SVRIS verifies a citation in five fixed steps: extract the metadata fields, normalize the strings, search authoritative sources for candidate documents, label every field, then apply deterministic rules to classify the result. Nothing in the chain is probabilistic, so the same citation always produces the same verdict — and every verdict shows which fields matched and which sources supplied the evidence.
SVRIS (SourceVerify Reference Identity Standard) verifies citations through a pre-determined, step-by-step process. Unlike AI systems that provide opaque verdicts, SVRIS makes every decision transparent and auditable.
Step 1: Extract Metadata Fields
The first step is parsing the citation to extract its component fields:
Title — The name of the work
Authors — List of author names
Year — Publication year
Venue — Journal, conference, publisher, or organization
Identifier — DOI, URL, ISBN, or other unique ID
Step 2: Normalize Strings
Before any comparison, all strings are normalized:
This ensures that minor formatting differences don't cause false negatives.
Step 3: Search for Candidate Documents
SVRIS queries multiple authoritative sources to find documents that might match the citation:
OpenAlex
Google Scholar
Google Search
Direct URL verification
Publisher databases
Step 4: Assign Field Labels
For each candidate document, SVRIS compares every field and assigns exactly one label:
MATCH — Fields are identical after normalization
CONTAINS — One is a substring of the other, or the citation uses a valid abbreviation (e.g., "PNAS" for "Proceedings of the National Academy of Sciences")
ABSENT — The citation doesn't include this field
UNCONFIRMED — Citation claims a value but evidence neither confirms nor contradicts it
CONTRADICTION — Evidence shows a different value (e.g., year off by 2+ years, no overlapping authors)
Step 5: Determine Classification
The final classification follows strict rules based on the field labels. Key requirements:
VERIFIED
Title or DOI must MATCH
No CONTRADICTION
Strong corroboration: 3+ MATCHes OR 2 MATCHes + 2 CONTAINS
VERIFIED WITH ERRORS
Document found with minor issues
Examples: weak corroboration, single contradiction with strong support, title CONTAINS with a MATCH
NEEDS HUMAN REVIEW
Ambiguous evidence requiring expert judgment
Examples: title MATCH alone, borderline with contradiction
UNVERIFIED
No matching document found
Or: multiple contradictions, no title/DOI match
Why This Matters
Every SVRIS classification can be audited. If the system says "VERIFIED," you can see exactly which fields matched and which sources provided evidence. This transparency is critical for academic integrity, publishing, and any workflow involving AI-generated citations.