PDF, TXT, DOCX and Markdown export

New export bundles require all four formats. They share a derived document (resume_document.py) built from the final cited Markdown. Supported source syntax is H1–H4 headings, paragraphs, ordinary bullets, inline emphasis and recorded HTTP(S)/mailto links. Unsupported tables, images, raw HTML, code and other structures fail explicitly instead of silently dropping content. PDF, TXT and DOCX display recorded URLs with their labels. Markdown preserves labels and destinations using standard link syntax; a rendered Markdown reader may display only the label. Section order, chronology and wording remain consistent. Exporting never rewrites facts or edits the career pack.

The resume wizard runs these commands for you. For manual operation, replace the example draft, plan and output paths with your reviewed application's paths:

python3 scripts/export_resume.py outputs/example-draft.md --plan data/plans/example-v2-plan.json --output outputs/example-v2-export
python3 scripts/export_resume.py --check outputs/example-v2-export/review/export-report.json

The output directory must be new. files/ contains only resume.pdf, resume.txt, resume.docx and resume.md. review/ holds the clean HTML preview, internal derived document with claim references, and export report. Send only the required files from files/, never the review directory. The report pins the draft, plan and actual submission bytes. Re-export to a new directory after editing. A standalone legacy draft may omit --plan and supply --paper-size A4|Letter and --page-limit N; this compatibility path does not confer planned-resume approval.

• PDF: real selectable text, correct recovered order, actual page count and compliance with the recorded limit. Source URLs are embedded as clickable hyperlinks, including each fragment when a link wraps across lines. Chrome/Chromium renders the document using an isolated temporary profile, with browser headers/footers disabled. No network assets are loaded. Visual inspection of final page breaks is still required.
• Markdown: clean UTF-8 resume.md with headings, paragraphs, bullets, emphasis and recorded links. Internal citations and review metadata are excluded; the cited working draft remains unchanged. The checker compares canonical Markdown structure, formatting, wording and link targets with the shared document.
• TXT: UTF-8, ordinary heading text, blank-line separation and - list markers.
• DOCX: editable WordprocessingML with native paragraph styles, heading outline levels and true list numbering. No layout tables, floating boxes, hidden review data or macros. Word may paginate differently from the PDF renderer; inspect DOCX in the intended word processor when a DOCX page limit is consequential.

Markdown, TXT and DOCX need only standard-library Python. PDF requires Chrome or Chromium and either Poppler (pdftotext, pdfinfo and pdftohtml) or macOS Swift/PDFKit for text, page-count and hyperlink checks. export_resume.py reports missing prerequisites or tool failures per format and exits nonzero. It retains successful formats and a failure report but does not label a partial bundle complete. Do not claim accessibility or employer ATS certification from text recovery. Report those checks separately.

The automated check compares recovered text against the shared document, allowing Unicode compatibility normalization, whitespace wrapping and list markers. For PDF only, a physical newline after an existing hyphen or inside an expected URL can be reconciled with the exact source token. Ordinary within-word spaces, missing word boundaries and changed URL characters still fail. The report records when this bounded reconciliation was needed. TXT and DOCX remain strict. Markdown is compared against its canonical structure, inline formatting, wording and link destinations. Word boundaries are preserved: SQL pipelines and SQLpipelines are different. Missing, added or reordered content fails. Native DOCX content and styles are also covered by regression tests. The export report records PDF tools, page count, content checks, hashes, and unperformed visual/accessibility/Word layout checks.

An orphan evidence comment is an authored Markdown error. Attach citations to the claim line or an indented continuation line. Keep accurate hyphenated words and recorded URLs when troubleshooting extraction; do not rewrite the candidate to accommodate an exporter defect.

Clickable PDF links

The shared document retains source URL targets and their visible text spans. Explicit Markdown links and full bare HTTP(S)/mailto URLs become HTML anchors before PDF generation. Balanced parentheses, query strings, fragments and encoded characters stay attached to the original destination. Use [label](URL) to disambiguate URLs ending in sentence punctuation. Bare hostnames do not acquire an invented protocol. TXT and DOCX retain the same visible URL text; this change verifies PDF hyperlinks, not word-processor automatic link detection.

After rendering, the exporter extracts PDF hyperlinks and their clickable text. Every logical source link must match its destination and all of its text, allowing multiple annotations for a wrapped link. Missing, incorrect, unexpected or partly clickable links fail the PDF export even if text extraction passes. Repeated destinations are checked as separate occurrences. Normal browser serialization of hostnames, Unicode and percent-encoding is allowed; changed paths, query values and fragments are not. The Poppler branch uses pdftohtml XML without its -p option, which would rewrite PDF link destinations; see the upstream utility manual.

Version-2 export reports retain annotation targets/text, verification results and file hashes. --check compares that recorded inspection with the pinned Markdown and checks the exported file hashes. It does not rerun the PDF extractor. Older reports cannot claim this verification; create a new export bundle when needed. Existing files are never regenerated automatically.

These checks do not visit destinations or establish website availability. A correct clickable URL can still lead to a removed page or require authentication. Copying wrapped URL text may retain a line break in some viewers; clicking the embedded hyperlink uses the complete destination independently of that text.

Run the real regression with Chrome and either supported extraction toolchain:

CAREER_TEST_REAL_PDF=1 python3 tests/test_pdf_links.py

This uses fictional data in temporary directories. The normal test suite runs the deterministic link and failure-path tests without launching a browser.

For multiple positions at one employer, employment grouping uses compact title/date paragraphs and H4 subsections in HTML, PDF and DOCX.

Editorial quality and page review

Follow resume-quality.md for role-sensitive selection, explicit editorial questions, review of omitted eligible evidence, PDF geometry diagnostics and the visual review required for the exact exported bundle. These extend existing process records; they never promote application judgments into career facts.

Markdown and report compatibility

Version-4 reports require PDF, TXT, DOCX and Markdown. Version-2 and version-3 reports retain their original three-format checks; version-1 reports still require a fresh export for verified PDF hyperlinks. Existing bundles are never rewritten. New briefs request four formats; saved three-format briefs remain readable and produce four formats when exported again. Markdown does not carry PDF pagination.

Markdown normalizes whitespace around emphasis delimiters without changing the visible wording. Emphasis that cannot be represented faithfully in supported Markdown fails explicitly instead of exporting extra literal asterisks.

Requirements by task

Generating a resume from career evidence uses Claude; rendering an existing draft uses local scripts. A PDF prerequisite or verification failure retains successful formats but leaves the four-format bundle incomplete. Read the report's per-format error, correct the cause and export into a new directory. Do not bypass a failed check or describe a partial bundle as fully verified.

For the complete private record rather than a resume, see your files and exports.

Known PDF verification limitation

In local testing on 2026-09-22, the optional real-PDF regression failed because Poppler recovered neighboring words or punctuation as part of a hyperlink's clickable text. The same failure occurred with the previous exporter. This does not establish that every PDF is affected, but installing the prerequisites alone does not guarantee a verified bundle.

If this happens, retain the export report and generated PDF for diagnosis. The successful Markdown, TXT and DOCX files remain available, but the bundle stays incomplete. Do not rewrite factual content or remove evidence merely to satisfy the extractor. Test any toolchain or rendering fix against the failing example before claiming the PDF is verified.

Read this guide on GitHub ↗