PDF to EDGAR HTML: why OCR is only the first step

A PDF can be a useful source for conversion, but extracting its text is not the same as preparing a reliable EDGAR document. Text order, financial tables and footnote relationships all need review. Whether a PDF itself is permitted in a submission depends on the applicable form and document requirements; it is not a universal substitute for required HTML or structured data. EDGAR format overview.

Identify the kind of PDF you have

A digital PDF may contain selectable text. A scanned PDF may contain only page images or an OCR layer. Mixed documents can contain both. Start with representative pages, including the densest table and the smallest footnote text, before estimating effort.

If an editable approved source exists, ask whether it can be supplied instead. That may remove avoidable reconstruction work, but it does not remove the need to confirm that the editable source matches the approved PDF.

OCR can produce plausible-looking errors

OCR mistakes are not limited to obvious gibberish. A zero and the letter O can be confused; parentheses can disappear; a superscript can attach to the wrong value. A paragraph can read smoothly while a material figure is wrong. Treat OCR as an extraction aid, not as a proof of fidelity.

In an illustrative example, a scanned table shows a parenthesized amount in one year column and a dash in another. OCR returns the amount without parentheses and moves the dash onto the adjacent row. Spellcheck would not reliably identify either problem. The reviewer needs a side-by-side comparison with the approved source.

Reconstruct tables deliberately

A PDF's visible column alignment does not guarantee that the extracted text carries a usable table structure. Rebuild the association between row labels, period headings and values. When information is unclear, raise a query instead of inferring the intended value.

• Check each header against the columns below it.
• Preserve units, scaling labels and currency descriptions.
• Confirm that footnote markers remain with the correct line item.
• Inspect rows split across pages or interrupted by repeated headings.
• Reconcile the final converted version to the approved source.

Keep a query log

Record page or section references for anything that cannot be read with confidence. A good query identifies the uncertain element and asks the authorized source owner to confirm it. It should not bury a guess inside a polished output.

Keep the confirmed response attached to the revision record. If the source owner supplies an updated PDF, establish whether the whole document or only specific pages replace the earlier baseline. This avoids creating a hybrid package assembled from inconsistent versions.

Account for tagging and package checks

HTML conversion and Inline XBRL preparation are separate scope decisions. Reconstructing a table visually does not assign the correct tags, contexts or units to its data. Confirm which specialist owns structured-data work and how it will be validated. SEC Inline XBRL explanation.

Ask for a scoped estimate

A meaningful conversion estimate should consider source quality, page count, table complexity, graphics, tagging scope and proof cycles. A guaranteed turnaround based on page count alone may hide uncertainty in a scanned or complex document. Send nonpublic files only through an agreed secure channel, and establish who will authorize the final release.