What is an ATS quality score?

An ATS quality score (0–100) measures how complete and machine-readable a resume is (contact details, skills, work history, education) computed from the parsed fields. It tells recruiters how much to trust the parse, and tells candidates how ATS-friendly their resume is.

What is semantic fit or vector similarity in resume screening?

Semantic fit compares the meaning of a resume and a JD, not just their words, using vector embeddings. It catches equivalent phrasing ('K8s' vs 'Kubernetes', 'led a team' vs 'engineering manager') that keyword filters miss.

What is skill depth weighting?

Instead of counting skill mentions, CandidRanker weights each matched skill by how long and how recently it was used, derived from the work-history timeline. Deep, current experience outranks a keyword listed once in an old course.

How are employment gaps treated in candidate ranking?

A gap of six months or more between work-history dates is shown as evidence in the candidate drawer, an interview talking point. It never reduces a candidate's rank. Fair-evaluation reviewers ask about exactly this; the answer is designed in.

What is a decision state and who sets it?

A decision state (shortlisted, selected, or rejected) is set only by a human recruiter, requires a reason, and lands in CandidRanker's audit log under the recruiter's name. AI never sets a decision state; that is the human-in-the-loop guarantee.

What is a recruitment audit log or decision trail?

The time-ordered record of every workflow event: criteria used, rankings produced, reasons given, decisions made, by whom, and when. It turns 'why was this candidate rejected?' from an awkward reconstruction into a lookup, the core of DPDP/GDPR-era defensibility.

What is a workspace or tenant in recruiting software?

A workspace is a fully isolated container for one team's or one client's candidates, JDs, and decisions. Nothing crosses workspace boundaries. The isolation model that lets agencies serve multiple clients from one account safely.

What is MCP (Model Context Protocol) in recruiting software?

MCP is the open standard that lets AI assistants use external tools. CandidRanker's MCP server exposes recruiting operations (list jobs, rank candidates, explain a ranking, pull a report) so assistants like Claude, or any LLM that supports MCP, can drive the hiring workflow conversationally under OAuth scoped to one workspace.

What is the difference between an API key and MCP access?

Both reach the same governed operations. API keys are workspace-scoped credentials for the REST API, suited to internal tools and integrations. MCP access uses OAuth consent per workspace and is built for AI assistants. Both inherit the same isolation, logging, and security.

What does dedup mean in resume processing?

Duplicate detection by content hash: the same resume uploaded twice is caught before any AI runs. No wasted quota, no double-counted candidates, no skewed funnel metrics.

What is a parse status and what happens when parsing fails?

Every uploaded file carries a parse status: pending, parsed, or parse_failed. Low-confidence parses are flagged for human review; failed files can be re-parsed from the stored copy without re-uploading. Transient failures retry automatically.

What is a JD lifecycle state?

A job description is active by default and marked fulfilled when the role closes. Fulfilled JDs disappear from ranking pickers but keep their full history: candidates, rankings, and decisions stay reviewable.

What is a hiring funnel in recruiting analytics?

The aggregated view of the decision log: how many candidates were uploaded, parsed, shortlisted, selected, or rejected, plus decisions per week, median time-to-decision, and time-to-fill. In CandidRanker it builds itself from the audit trail, no spreadsheet.

What is a Hiring Decision System?

MinMaxHR's category term for the evaluation layer that sits next to an ATS. The ATS manages workflow: postings, stages, records. The Hiring Decision System handles the decision step: ranking candidates against roles with explainable scores, and recording the human decisions that follow.

What does explainable ranking mean?

Every score comes with recruiter-visible reasoning: which criteria matched, which didn't, and the per-dimension evidence. No black-box numbers, which is what makes a ranking defensible to hiring managers, candidates, and auditors.

← All resources

Glossary · Recruiting terms

The CandidRanker glossary: every term in AI candidate ranking, explained

Published 9 July 2026 · Updated 11 August 2026

Sixteen terms, defined the way they actually work inside CandidRanker. Learn them once and every screen in the product reads itself, as does every conversation about AI recruiting.

This glossary defines the core vocabulary of AI candidate ranking: the scores, the workflow states, the governance records, and the integration standards like MCP that connect recruiting data to AI assistants.

The scores

Two numbers appear throughout CandidRanker, and they answer different questions. The ATS quality score (0–100) asks: how complete and machine-readable is this resume? The match score (0–100) asks: how well does this candidate fit this specific job description? A beautifully formatted resume can be a poor fit, and a messy scan can hide a top candidate, which is why both scores exist.

The evaluation machinery

The match score decomposes across eight weighted dimensions: skills, experience, tools, education, title, certifications, projects, and semantic fit. Semantic fit is powered by vector similarity: a meaning-level comparison between the candidate and the JD that catches 'K8s' when the JD says 'Kubernetes'. Skill depth weights each matched skill by how long and how recently it was used, so six years of daily Python outranks one course in 2019.

All of this machinery is visible on every candidate CandidRanker ranks. Nothing scores in the dark.

The workflow states

Every file moves through a parse pipeline (pending → parsed → parse_failed, with a retry after transient failures). Every candidate carries a decision state (shortlisted, selected, or rejected) set only by a human recruiter, with a reason, never by the AI. Every JD has a lifecycle state: active or fulfilled; fulfilled roles stop appearing in ranking pickers but keep their history.

These states are CandidRanker's on-screen vocabulary: the same words recruiters see next to every file, candidate, and role.

