Operator story

Nosh, the citation gardener.

Nosh is a research librarian: Perl, Lisp, Zotero, Catmandu-shaped metadata flows, PURL servers, and enough bibliography pain to want help. Nosh does not know Go and should never need to know Sidecat internals. This page follows one real metadata job through the whole product — it is the first full path Sidecat is built and judged against, and the working material behind the design of Sidebird and Sidebar.

Scholarship grows through citation trails, not junk drawers

Nosh’s durable belief is simple: a source needs a card — why it matters, where it belongs, what claim it supports. Agents are welcome to help, but mutation of a trusted catalog needs clear intent, a preview, an approval, and a receipt. And a path that worked once should become replayable and reusable, not a script in a folder with a warning in the filename.

One job, start to finish

1. The mess. A retiring web host is taking a PURL namespace with it. About three thousand persistent-URL rules need to move, the bibliography group’s Zotero library is full of stable links that run through them, and the exported MARC records embed the old hostnames. Nosh has fixed things like this before — with one-off Perl scripts that worked once and were never trustworthy again. A folder of scripts nobody can rerun.
2. A preview, not a script. Nosh writes the first check as a small workbench routine: read the caller-supplied PURL rules, the Zotero item context, and a Catmandu-shaped record preview, and report which links would break under the proposed rewrite. It touches nothing. Because it ran in the workbench, the shape is replayable — an agent can rerun it tomorrow against a new export and get a comparable answer. A replayable routine with a recorded result.
3. From routine to instrument. The preview proves useful, so it becomes a package candidate. Admission is where a private experiment turns into a shared instrument: the package states its purpose, what it reads, what it writes (nothing, yet), and the invariants it promises. Nosh can read that statement without reading a line of implementation. A package admission packet a librarian can judge.
4. Review what happened. Wearing the review hat, Nosh opens the run evidence in Sidebird: which rules were flagged, which Zotero collections were touched by the preview, what the receipts say actually ran. Something looks off in the MARC preview — so Nosh flags a question anchored to the exact page, and the answer comes back on that same page. Discussion stays attached to the thing being discussed. An anchored question, answered where it was asked.
5. Promote what needs deciding. The clarified concern becomes a promotion: a packet that carries the anchors, the notes, and Nosh’s reason into planning. Promotion is a handoff, not an approval — nothing is scheduled, nothing runs. It just means the question has graduated from “help me understand this” to “someone should decide something.” A Sidebar packet with its review context intact.
6. Approve what happens next. Wearing the approval hat, Nosh reads the packet in Sidebar next to the program’s north star, its charter, and the capability map — is this rewrite read-only, does it write Zotero, does it publish resolver state? Nosh approves the bounded next step in their own words, and the decision is recorded with its obligations. A recorded decision with a reason and obligations.
7. Run under the decision. Only now does the admitted path run — with the scope the decision named and evidence flowing back to Sidebird, where the loop closes: Nosh can see what happened, ask the next question, and the story of why the work exists is still attached to it. Receipts, and a trail the next librarian can follow.

The two hats

The same person did the reviewing and the approving in that story, but the product never collapses the two jobs. In Sidebird, Nosh looks backward — evidence, replay, receipts, questions about what happened. In Sidebar, Nosh looks forward — plans, packets, boundaries, decisions about what should happen next. Keeping the hats separate is what keeps a review surface from becoming a planning cockpit, and an approval surface from becoming a replay dashboard.

What this story demands of the product

Nosh is design material, not marketing. Each claim above is a demand a Sidebird or Sidebar decision can be checked against: