Scholarly Correction Monitor

Watches a bounded DOI watchlist for Crossref-deposited corrections, retractions, expressions of concern, withdrawals, and other updates, and reports what changed since the prior run. Built for research teams, librarians, systematic-review maintainers, and evidence-product builders.

This is an unofficial metadata monitoring tool. It is not affiliated with or endorsed by Crossref or Retraction Watch.

Full behavioral contract: PRODUCT-SPEC.md.

Why two Crossref requests per DOI

Crossref's own documented data model places the update-to field on the notice record (the correction/retraction/etc.), where it points backward at the DOI it updates -- not on the original work. A direct lookup of an original DOI therefore frequently exposes no update-to at all, even when deposited updates exist:

• Direct lookup of 10.1177/1758835920922055 (GET /v1/works/10.1177%2F1758835920922055) returns no update-to.
• Crossref's update-discovery query, GET /v1/works?filter=updates:10.1177/1758835920922055, returns notice 10.1177/17588359231172420, which carries the retraction record.

This Actor fetches both endpoints for every watched DOI: the direct lookup for bounded bibliographic metadata, and the update-discovery query as the sole, authoritative source of currentStatus and deposited updates. no-deposited-update in this Actor's output means both sources were checked and neither exposed a recognized update -- never a direct-lookup-only guess. Even a direct 404 still runs bounded update discovery; this prevents a missing direct record from hiding a notice that explicitly targets that DOI. If a watched DOI is itself an update notice, that edge case is flagged with a warning rather than misreported as the notice's own status (see PRODUCT-SPEC.md).

Input

Output

One dataset item per watched DOI: bibliographic facts (title, publisher, type, issued date, container title), currentStatus, normalized deposited updates, changeType versus the saved baseline, content/status/metadata hashes, warnings, retry count, and an error code where applicable. Full field list: PRODUCT-SPEC.md#output-contract and .actor/dataset_schema.json.

A run summary and the current watchlist's baseline snapshot are stored in the run's default key-value store. Authoritative state uses one hashed key-value record per DOI, so overlapping runs on different DOI subsets cannot overwrite one another. No abstracts, author emails, or raw Crossref payloads are ever stored -- see SECURITY.md.

Pricing

Pay per event: the synthetic apify-actor-start event (Apify default pricing) plus doi-result at $0.003 per DOI that was successfully fetched or explicitly resolved as not-found. Source errors after retries are emitted as free diagnostics, and DOIs skipped once the run's spending limit was reached are never fetched or charged. Paid capacity is reserved before source work. See PRODUCT-SPEC.md#billing and LIMITS.md.

Local development

Requires Node.js 22+.

npm install
npm run lint
npm run typecheck
npm test              # unit + integration tests (mocked HTTP, no network)
npm run fixture-run    # deterministic offline run through the real request/parsing pipeline
npm run smoke           # bounded LIVE Crossref smoke -- real network, run manually only
npm start                # runs the Actor locally (needs local Apify storage emulation)

npm run smoke is not part of npm test and is never run in CI; it makes a small number of real, polite requests to the live Crossref API using the two DOIs documented in PRODUCT-SPEC.md.

More docs

• SECURITY.md -- permissions, data handling, privacy.
• LIMITS.md -- technical and product limits, including the no-deposited-update limitation.
• CHANGELOG.md -- version history.

Rollback

If a published build misbehaves, revert traffic to the last known-good build on the Apify platform (Actor -> Builds -> select the previous tagged build -> "Publish this build" / repoint the latest tag), which requires no code change. To roll back the source itself,

git revert

Support

Support: support@flintglade.com