deletion of extracted data, rankings and stored files is available today.