Help

What is Diffesq?

Diffesq compares two files and shows you exactly what changed — text, JSON, YAML, XML, CSV, and more. All comparison happens inside your browser; your files are never uploaded to any server.

How do I compare two files?

Drop or choose a file on each side of the app. Diffesq detects the format automatically (you can override it), runs the comparison locally, and renders the differences. Use the Prev/Next controls to step through changes.

How do I export a report?

Click Export and choose a format (HTML, Markdown, JSON, or PDF). Free users get one watermarked sample; Pro unlocks clean, unwatermarked exports and lets you export multiple formats without reopening the menu.

Comparing Word (.docx) documents

Diffesq compares two Word documents structurally — paragraphs, headings, lists, tables, formatting, headers/footers, notes, comments, fields, and images — and shows the differences two ways: a synchronized side-by-side view of the rendered pages and a consolidated redline. A change panel, structure map, and summary help you navigate and understand what changed. As with every format, the two files are compared entirely in your browser and never leave your device.

Which Word features are supported?

Rendering runs in one of two modes, and Diffesq always tells you which one is in use:

• High fidelity — paragraphs and runs, standard character/paragraph formatting, headings, numbered and bulleted lists, tables (including merged cells), page geometry (size, orientation, margins, columns), manual page and section breaks, headers/footers, page-number fields, common fonts with fallback, and inline images.
• Approximate — when a document uses features the high-fidelity renderer can’t fully reproduce (e.g. content-overflow pagination, floating/anchored objects, or legacy drawing markup), that pane switches to an approximate paginated view with a clear label and a list of the affected features.
• Inventoried but not rendered — SmartArt, charts, equations, content controls, and embedded media are recorded and disclosed, never silently dropped.

The comparison itself is independent of rendering: the structural results are computed from the document model and stay correct even when a pane falls back to approximate mode.

What about tracked changes?

If either document already contains tracked revisions, Diffesq compares each document’s original, pre-revision state and shows a notice saying so. Existing tracked insertions/deletions are not presented as newly detected differences. (Final-state and revision-aware comparison are planned for a future release.)

Exporting a DOCX comparison

Pro users can export a DOCX comparison as an HTML report, PDF, JSON manifest, or CSV summary. Each report records the file names, date, engine and schema versions, the active preset and applied ignore rules, the rendering fidelity, warnings, the summary, and the change details — plus a statement that the files were compared locally. Free users don’t get DOCX exports; reports generate on your device and are never produced server-side.

Comparing datasets (Parquet, SQLite, JSON Lines, Arrow, CSV)

Dataset comparisons show three layers: a schema diff (columns and — for SQLite — tables, indexes, and views), a profile diff (row counts, null rates, distinct counts, min/max, mean and standard deviation, side by side with changed metrics flagged), and a row-by-row diff. You can also compare across formats — a CSV against a Parquet file, say — with columns matched by name and compared on a common type.

How row matching (key selection) works

Rows are matched on a key column — a column that is unique and non-null in both files, like order_id. Diffesq suggests one automatically (SQLite primary keys rank first) and you can pick any shared column, combine two into a composite key, or switch to positional matching — which is labelled unreliable because it misreports changes whenever row order differs. With a good key, a reordered file with identical data correctly shows zero changes.

Numeric tolerance and comparison options

Floating-point pipelines often produce values that differ only in the last decimal places. The comparison options let you set an absolute or relative tolerance (for example 1e-9) so that noise below the threshold doesn't count as a change; the default is exact. You can also ignore case, trim whitespace, treat NULL as an empty string, and exclude columns from comparison while keeping them visible.

Why very large datasets show fewer rows

Everything runs in your browser's memory, so row-by-row comparison has a ceiling: up to about 2 million cells (roughly 500,000 rows of a 4-column table) every row is compared — typically in a few seconds. Beyond that, the first 100,000 rows are compared and the result says so. Beyond about 20 million cells the row diff steps aside entirely. In every case the schema and profile layers still cover the complete dataset — column statistics are computed over all rows even when the row grid is sampled — so "did anything meaningful change?" always has a full answer.

Accessibility

Every DOCX view and control is keyboard-operable and screen-reader friendly: each change is announced with its type, location, and content; progress, cancellation, and errors are announced via live regions; every status carries a non-color indicator (icon, symbol, or count) so nothing depends on color alone; and animations respect your “reduce motion” setting.

Known limitations (this release)

• Documents compare, never change — there’s no editing, merging, or accept/reject of changes, and no modified-DOCX or redlined-DOCX export (reports are HTML/PDF/JSON/CSV).
• Password-protected and macro-enabled (.docm) documents aren’t supported.
• Word documents: target size is up to 25 MB / ~500 pages per file.
• Folder comparison matches files by path and content; it doesn’t follow renames across folders or compare file histories.
• The image duplicate finder scans one folder at a time, and cleanup (moving or deleting duplicates) needs a browser with write access to folders — currently Chrome and Edge.
• Tracked-changes documents compare in their original, pre-revision state (final-state comparison is planned).

Where is my license key?

After purchase, your license key (format DIF-XXXX-XXXX-XXXX-XXXX) is shown on the confirmation screen. A license activates on any number of devices. Lost your key? Email [email protected] from your purchase email address.

Is my data private?

Yes. Your files are processed entirely on your device and never leave it. The only thing that can leave your device is your saved comparison profiles and filter selections, and only if you turn on preference sync — it is off by default, available on Pro, and you can view or delete anything stored at any time. See our Privacy Policy for details.

Still stuck?

Visit Support or email [email protected].