Finding similar documents, and comparing them

The Documents tab groups files whose text is largely the same — a draft and its revision, a report saved twice with a paragraph changed, the same spreadsheet exported by two people. It then lets you read the two side by side before you decide.

Nothing is ticked for you. See Reviewing similar results.

What can be read, and what cannot

Files that can contain macros — .docm, .xlsm, .pptm — are read exactly like their ordinary counterparts. Only the writing is compared; the macros themselves are ignored.

Old Word and PowerPoint files are excluded outright. They are not read, not matched, and not reported — a .doc simply never appears on this tab. Re-save it as .docx if you need it included. Old Excel workbooks are not in that group: .xls is read normally. Excel's binary .xlsb format is, though — save it as .xlsx or .xlsm to have it compared.

Web pages and settings files are left out for a different reason: their tags and field names count as words. Two unrelated pages built from one template, or two settings files that differ only in a handful of values, would come out looking almost identical. Rather than report matches you would have to dismiss one by one, Trawler does not compare these at all. They are still found by the ordinary duplicate scan, so identical copies are still detected.

.rtf files are not compared here. They are still found by the ordinary duplicate scan, so identical copies are still detected — but they are not read for similar text. The reason is that a Rich Text file stores its pictures as text, and there is no reliable way yet to tell the pictures from the writing; on a real document that made two clearly different versions look identical. A wrong match is worse than no match, so this is deliberate. Save as .docx or .odt to have a document compared.

Files over 100 MiB are skipped. PDFs are exempt from that limit and are read whatever their size.

Ebooks with copy protection

An .epub you bought from a shop that applies copy protection cannot be read. Trawler does not guess at it: the book is skipped and recorded in Scan report → Scan errors under the type Documents.

This is deliberate, and it is the safer of the two behaviours. A protected book's text is scrambled rather than absent, so reading it anyway would produce groups of unrelated titles that look like genuine matches — and that verdict would then stick. Books without copy protection are read normally, including ones that only scramble their embedded fonts.

As with every tab under Similar files, only files your scan already found are considered. If the scan filtered out document types, or never covered the folder, there is nothing here to match.

Scanned PDFs

A PDF that is only page images — a scan, a photographed contract — has no text to read. Trawler does not guess: the file is skipped and recorded in Scan report → Scan errors under the type Documents, with the reason "no extractable text (scanned or image-only pdf)".

It is not retried on the next match. It is tried again once the file itself changes, or after a fresh scan.

Running a match

Open Similar files, choose the Documents tab, and use Find similar documents.

The Matching bar carries one control: a Similarity slider from 30% to 100%, set to 70% by default. It runs lower than the other tabs on purpose — two revisions of the same report often share far less text than you would expect. Hover the slider to read the current setting; while you drag, the figure follows the handle.

Moving the slider normally re-groups what is already there. When a change means the documents have to be read again, a bar appears reading "Settings changed since the last document match." with Re-run and Revert.

Results list as "N documents · size reclaimable" per group, with Words and Match columns per file. On the free version only the first 10 groups are shown, with a note saying how many more there are.

Match is read against the file at the top of the group — the one Trawler suggests keeping — so that file always reads 100% and every other figure tells you how much of it that file contains. A file whose text is identical to the top one also reads 100%. Where a group was built through a chain of resemblances, a file near the bottom can read lower than the slider you set: it matched something in the group rather than the file being kept, which is worth a look before you discard it.

The side-by-side compare view

On a group header, choose Compare side-by-side. The compare surface fills the panel: two panes of text, differences highlighted, headed Compare · Side-by-side text differences.

The left pane is A and the right is B. By default A is the newer of the first two files — the one you would probably keep. Back returns to the list.

The two panes are line-aligned: where one side has a line the other does not, the other side gets a blank row, so line 40 always faces line 40.

The controls

F3 jumps to the next difference and Shift+F3 to the previous one, wrapping at the ends. That works in the text view only — the Original page view has no per-line markers to jump between.

Auto-advance moves you along; it never decides anything. Advancing is navigation, not a verdict.

The comparison details

The ⓘ toggle shows or hides a details table beside the panes: Name, Folder, Size, Modified, Words and Pages. Values that differ between the two are shown in bold. Pages is a PDF figure; other formats show a dash.

Every row here is a fact about one file, so each one can differ between the two columns. The match percentage is not in the table, because it describes the pair rather than either file — you read it once, in the results list, against the file the group is keeping.

If you open the compare view on files that have not been matched yet, the panes read "Run the match to compare these files". That is a normal state, not an error.

Related

• Reviewing similar results — Keep and Discard, and what happens next
• Similar videos — and the FFmpeg setup video matching needs
• What can be undone, and what cannot — before you remove anything