Every source TAVONEL reads, and what survives the read.

One list decides what the upload route accepts, what the file picker offers, what a rejection says and what this page prints. They cannot disagree, because they are the same list. Each row states its support tier and what survives into the Compiled World; anything not listed here is refused at upload rather than accepted and quietly mishandled.

Every accepted format is read the same way. It is sanitized to PDF and read by the OCR reader, so every row below preserves the page, the paragraph text and the exact region that binds a result back to its source. What that path does not do, it does not do for any of them: No native structure reader yet · No table or formula extraction · No visual reconciliation of the native file · Up to 5 MB per source · Up to 80 pages per source.

Accepted
11of 12 listed formats are accepted at upload.
Preserved
3fields cross into the compile request, the same three for every accepted format.
Converted
10of the accepted formats are turned into another format before they are read.
Omitted
1listed here and turned away at upload, with nothing about them compiled.
This table is generated from the same list the upload route enforces.
SourceSupport tierWhat is preservedSpecific to this format
.pdfapplication/pdfBest effort
  • Page
  • Paragraph text
  • Exact region
  • Read as sanitized PDF, then OCR
.docxapplication/vnd.openxmlformats-officedocument.wordprocessingml.documentBest effort
  • Page
  • Paragraph text
  • Exact region
  • Converted to PDF before reading
.odtapplication/vnd.oasis.opendocument.textBest effort
  • Page
  • Paragraph text
  • Exact region
  • Converted to PDF before reading
.jpg .jpegimage/jpegBest effort
  • Page
  • Paragraph text
  • Exact region
  • Converted to PDF before reading
.pngimage/pngBest effort
  • Page
  • Paragraph text
  • Exact region
  • Converted to PDF before reading
.tif .tiffimage/tiffBest effort
  • Page
  • Paragraph text
  • Exact region
  • Converted to PDF before reading
.gifimage/gifBest effort
  • Page
  • Paragraph text
  • Exact region
  • Converted to PDF before reading
.xlsxapplication/vnd.openxmlformats-officedocument.spreadsheetml.sheetBest effort
  • Page
  • Paragraph text
  • Exact region
  • Converted to PDF before reading
  • Spreadsheets have no defined page unit
.odsapplication/vnd.oasis.opendocument.spreadsheetBest effort
  • Page
  • Paragraph text
  • Exact region
  • Converted to PDF before reading
  • Spreadsheets have no defined page unit
.pptxapplication/vnd.openxmlformats-officedocument.presentationml.presentationBest effort
  • Page
  • Paragraph text
  • Exact region
  • Converted to PDF before reading
.odpapplication/vnd.oasis.opendocument.presentationBest effort
  • Page
  • Paragraph text
  • Exact region
  • Converted to PDF before reading
.zipapplication/zipNot readnothing beyond the shared read
  • Expanded in your browser before upload
  • Each member is checked against this table
  • Encrypted or nested archives are refused
  • Up to 128 files and 500 MB expanded
  • The archive itself is never compiled
  • Not included in a free evaluation
What each support tier means
Verified, native reader
Read by a native reader for the format, with a qualification receipt behind it. No format has reached this tier yet.
Verified, native and checked
Read natively and cross-checked against a render or OCR pass, with a qualification receipt behind it. No format has reached this tier yet.
Best effort
Extracted by a general-purpose path. Useful, and not a guarantee that every structure in the source survived. 11 formats here today.
Metadata only
Handled at the type, metadata or container level only. No content is read. No format has reached this tier yet.
Needs review
Encrypted, damaged or proprietary in a way that needs a person before anything is compiled. No format has reached this tier yet.
Not read
Refused. Nothing about the source is compiled. One format here today.
How a row is filled in

What is preserved is what the compile request carries today, not what the file format contains. A spreadsheet’s cells and formulas are in the file and are not in that list, so the row says so rather than implying otherwise.

Known limitations are read out of the code that enforces them, not written as caveats. An archive is expanded in the browser and its members are validated one at a time; the archive itself is never compiled, which is why it appears below the tier that refuses it and still appears in the file picker.

A tier is earned by a measurement. A format moves up when a native reader for it exists and a qualification run produces a receipt with the date it was produced. That is a measurement, not a decision, and this page changes when the measurement does.

Legacy binary HWP (application/x-hwp, .hwp) is not in the table. It has no reader here, and an unlisted format cannot be held for review: the upload route refuses it before any file is stored. Listing it under review would describe a queue that does not exist, so it is refused with everything else that is absent, and it appears here instead.

A format is not a connector. Every row above is about reading one file that has already arrived. Supported, for a connected system, means the whole lifecycle: create, update, rename, move, delete, permission change, tombstone behaviour, source-version propagation, world update and ACL enforcement. The current readers checkpoint provider changes, suspend a bound source when deletion or lost access is observed, and fail closed before governed use. What each connector does at each of those events is on Integrations.