Resume PDF Text Extraction: November 2026 Renderer Results

Published 8 November 2026 · Nodesify product team

What does this collection show?

Three synthetic inputs were exported through five application templates in an isolated Linux candidate using the production PDF code path, including its hosted fonts and render-document page. The 15 cases below retain the PDF, raster pages, input snapshot and text extracted with PDF.js 6.3.289. These are local string checks, not employer ATS acceptance rates, hiring outcomes or independent career review.

This candidate includes working changes and is not a claim about a deployed release. The July engine comparison is retained separately. Different inputs, extractors and software versions mean the two collections do not establish an improvement or regression between releases. WeasyPrint and DOCX were not measured in this collection.

Download the inputs and results

• Complete collection ZIP
• Per-case JSON results and CSV results
• Renderer runtime and image provenance
• Collection procedure and limitations and SHA-256 checksums
• Fixture source, collector source and scoring source

All names, employers, achievements and credentials in these inputs are fictional teaching material. For an example intended for inspection before purchase, use the separate PDF, Word and TXT samples.

How should I interpret the checks?

The collector reads each PDF.js text item and appends a space, or a newline when its end-of-line flag is set. It separates PDF pages with form-feed markers. This procedure can insert spaces inside words split into several glyph runs. A missing string in the report therefore does not by itself establish missing text in the PDF.

Selected unique words

Matches normalized unique words from selected input fields; this does not verify every character or its order.

Section heading strings

Finds seven heading strings, without establishing reading order or field assignment.

Contact strings and link labels

Checks email, normalized phone digits and visible link labels, not hyperlink destinations or ATS fields.

Selected names, skills and dates

Matches selected normalized phrases or word-boundary tokens, without checking their semantic relationships.

Selected Unicode tokens

Checks nine tokens in the Unicode fixture, rather than all languages or characters.

Long-fixture marker and opening phrase

Checks the marked opening phrase near its marker; it does not establish complete bullet continuity across pages.

Normalization collapses whitespace and page-break characters. A match cannot establish visual placement, correct associations or field extraction by a commercial ATS. Keep partial scores and missing-token details visible; a percentage is not a pass certificate.

A worked observation: readable words can fail a string check

In the standard Classic extraction, the collector writes Certi fi ed rather than Certified. In the Unicode Elegant extraction, it writes Certi fi cations, so the heading-string check matches six of seven headings. The raster page visibly contains the heading. This illustrates a boundary between visual presentation and this extraction procedure, rather than evidence of an employer rejecting a resume. Inspect the Unicode Elegant page, its extracted text and the scoring implementation together. The review notes state which images were inspected.

Inspect each exported resume

Which environment produced these files?

Captured at 2026-11-07T18:12:11.080Z. The collector used v22.23.4 and PDF.js 6.3.289. The runtime download identifies the separate Linux renderer environment.

Candidate image ID: sha256:2b4672bc542748c42fc4824ddf95db295dace7e13a1e493079509ae8f1a84460. The image identifies the captured build; the base source revision alone does not identify its working changes.

What should I check in my own resume?

Inspect the actual file you will submit. Compare its words, dates, links, reading order and page breaks with your intended job version, then inspect populated application fields. Follow the exported-text checklistand read the methodology for the difference between extraction, content scoring and employer screening.