check() in the VeriWorkly ATS Engine turns a resume into a report in a fixed series of steps. Knowing them explains why a field is empty, why a rule is missing from a report, and why the same resume can score differently as pasted text and as a PDF.

The pipeline

1. Prepare the input. The input becomes one text, and a validated document when it was structured. Size guards apply here: resume text is cut at 50,000 characters. Invisible characters are counted before they are removed.
2. Localise the policy. With locale packs attached (withLocales), the languages, the region and the date order are detected or taken from your languages and region options.
3. Read the resume. Text is normalised once (NFKC, digits of every script, invisible characters). Wrapped lines are rejoined and letter-spaced words are read back. Headings split the text into sections. The parser then recovers the record, and the text-level checks run: integrity, timeline, skills and writing.
4. Score the rules. Rules whose evidence exists are evaluated. Rules without evidence are dropped. See Scoring.
5. Match the posting, when you pass jobDescription. See Job matching.
6. Assemble the report. Fixes, strengths and categories are built, and the report is stamped with the engine version and policy fingerprint.
7. Add advice. File, age and target-ATS notes are read last, after the score is final, so they cannot move a number.

Accepted inputs

check accepts a string, a structured AtsResumeDocument (built with the /document helpers), or a JSON Resume (a basics object plus a work array, or an education entry with an institution). A structured input is read directly, with no parsing. A string is parsed. Any other object is flattened to text, with its keys on their own lines, and then parsed like a string.

import { check, DEFAULT_POLICY } from "@veriworkly/ats-engine";

const fromText = check(resumeText, DEFAULT_POLICY);
const fromJson = check(
  {
    basics: { name: "Mei Lin Chen" },
    work: [{ name: "Acme", position: "Engineer", startDate: "2021-03" }],
    skills: [{ name: "Go" }],
  },
  DEFAULT_POLICY,
);

See Structured input.

Reading order

Order is what breaks a multi-column layout, more than columns as such. Lines from a PDF are ordered by where they sit on the page and not by the order the file paints them, column by column. A running header, footer or page number printed on every page is read once or not at all. A Word document's header and footer are read once as well.

To see the text the way the engine read it, pass includeLines:

import { check, DEFAULT_POLICY } from "@veriworkly/ats-engine";

const report = check(resumeText, DEFAULT_POLICY, { includeLines: true });

for (const line of report.lines ?? []) console.log(line);

report.lines holds at most 500 lines and is the resume's full text, so it is off by default. Showing it to a candidate is the quickest way to explain a scrambled read.

What gets recovered

The parsed record, report.parsed, holds what an ATS would store: name, email, phone, links, roles (title, employer, start, end, current), education (school, credential, ISCED level), skills, certifications, spokenLanguages (with CEFR level), monthsOfExperience and highestIsced.

Certifications and spoken languages are read from their own sections, and from a "Languages:" or "Certifications:" line among the skills. A "Languages: Go, Rust" line stays skills.

Provenance

parsed.provenance says, for each group (name, email, phone, roles, education, skills, certifications, spokenLanguages), where it came from.

A report can then say which values a parser read and which a model read, which are different claims. See the AI guide.

Layout signals from files

Text alone cannot show columns, tables, photos or hidden text. extractResume and readResumeFile from /node return a layout alongside the text: column ratio, ruled table count, page count, images of photo size, hidden text, pages with no text layer, file metadata, and for a PDF whether it is encrypted or for a Word document its tracked changes and comments.

Hidden text means characters drawn so a reader cannot see them while an ATS still reads them: in the page's colour or near it, too small to read, off the page, invisible, or under an image.

What pasted text cannot be judged on

Pasted text and structured documents have no geometry. Five rules depend on it and are left out of the report when no layout is passed:

• ats-v2.format.columns
• ats-v2.format.tables
• ats-v2.format.photo
• ats-v2.parse.imageOnly
• ats-v2.integrity.hiddenText

A dropped rule is neither passed nor failed, and it leaves the score's denominator, so the score means the same thing with or without it. A pasted resume can still have a two-column PDF behind it, so a score from text says nothing about those five.