Products
Data Framer

Release notes

What each version of Data Framer deploys. The version in the page footer links to its entry here, so what is running and what it changed are one click apart.

1.19.0

2026-09-18ReleaseThis build

Fixed

  • Legacy Word (.doc) and Excel (.xls) files are refused at upload, and the message names the accepted formats. They used to be accepted with no parser behind them, so the file scanned clean, was stored, and sat at "Safe" for ever while extraction failed silently.
  • A document that cannot be read is marked "Could not be read" with a reason written for the person who uploaded it: a damaged or password-protected file, or a scanned PDF with no text layer, which now says so and suggests uploading it as an image. Such a document is not retried, since it would not read better on the third attempt.
  • A failure in the model provider or in storage is still retried; if the last attempt fails too, the document is marked with a reason rather than left looking fine.
  • Failed documents appear in the document list with their reason, and the step tracker counts them, so nothing that was uploaded can disappear from view.
  • Downloading a source document whose name contains characters outside Latin-1 (Polish letters, for example) returned an error. The file name is now sent in the RFC 5987 form with an ASCII fallback.

Changed

  • Removing a document is its own state, shown as "Removed", rather than sharing the "Failed" state. Documents that were in the old shared state before this release are treated as removed, so none resurfaces without a reason to show.
  • The upload form lists the accepted formats: PDF, Word (.docx), Excel (.xlsx), CSV, text or image.
  • A stored document that failed extraction can still be downloaded, since the first question is what was uploaded.

Added

  • This release notes page at /releases, linked from the version badge in the footer. Every version bump must carry an entry here; the build refuses one that does not.

1.18.2

2026-09-18Hotfix

Fixed

  • The template picker on New product described every family as a "Tokyo demo template". Each now reads "Product data template for ..." with the family and standard it covers. The template backfill carries the wording to templates that were seeded before this fix.

1.18.1

2026-09-18Hotfix

Changed

  • The Node runtime is pinned to 22.x for every environment, and the malware-scan image used by the test suite is pinned to a fixed version. Nothing visible changed; both had been matching by coincidence rather than by configuration.

1.18.0

2026-09-17Release

Changed

  • The product page follows the business process from top to bottom: where the product is in its five steps, the step that belongs on this page, the result of that step, how good the data is, and the destructive actions last. An approved version no longer shows an upload control the server would refuse.
  • The dashboard's product cards gained a search box, a status filter and a next-action line, and each product family carries its own glyph.
  • The page header carries the product family; the page heading carries the product name. They used to both carry the product name.

Versions before 1.18.0 predate these notes; their content is in the repository's commit history.