How the Ryzio ATS score is calculated

What the score means—and what it does not

The Ryzio ATS score is a deterministic screening-readiness diagnostic. It evaluates the text and layout signals Ryzio can extract from a resume, then applies a published set of checks. The same extracted resume and job-description input produce the same audit. AI does not choose the score.

It is not a score issued by Greenhouse, Lever, Workday, Taleo, iCIMS, SmartRecruiters, or an employer. It does not reproduce a company's configured screening questions, recruiter decisions, or ranking rules, and it cannot guarantee an interview.

Inputs and processing boundary

The public checker accepts a PDF or DOCX up to 5 MB. Ryzio extracts text and available layout signals, checks whether the file resembles a resume, converts the extracted content into a structured internal representation, and runs the audit. The original uploaded file is processed for that request and is not saved to a user dashboard.

A job description is optional. When at least 50 characters are supplied, Ryzio evaluates keyword overlap, repeated requirements, and role alignment. Without a job description, the two checks that require one are marked not scored and removed from the total, and the score is expressed as a percentage of the points that could be measured. They are not awarded partial credit, because an unanswerable check is not a passed check. A score produced without a job description is therefore not directly comparable to one produced with a job description.

Checks included in model phase4-40-30-20-10

The current model totals 100 points. Individual findings expose the maximum points, earned points, status, observation, and suggested fix. The four category totals are shown below.

• Keyword match · 40 points: job-description overlap, repeated critical requirements, and role-title alignment.
• Content · 30 points: action-led bullets, measurable outcomes, outcome language, vague filler phrases, bullet density, and skills formatting.
• Structure · 20 points: supported file type, extractable PDF text, column/layout signals, section presence, standard headings, dates, and complete experience entries.
• Contact and file-readiness · 10 points: name, email, phone, location, links, and basic format signals.

How job-description matching works

Ryzio builds a normalized keyword index from the resume and compares it with terms and phrases extracted from the job description. Alias groups reduce false misses for common equivalents. Repeated and contextually important requirements receive more attention than isolated generic words.

The comparison measures text evidence, not capability. A missing requirement may mean the resume uses different wording, the candidate omitted relevant experience, or the candidate does not have that skill. Ryzio should never turn a missing term into an invented claim.

Known limitations

The checker does not connect to a real employer account or observe how a specific ATS has been configured, and it does not submit the file to any vendor's live parser. Vendor results describe whether the file trips a documented parsing risk; a clear result means no known risk was found, not a guarantee of compatibility.

Layout inspection depends on the file type. PDFs give per-item positions, fonts, sizes, and page count, so column structure and font checks are measured directly. DOCX has no fixed pagination and no positional data, so column structure is inferred from the document's tables, page count is estimated from text length, and fonts are read from the document's style definitions. When a signal cannot be measured for a given file, the check is marked not scored rather than passed.

Keyword extraction and aliases are heuristic. A specialized term, unusual abbreviation, or multilingual resume may need human interpretation. The model rewards measurable outcomes and conventional structure, but those preferences should not override accuracy, accessibility, or the norms of a particular profession.

Versioning and change policy

The active model identifier is phase4-40-30-20-10. Ryzio should change weights or thresholds only when the implementation, tests, and this public page are updated together. A future model change should receive a new identifier so scores from materially different methods are not presented as directly comparable.

Last methodology review: 16 July 2026. The current public baseline was checked against the running code and the primary vendor sources linked in the supporting guides.