fix(pubs): carry open-access metadata through the discovery review queue

[JamesKane]

A paper promoted from the OpenAlex discovery queue lands on /references with no Open access badge and no citation count, until the nightly by-DOI enrichment job happens to re-fetch it. In dev the un-badged rows are exactly the recently-promoted ones.

Root cause

The search response already carries open_access.oa_status and cited_by_count — the wire Work type even parsed them — but Candidate dropped both, pubs.publication_candidate had nowhere to put them, and promote_candidate's INSERT omitted them. So a promoted publication starts with open_access_status = NULL, which the list template maps to "no badge".

Now carried end to end, from both the discovery job and the public "suggest a paper" form. When promotion reuses an existing publication it gap-fills only — a row the curators already own is never overwritten.

Two related defects on the same path

Publications with an OpenAlex id but no DOI were unreachable by enrichment. The work-list is WHERE doi IS NOT NULL, but a candidate is promoted on its OpenAlex id alone, so those rows keep NULL citations/OA forever. Adds work_by_id and a second enrichment pass. Confirmed against live OpenAlex on the dev row that hit this (W7111817982: no DOI, oa_status: green).

DOIs from discovery were stored as resolver URLs. OpenAlex returns https://doi.org/10.x and it was stored verbatim, so the reference list rendered href="https://doi.org/https://doi.org/10.x" (broken link), and exists_by_doi couldn't match the bare DOI the public form normalizes — meaning a paper already in the catalog could be queued a second time. Ingest now normalizes; normalize_doi moves to du-external and is shared with the submit form.

Migration 0073

Adds cited_by_count / open_access_status to pubs.publication_candidate and backfills URL-form DOIs to bare form in both tables. publication.doi is UNIQUE, so the publication update skips any row whose bare form is already taken by another publication — a pre-existing duplicate pair survives un-normalized instead of aborting the migration. Dev and cutover both had zero such collisions; worth re-checking on prod after deploy.

Applied to dev: 30 publications and 5,934 candidates normalized, 0 URL-form rows left.

Notes

• upsert_candidate now takes a NewCandidate struct — the positional list was at seven arguments and this would have made nine.
• Existing prod/dev rows that are already un-enriched still need one publication-update run to backfill (requires OPENALEX_MAILTO); new promotions carry the badge at accept time.

Test

cargo test --workspace against a fresh empty database: 54 binaries, 239 tests, 0 failures. cargo clippy --workspace -- -D warnings clean. New coverage: candidate keeps OA/citations through promotion (live-DB test), DOI normalization, and the search-payload parse.

🤖 Generated with Claude Code