Scholarly Record Reconciler

Reconcile DOI, PMID, or bounded bibliographic queries across structured Crossref, NCBI PubMed, and OpenAlex metadata while preserving provider disagreements.

What this does not do

• No full-text retrieval.
• No citation-count ranking or quality score.
• No medical or legal interpretation.
• No automatic correction of provider records.
• No title-only mass search beyond the strict candidate cap.
• No claim that a canonical candidate is legal or scholarly ground truth.

Sources and authorization

Crossref REST and NCBI E-utilities are public structured metadata services. OpenAlex currently requires an API key; provide it through the secret input field. A missing or invalid key becomes visible partial coverage and is never exposed. There is no HTML scraping or hidden fallback.

Input

Each of 1-100 records contains exactly one doi, pmid, or query. Queries require title and may include firstAuthor, year, and venue. Select at least two sources. Example:

{"records":[{"doi":"10.1038/s41586-021-03819-2"}],"sources":["crossref","ncbi","openalex"],"includeRaw":false,"maxCandidatesPerSource":3,"concurrency":1}

Output

Each record produces provider terminal states, normalized provider records, field comparisons, explicit conflicts, a canonical candidate only when a unique deterministic rule succeeds, and the rationale. Ambiguous candidates are retained without canonicalization. Abstract text is never copied to the dataset; only availability is recorded. Optional raw JSON is bounded and written to the run KVS with a checksum manifest.

Pricing

reconciled-record costs $0.01 per terminal reconciled record. A result is charged only when at least two selected providers reach a terminal state and deliver reconciliation, ambiguity, or definitive absence evidence. Invalid input or fewer than two terminal providers is uncharged. Platform usage is additional.

State

The Actor is stateless. Canonical hashes exclude timestamps, run IDs, retries, and platform metadata.

Partial and negative results

Authentication failures, exhausted rate limits, timeouts, and malformed provider responses remain visible. Two terminal providers can still deliver a charged partial reconciliation; fewer than two cannot. Definitive cross-source absence for an exact identifier is useful terminal evidence and is charged.

Limits

Maximum 100 records, 10 candidates per provider, and concurrency 6. Launch resources are 256-1024 MB (512 MB default) with a 300-second timeout.

Privacy and deletion

The OpenAlex key is secret and never echoed. Crossref contact email is reduced to a provided/not-provided flag. Raw provider records are stored only when requested in the run KVS. Delete the run dataset and KVS to remove outputs.

Source rights and attribution

Provider metadata remains attributed by source and evidence locator. NCBI's disclaimer and copyright notice apply; copyrighted abstracts are not retained. See SOURCE_POLICY.md.

Limitations

Exact identifiers and exact normalized title plus author/year evidence can support a candidate. Fuzzy title similarity never establishes identity. Provider metadata can be incomplete, delayed, or conflicting.

Support and changelog

Email support@flintglade.com with the run ID, build ID, and stable error code, but never send API keys or private raw records. See CHANGELOG.md.