The governance records

The audit log is the time-ordered record behind everything: who ranked what, against which criteria, who decided, and why. The analytics funnel aggregates that log into the numbers leadership asks for (decisions per week, time-to-decision, time-to-fill) with no manual tracking. Dedup catches duplicate resumes by content hash before any AI runs, so batches and metrics stay clean.

Together these records are CandidRanker's governance layer: the log, the funnel, and clean inputs.

The platform and integration layer

A workspace is an isolated tenant: its candidates, JDs, and decisions never cross into another workspace, which is how agencies keep clients separated. API keys are workspace-scoped credentials for the REST API. And MCP, Model Context Protocol, is the open standard that lets AI assistants like Claude, and any LLM that supports it, operate CandidRanker conversationally: list jobs, rank candidates, explain rankings, pull reports.

Why the vocabulary matters

Shared terms make shortlist reviews faster, everyone reads the same score the same way.
Precise states (parsed, shortlisted, fulfilled) make the audit trail unambiguous.
Knowing the integration terms (API key, MCP) turns 'can it connect?' into a five-minute answer.

It is the vocabulary CandidRanker uses on every screen, in every report, and in its MCP tools.

Frequently asked questions

What is a match score in AI recruiting?: A match score (0–100 in CandidRanker) measures how well a candidate fits a specific job description. It decomposes across eight weighted dimensions (skills, experience, tools, education, title, certifications, projects, and semantic fit) and is deterministic: same resume, same JD, same score, every time.
What is an ATS quality score?: An ATS quality score (0–100) measures how complete and machine-readable a resume is (contact details, skills, work history, education) computed from the parsed fields. It tells recruiters how much to trust the parse, and tells candidates how ATS-friendly their resume is.
What is semantic fit or vector similarity in resume screening?: Semantic fit compares the meaning of a resume and a JD, not just their words, using vector embeddings. It catches equivalent phrasing ('K8s' vs 'Kubernetes', 'led a team' vs 'engineering manager') that keyword filters miss.
What is skill depth weighting?: Instead of counting skill mentions, CandidRanker weights each matched skill by how long and how recently it was used, derived from the work-history timeline. Deep, current experience outranks a keyword listed once in an old course.
How are employment gaps treated in candidate ranking?: A gap of six months or more between work-history dates is shown as evidence in the candidate drawer, an interview talking point. It never reduces a candidate's rank. Fair-evaluation reviewers ask about exactly this; the answer is designed in.
What is a decision state and who sets it?: A decision state (shortlisted, selected, or rejected) is set only by a human recruiter, requires a reason, and lands in CandidRanker's audit log under the recruiter's name. AI never sets a decision state; that is the human-in-the-loop guarantee.
What is a recruitment audit log or decision trail?: The time-ordered record of every workflow event: criteria used, rankings produced, reasons given, decisions made, by whom, and when. It turns 'why was this candidate rejected?' from an awkward reconstruction into a lookup, the core of DPDP/GDPR-era defensibility.
What is a workspace or tenant in recruiting software?: A workspace is a fully isolated container for one team's or one client's candidates, JDs, and decisions. Nothing crosses workspace boundaries. The isolation model that lets agencies serve multiple clients from one account safely.
What is MCP (Model Context Protocol) in recruiting software?: MCP is the open standard that lets AI assistants use external tools. CandidRanker's MCP server exposes recruiting operations (list jobs, rank candidates, explain a ranking, pull a report) so assistants like Claude, or any LLM that supports MCP, can drive the hiring workflow conversationally under OAuth scoped to one workspace.
What is the difference between an API key and MCP access?: Both reach the same governed operations. API keys are workspace-scoped credentials for the REST API, suited to internal tools and integrations. MCP access uses OAuth consent per workspace and is built for AI assistants. Both inherit the same isolation, logging, and security.
What does dedup mean in resume processing?: Duplicate detection by content hash: the same resume uploaded twice is caught before any AI runs. No wasted quota, no double-counted candidates, no skewed funnel metrics.
What is a parse status and what happens when parsing fails?: Every uploaded file carries a parse status: pending, parsed, or parse_failed. Low-confidence parses are flagged for human review; failed files can be re-parsed from the stored copy without re-uploading. Transient failures retry automatically.
What is a JD lifecycle state?: A job description is active by default and marked fulfilled when the role closes. Fulfilled JDs disappear from ranking pickers but keep their full history: candidates, rankings, and decisions stay reviewable.
What is a hiring funnel in recruiting analytics?: The aggregated view of the decision log: how many candidates were uploaded, parsed, shortlisted, selected, or rejected, plus decisions per week, median time-to-decision, and time-to-fill. In CandidRanker it builds itself from the audit trail, no spreadsheet.
What is a Hiring Decision System?: MinMaxHR's category term for the evaluation layer that sits next to an ATS. The ATS manages workflow: postings, stages, records. The Hiring Decision System handles the decision step: ranking candidates against roles with explainable scores, and recording the human decisions that follow.
What does explainable ranking mean?: Every score comes with recruiter-visible reasoning: which criteria matched, which didn't, and the per-dimension evidence. No black-box numbers, which is what makes a ranking defensible to hiring managers, candidates, and auditors.

Related resources

How CandidRanker works. The five-step workflow these terms live in.
Recruiting inside an AI assistant. The MCP integration, end to end.
Explainable candidate ranking. The ranking workflow in operational detail.
Recruitment audit trail. The governance record, in depth.