The step after a batch has run, and the one that proves the model's central
decision worked: a certificate and the work it reviews have to be in the
*same* graph of the split query service, or a query cannot join them. The
search index says whether an item exists; only SPARQL says whether it is
reachable from the graph its certificate lives in.
Usage
verify_wikidata_export(
dir = "../register",
records = NULL,
update_register = FALSE
)
Arguments
- dir
the register repository to read from
- records
already-read records, as from [read_register_records()]
- update_register
whether to write the items found back into
`register.csv`, see [update_register_wikidata()]. Off by default: this is
a check, and editing the register is a separate decision.
Value
a `data.frame` with one row per certificate, invisibly: its
certificate ID, its report DOI, its item if it has one, whether the query
service can see it, and the work it states `review of` on
Details
Run it after the query service has caught up - hours rather than minutes,
and the scholarly endpoint is the slower of the two. An item that exists but
is not yet visible here is a lag, not a failure; the same result days later
is a failure.
Examples
if (FALSE) { # \dontrun{
verify_wikidata_export("../register")
} # }