Citation Link and Metadata Auditor

Audit explicit DOI, PMID, and public HTTPS citations for resolvability, canonical identifiers, redirect behavior, metadata consistency, content type, and source-reported correction or retraction signals.

What this does not do

• No paper-content retrieval.
• No citation-quality or truth score.
• No automatic manuscript changes or link repair.
• No medical interpretation.
• No crawling references from a page.
• No guarantee that provider metadata is complete.

Sources and authorization

The Actor uses public Crossref REST metadata, NCBI E-utilities metadata, and a bounded request to a user-supplied public HTTPS URL. URL checks start with HEAD and use GET Range: bytes=0-65535 only when HEAD is unsupported. Every redirect hop is revalidated against the SSRF policy. The Actor does not retain article or page bodies.

Input

Each of 1-1,000 entries contains exactly one doi, pmid, or url, and may include expectedTitle and expectedYear. Checks can be limited to identifiers, HTTP, metadata, or corrections. Concurrency is 1-16 and each response is capped at 2 MB.

{"citations":[{"doi":"10.1038/s41586-021-03819-2"}],"checks":["identifiers","metadata","corrections"],"followRedirects":true,"concurrency":1}

Output

One result row is written per citation, followed by one uncharged run summary. Result rows include canonical identifiers, redirect chain, HTTP evidence, bounded metadata, deterministic expected-field comparison, source coverage, source-reported warnings, evidence locators, and a recommended human action. KVS includes source-contract and artifact manifests.

{"status":"ok","data":{"substatus":"healthy","canonicalIdentifiers":{"doi":"10.1038/s41586-021-03819-2"},"sourceWarnings":[],"recommendedHumanAction":"No citation-specific action indicated by the selected checks."},"charge":{"event":"citation-audit","chargeable":true}}

Pricing

The primary event is citation-audit at $0.01 per terminal citation audit. Healthy, redirected, metadata-mismatch, definitive not-found, source-warning, and useful partial results are chargeable. Invalid input, blocked URLs, transient DNS/TLS failures, provider outages, and exhausted rate limits are not charged. The maximum event cost is the number of admitted terminal audits multiplied by $0.01; Apify platform usage is separate.

State

The Actor is stateless. It creates no baseline and commits no monitor state. Duplicate inputs are independently represented so billing and result cardinality remain explicit.

Negative and partial results

A verified 404/410 is a useful terminal result, not a run failure. A source warning is always labeled as provider-reported and never restated as a Flintglade determination. When one selected source succeeds and another fails transiently, the row is partial only if it still provides useful evidence.

Resources

Minimum 256 MB, default 512 MB, maximum 1024 MB, and a 300-second hosted timeout. Requests follow a five-hop redirect cap, 2 MB response cap, and maximum concurrency of 16.

Data handling, privacy, and deletion

Input and results are stored in the run's Apify storages under the caller's account retention settings. Optional Crossref mailto is treated as secret and is never echoed. URLs with credentials are rejected and query values are not used as display identifiers. Delete the run, dataset, and key-value store in Apify to delete stored output.

Rights and attribution

Crossref metadata is attributed to Crossref; PubMed metadata is attributed to NCBI. Explicit URL evidence is limited to technical metadata authorized by the user-supplied locator. Provider names and marks are not used to imply endorsement.

Limitations

Provider metadata can lag publishers or omit relations. A reachable URL can still contain incorrect content. Title similarity is reported separately from exact normalized equality and never auto-corrects a citation. Source outages and hostile destinations can make an audit partial or unchargeable.

Support: support@flintglade.com