Sources capture their own bibliography — authors, publisher, date

Why Care?

A source isn't just a URL — it's a citation. To ground anything in it you need the author, the publisher, and the date. Until now the curator caught the title and nothing else, so every source needed those typed in by hand. That's exactly the tedious, error-prone work the tool exists to remove.

Now: fetch a source and its author(s), publisher, and publication date appear on their own, pulled from the page's structured metadata. You still own them — every field is editable — but you start from filled-in, not blank.

What's New?

• Auto-filled bibliography. source.add and source.fetch extract author, publisher (og:site_name), and published date from Jina and write them to both the markdown frontmatter and the SurrealDB registry.

• Authors is an array. One author → a one-element list; multiple authors → multiple entries. (Jina hands back a string for one and an array for several; we normalize.)

• Editable everything. Title, Filename, Author(s), Publisher, Published date — all in the source form, all hand-correctable, each writing through to file + registry.

• Green save-pulse. Hit Enter (or click away) in a field and its border pulses green, then fades — a "saved ✓" right where you typed. No pulse if the save errored.

• Renamed files stay connected. Fixed a gap where an in-session source could lose its link to its file on disk.

Under the Hood — the one-line bug that hid everything

The bibliographic data was always in Jina's response — we were just asking for the wrong format. r.jina.ai returns a markdown preamble (title + date only) or a JSON body whose data.metadata carries author and og:site_name. We switched to JSON, wrote all the extraction code... and it did nothing. The author and publisher kept coming back empty.

The cause: a stale Accept: 'text/markdown' header. An earlier edit replaced the response-handling code but the window started one line below the header, so the request still asked for markdown. Every fetch ran the JSON parser against a markdown body, JSON.parse threw, and the code fell to the partial fallback path. All the right code, dead behind a failing parse. One character — application/json — and authors, publisher, and date all lit up:

authors:
  - "Alana Semuels"
publisher: "The Atlantic"
published_date: "2016-06-02"

The lesson logged for next time: when a feature is "implemented but inert," check the request, not just the handler — and verify the running container actually has the line you think it does.

The "renamed files won't connect" fix

A source added during a session came back from the resolver without its source_slug, so its Filename field sat empty and rename had nothing to point at — even though the database knew the file perfectly well. Two fixes: the source.add / fetch / retry responses now echo source_slug, and the UI derives the slug from corpus_path as a fallback. A file that exists never shows a blank filename again.

What's Next

• A paid Jina key (JINA_API_KEY in augment-it/.env) for more consistent extraction and higher rate limits — the free tier honors the JSON request less reliably.

• The still-pending interstitial detector and auto text-extraction from attached PDFs from the prior entry, [[2026-06-29_01_Strategy-Curator-Source-Surface-Curate-Sources-By-Domain-Type]].

Files Touched

content-ingest's jina.ts (JSON extraction) and corpus.ts (bib frontmatter + authors list), the resolver's domains.ts (registry columns + response echo), and the strategy-curator UI (curation.svelte.ts, SourceDetail.svelte, types.ts, app.css). Spec bumped to 0.0.0.6 with an Increment 4 section.