Skip to content

ADR-0023 — Picks wait in one list per project; a governed dataset grows only by a version you make (0.6.0)

Status: Accepted (owner, 2026-11-01), with RFC-023; built for 0.6.0. Date: 2026-11-01 Detail: RFC-023, PRD-014 (FR8 to FR10), UXS-002 (the sheet)

Context

After many evaluations of production data, the owner makes the next version of a key dataset from a few production records that would most help. A good record is often noticed before its target dataset is known, across many captures and many evaluations, and nothing kept it until the version was made.

Decision

  • A pick marks a production item as a candidate for one or more governed datasets, or none yet, with tags and a note. Picks wait in the project's one list, data/picks.jsonl, in git; the list is append-only, the latest line for an item wins, and a dropped or used pick keeps its history.
  • A governed dataset grows only by a version you make on purpose: dataset-extend V1 --dataset-id V2 with the picks you choose, the lineage recorded per item. Its references are then promoted as for any dataset; nothing is promoted by the extension.
  • The pages never write (PRD-004, ADR-0004): the sheet and the dataset page select and give the pick and dataset-extend commands.

Consequences

  • Candidates accumulate across captures and evaluations without deciding their target early.
  • A version records where every added item came from, so its origin is never lost.
  • Picking in the browser means copying a command; labelling or picking by clicking is a later question.

Alternatives considered

  • One pick list per governed dataset. Rejected by the owner: a good item is often seen before its target.
  • Building a version, or picking, from the pages. Rejected: the pages never write.