DBLP reliability contract

DBLP reliability contract

Resmon uses two documented DBLP routes for two distinct questions:

• General keyword searches use https://dblp.org/search/publ/api. The caller’s query remains the REST q value and pagination uses f and h.
• Person searches use https://sparql.dblp.org/sparql. The query resolves an exact escaped rdfs:label and selects publications linked with dblp:authoredBy. It does not treat editor-only works as authored works and does not switch to SPARQL because REST returned a challenge.

Both routes share one 0.5 request/second limiter, a truthful Resmon User-Agent, one cooperative 45-second operation budget, and at most one retry per request. A valid Retry-After wait is used only when it fits inside that budget. HTML challenge pages are failures and are neither retried nor bypassed.

Author meaning and pagination

Exact labels can identify several DBLP people, while aliases can be missed. Resmon therefore retains DBLP person IDs on returned authors and does not claim that a name uniquely identifies one human.

The inner SPARQL page contains distinct publication/title/year tuples and omits the matching person from its projection. This prevents two same-label people linked to one publication from consuming two page slots. The outer query joins all authors. Resmon groups those rows by the existing DBLP record key, removes duplicate author rows, and advances the page from the count of distinct raw publication IRIs rather than the number of normalized survivors.

Identity, fields, and dates

Publication and author identifiers must be URI bindings. Titles, author names, and optional venues must be literal bindings. Years must be xsd:gYear literals. An optional DOI must be an RDF URI under https://doi.org/, matching the public SPARQL service’s observed JSON binding shape and DBLP’s KG tutorial. Malformed rows are recorded as parse failures; valid records from the same page remain available.

The existing DBLP key such as conf/... is the stable external_id. The full https://dblp.org/rec/... IRI is the URL. DOI and venue remain optional and no missing value is manufactured.

DBLP provides year precision. YYYY-01-01 is a storage encoding, not evidence of a known month or day. A partial boundary year is excluded. If a requested window contains no complete calendar year, Resmon makes no DBLP request and records that the window cannot be answered at DBLP’s precision.

Evidence limit

A bounded diagnostic on 2026-09-17 observed HTML challenge pages from both REST search endpoints and useful JSON bindings from the public SPARQL service. A second one-request diagnostic using the candidate projection observed ten DOI URI bindings, ten DBLP publication and author URI bindings, literal venue/name/ title fields, and ten xsd:gYear values. Those observations support the adapter choice and parser contract. They do not establish the cause of the REST challenge, guarantee future hosted availability, or prove every name and publication shape.