# Local document export experiment — protocol v1

Registered before running the exports. Question: for original text/shape documents, how do selectable vector content, raster resolution and JPEG quality change file size and decoded pixels? This is a first-hand file-export experiment, not a LinkedIn delivery, human reading, accessibility-conformance or engagement study.

Inputs: four original four-page square documents, 540 PDF points per side. Separate content classes: a sparse explanatory brief, a dense numerical table, a process diagram and a deliberately detailed synthetic pattern. All examples and numbers are constructed; no account, customer, third-party artwork or private data is used.

Seven export conditions for each document: vector original; PNG at 1080, 720 and 540 pixels per side; JPEG at 1080 pixels/quality 90 and 70; JPEG at 720 pixels/quality 70. PNG resolution contrasts hold encoding constant; JPEG-quality contrasts hold 1080 resolution constant. Cross-format contrasts change several representation properties and must not be interpreted as isolating one cause. JPEG uses 4:4:4 chroma sampling. No post-result deletion or preferred-format substitution.

Measure all 28 PDFs: bytes, page count, per-page text drawing-operation presence and embedded-image count, SHA-256, and RGB pixel RMSE at a fixed 360px raster render compared with the same page of its vector original. Average page RMSE within each four-page document, also retain every page result. Zero RMSE does not prove accessible content; a larger error is not a measured readability failure. No hypothesis test or population confidence interval is appropriate for these four constructed documents.

Use installed pdf-lib 1.17.1, sharp 0.35.5 and existing Poppler pdftoppm; record actual versions and rendering/encoding settings. Fixed metadata dates, deterministic source definitions, runnable generator, dataset, complete 28-file bundle and representative original/variant views are published. View decoded pages to catch export mistakes; report any procedure correction and retain its reason. Inspect the actual public artifacts after release. Do not run a paid service, real account publication or user experiment for this test.

Interpretation should help a reader decide which export to inspect and why retaining the editable source matters. It may establish results only for the measured files and pipeline. It cannot establish an optimal LinkedIn size, effects of its recompression, assistive-technology behavior, human preference or reach.