Inspiration

A bookmark remembers a page, not the paragraph, table or code example you needed. When a useful reading collection moves between people, context can disappear—and an archive can look complete even when its sources are missing. Final Edition helps the person preparing a collection and the person reading it see what was kept and what was lost.

What it does

Final Edition turns an owner-selected Linkding export and saved HTML captures into one searchable reading edition. The owner explicitly pairs captures, reviews omissions and chooses a reading-only or complete copy. Unresolved, ambiguous and unsupported records remain visible rather than being silently guessed away.

Passage Finder searches remembered words and opens the matching retained paragraph, table or code example. Previous/Next moves between matching passages. “Save excerpt + source” downloads the passage with source URL, original-byte digest and disclosed omissions. Supported table excerpts retain context through a clearly labeled generated cell grid; layouts that cannot be handled are refused rather than flattened misleadingly.

The Python builder runs locally and exports static HTML. Recipients can read the packaged supported content without the builder or original bookmark server. This is not a crawler, hosted service or an AI answering system. Both profiles may contain private content: reading-only excludes fields, but is not redaction or permission to share.

How GitLab and the assisted agent fit

GitLab holds the MIT-licensed source, review merge requests, CI checks and downloadable generated editions. A read-only, mention-triggered GitLab Duo flow assists review: it examines evidence and can recommend a bounded repair while leaving application, tests and release decisions to people.

In a recorded historical test, Duo diagnosed a seeded reader-reset regression and proposed a patch and regression test. It did not apply or run them; the repair was independently applied and validated. A wrong-source-identity test produced an abstention. These are individual observed sessions, not a general success-rate benchmark, and the historical outputs retained limitations. They do not qualify the later table fixes. The deliberate manual release gate remains closed.

What is verified—and what is not

The latest review documentation head is ebcb0994 on review/windows-portability-20261024, in draft MR !3. The source-and-test snapshot is c24b5c2a, with final runtime change ff2e1620. Its local check passed 180 Python, 29 builder and 74 reader tests, including 26 table tests. Reader interaction checks use a limited Node DOM double; these are not new browser or live-model results.

The newer table fixes use JSON-quoted cell grids to prevent embedded delimiters, multiline captions and formatting whitespace from inventing apparent rows or columns. These checkpoints skipped CI. Browser/download qualification of their exact regenerated bytes remains incomplete.

The demo, actual hosted-browser observations and successful four automatic CI jobs belong to earlier source c6543d6, pipeline 2991053242, not the newer table fixes. Those bounded browser observations covered search, passage/table/code navigation, history, metadata-only and missing-source handling, excerpt downloads and a narrow viewport. Broader accessibility, security, capacity, cross-browser and distributed-file cold-open qualification remain open. No merge, deployment or release is claimed.

What we learned and what comes next

A useful preservation tool must make incomplete context visible, not just produce a polished file. Generated excerpts also need a truthful boundary between source text and transformed table data. Next is browser verification of the exact newer artifacts and independently reviewed release evidence—not expanding the claims beyond what was tested.

Source and evidence

Built With

Share this project:

Updates

Submission history