FIELD GUIDE / FILE CONVERSION
Choose a conversion that preserves what you actually need.
Changing a filename extension does not change a file’s structure. A real conversion reads the source format, builds a controlled representation, and writes a new format—with deliberate tradeoffs in layout, semantics, image quality, and metadata.
- 19
- bounded conversion routes
- 30 sec
- longest worker timeout
- 16 MiB
- maximum generated output
01 / BEFORE YOU CONVERT
Start with the destination, not the source.
The best target format depends on what will happen next. If a person needs to read a document, PDF is often useful because page boundaries are stable. If software needs the data, JSON, CSV, or YAML is usually more appropriate. If someone needs to edit prose, Markdown or plain text is easier to inspect than a binary office file. Conversions that optimize for one goal can discard information needed for another.
Preserve meaning
Use Markdown-to-HTML for headings, lists, emphasis, and links. Use Word-to-HTML when paragraph and table structure matters more than exact page layout.
Preserve values
Use spreadsheet-to-CSV or JSON when the cell values matter. Styling, formulas, merged cells, charts, comments, and extra worksheets are not equivalent to data rows.
Preserve appearance
Use PDF when pagination is important, or PNG for lossless image output. A text extraction route cannot reproduce the visual layout of a scan.
Reduce size
WebP or JPEG can be smaller than PNG for photographs. JPEG does not support transparency; Switchboard flattens transparent areas to white when producing JPEG.
02 / WHAT EACH FAMILY DOES
Know what is preserved—and what is intentionally left behind.
| Family | Useful for | Important boundary |
|---|---|---|
| Markdown / HTML / text | Readable documents, publishing, content cleanup | Raw HTML in Markdown is escaped. Script and style elements are removed from HTML cleanup routes. |
| Word / PDF | Extracting readable text or simple web structure | DOCX output focuses on paragraphs and tables. PDF extraction requires a real text layer; image-only scans need OCR, which this service does not perform. |
| JSON / YAML | Configuration and structured records | YAML is parsed in safe mode. CSV output requires records that can form a table; arbitrary nested data may not map cleanly. |
| CSV / Excel | Tabular analysis and handoff between tools | Excel-to-CSV reads the first worksheet and exports values. CSV-to-Excel prefixes formula-like cells so imported text is not treated as a spreadsheet formula. |
| Images | Web delivery and compatibility | PNG is lossless; JPEG and WebP use quality settings. Animated images are treated as still-image inputs rather than preserved animations. |
The converter checks binary signatures and container structure for formats such as PDF, DOCX, XLSX, and images. A misleading extension is not enough to make unrelated bytes valid. Text routes require UTF-8 input, which avoids silent character-set guesses that can corrupt names, punctuation, or non-English text.
03 / PRIVACY MODEL
Send only the data the conversion needs.
Switchboard processes an upload for the current request and does not intentionally retain the source or converted output. The web framework may briefly spool a larger multipart upload to private temporary storage while the request is active. Conversion responses are delivered as downloads with Cache-Control: no-store. Operational logging records a request reference, the chosen conversion, input byte count, and outcome—not the file body.
Those controls reduce data exposure, but they do not make a public conversion service the right destination for every document. Remove information that is not necessary for the task. For regulated records, trade secrets, private keys, passwords, medical records, or documents governed by a contractual processing requirement, use an approved organizational workflow instead.
- Make a working copy.Redact unrelated names, identifiers, account numbers, comments, and hidden sheets before uploading.
- Choose the narrowest route.Extract plain text if formatting is unnecessary; avoid moving extra content through a richer format.
- Inspect the download locally.Confirm that headings, rows, characters, formulas, and transparency behaved as expected.
- Handle the output deliberately.A converted file can still contain sensitive content. Store and share it under the same data-handling rules as the source.
04 / LIMITS AND FAILURE MODES
Bounded work makes a public converter more predictable.
Each route has its own input ceiling and execution timeout. Lightweight text routes accept smaller inputs; the largest document, spreadsheet, and image routes accept up to 10 MiB. The public edge caps the complete multipart request at 12 MB, generated output is capped at 16 MiB, and the longest conversion timeout is 30 seconds. DOCX and XLSX inputs are ZIP containers, so their member count, expanded size, and compression ratios are checked before processing.
A rejection does not necessarily mean the file is malicious or damaged. It may be empty, too large for the selected route, encoded with something other than UTF-8, mislabeled, password-protected, structurally incomplete, or too complex to finish within the public worker limit. Try exporting a smaller copy from the source application, selecting only the required worksheet or pages, saving text as UTF-8, or using a more direct target format.
05 / REVIEW CHECKLIST
Validate the output before it enters another workflow.
- Compare row counts and column headers after spreadsheet or structured-data conversion.
- Check dates, decimal separators, leading zeroes, large identifiers, and non-English characters.
- Confirm that links point where their visible labels suggest before publishing converted HTML or Markdown.
- Open office and PDF downloads in a fully patched viewer and keep endpoint protection enabled.
- Review image dimensions, transparency, color, and file size rather than judging only the thumbnail.
- Do not assume conversion validates the truth, provenance, accessibility, or legal status of the content.
Switchboard currently has no advertising, analytics, or third-party tracking tags. If advertising is added in the future, it will be considered only for an explicit allowlist of public guides such as this one—never the upload or download workflow—and only after the privacy disclosure, consent controls, and route-specific Content Security Policy are ready.