Why did your PDF-to-Excel conversion come out wrong?

How converted tables actually fail

Some failures announce themselves: shifted rows, stray cells, a table that clearly lost its spine. The dangerous failure is quiet. An empty quarter gets padded with zeros, or a value bleeds across columns, and the sheet looks finished. Whoever totals it later inherits an invented number with no mark on it.

Before the diagnosis steps, one status note: you can open these pages and sign in to the workspace today, but that does not mean external platform accounts are connected, and it does not mean the product has formally launched. Everything below runs on the PDFs you supply.

What you need before you diagnose

The source PDF open beside the suspect sheet, and the specific row or section that triggered suspicion. A vague sense that something is off diagnoses slowly; one concrete mismatch, like a total that will not reconcile, diagnoses fast.

Know which pages of the report came from a scanner. Reports assembled from mixed sources fail differently on each kind of page, and knowing the provenance cuts the search in half.

Step 1: Test whether the page has a text layer

Try selecting text in the PDF itself. If nothing selects, the page is an image and no converter can read it directly. We refuse those rather than return a table we invented. Recognising the text first is a separate job you would run elsewhere; expecting table structure out of pixels is the fastest way to waste an afternoon.

Step 2: Compare row shapes against the source

Walk the affected section row by row. A short row usually means a real empty cell survived honestly; a suspiciously uniform five-column sheet where the source shows blanks means something got padded. Count columns on both sides before changing anything.

Step 3: Rebuild merged period headers by hand

When Q1, Q2, Q3 spanned three columns in the PDF but landed as one merged label, split them back out manually. Do this once per header style, then reuse the correction for every report in the batch that shares the template.

Step 4: Treat padded zeros as suspects

Any zero you cannot find printed in the source gets deleted, not explained. Zeros propagate into sums and averages invisibly, and they are the single most common way a converted sheet lies while looking tidy.

Step 5: Record the cause for the next batch

Write down which failure mode it was: scan, merged header, or padded gap. Next month's reports from the same source will fail the same way, and a one-line note turns a diagnosis into a ten-second check.

Verification

The repair holds when the previously broken section reads correctly against the source and a spot check elsewhere in the sheet finds no new surprises. If fixing one table forced edits in three others, stop and reconsider which failure mode you actually had.

Limits worth stating plainly

Tables are not cell-perfect. Check every number against the source page before you rely on it. A whole multi-page report converted in one pass is not reliable today. Rows from different periods can land in one cell and a column can be dropped. Work one table at a time. Audit-style statements with dollar signs, em-dash blanks, and parenthesised negatives are not handled. Data rows can be lost. Scanned PDFs with no text layer are refused rather than guessed at. There is no OCR path you can rely on here today. A document with no printed table is not something to convert. If you are handed a sheet built out of running prose, discard it and tell us.

Tablefold keeps uncertain cells flagged rather than silently resolved. When a repaired sheet ever disagrees with the source PDF, the source wins, and the flag history tells you where to look first.

What Tablefold does in this workflow

Tablefold detects the grid structure, declines to invent values for empty spans, and marks multirow headers and low-confidence cells instead of resolving them quietly. The flags are the diagnostic input for everything above.

The repair decisions stay yours: what to fill, what to leave empty, and when a page needs the scan detour. The product makes the failures visible early, which is what keeps them cheap to fix.