Composing wikis answers "query everything together". This page answers the other half: when two wikis both describe the same person, how many nodes does the merged graph hold, and how do you make it one.
The rule that drives everything: a page's RDF subject comes from the root wiki's graph.context.wiki, not from the base of the repository the page came from. An installed source's pages are subjects under the host wiki's base, with their triples kept in that source's named graph.
The walkthrough below was run end to end against two throwaway wikis. Query output is abbreviated: IRIs are shown as CURIEs and columns trimmed.
Set up
A host wiki (https://me.example/) with a curated page, and a generated source wiki (https://acm.example/) with a page for the same person under a different filename.
# host/wiki.yml
wiki:
input: [wiki]
graph:
context:
wiki: https://me.example/
schema: https://schema.org/
sources:
- name: acm
type: git
url: file:///path/to/acm-memory # or a GitHub URL
ref: main
path: wiki
# host/wiki/Ethan_Davidson.md
---
'@type': schema:Person
schema:name: Ethan Davidson
schema:birthDate: 1999-01-01
---
# acm-memory/wiki/Ethan_T_Davidson.md
---
'@type': schema:Person
schema:name: Ethan Davidson
schema:jobTitle: Officer
---
wiki install
wiki graph list
name kind uri commit required_by
root root https://me.example/graphs/root
acm source https://me.example/graphs/source/acm a447973c60d1 root
Same basename: one subject, and a failing check
If the source also calls its page Ethan_Davidson.md, both pages resolve to the same IRI, https://me.example/Ethan_Davidson. There are not two nodes to reconcile: one subject holds both pages' triples, each in its own named graph.
wiki query --no-inference 'SELECT ?g ?p ?o WHERE {
GRAPH ?g { <https://me.example/Ethan_Davidson> ?p ?o } }'
https://me.example/graphs/root schema:name Ethan Davidson
https://me.example/graphs/root schema:birthDate 1999-01-01
https://me.example/graphs/source/acm schema:name Ethan Davidson
https://me.example/graphs/source/acm schema:jobTitle Officer
Source pages are built like root pages, so the same basename is also an output collision, and wiki check fails:
Errors:
- Output collision on output path '/wiki/Ethan_Davidson/': page …/wiki/Ethan_Davidson.md conflicts with page …/.wiki/sources/acm/repo/wiki/Ethan_Davidson.md.
- Output collision on public URL '/wiki/Ethan_Davidson/': …
Resolve it one of two ways:
- Rename the source page upstream (see contributing changes to a source), then reconcile with
owl:sameAs, below. - Exclude it in the host's config:
wiki.exclude: ['.wiki/sources/acm/repo/wiki/Ethan_Davidson.md'].wiki check --strictthen passes, and the source's page is dropped from the graph too, so its triples are gone.
Different basenames: two subjects
With Ethan_Davidson.md in the host and Ethan_T_Davidson.md in the source there are two subjects, https://me.example/Ethan_Davidson and https://me.example/Ethan_T_Davidson, and wiki check --strict passes. Nothing links them yet; every query returns the person twice.
Reconcile with owl:sameAs
Point the curated page at the generated one:
# host/wiki/Ethan_Davidson.md
---
'@type': schema:Person
schema:name: Ethan Davidson
schema:birthDate: 1999-01-01
owl:sameAs: https://me.example/Ethan_T_Davidson
---
With inference on (the default), the bundled OWL 2 RL rules treat the two subjects as one: each now answers with the union of both pages' properties.
wiki query 'SELECT ?p ?o WHERE { <https://me.example/Ethan_Davidson> ?p ?o }'
rdf:type schema:Person
schema:name Ethan Davidson
schema:birthDate 1999-01-01
owl:sameAs https://me.example/Ethan_T_Davidson
schema:jobTitle Officer
The reverse subject, Ethan_T_Davidson, answers with birthDate too.
schema:sameAs is not enough
schema:sameAs is an ordinary property: it is recorded, and nothing merges. A query over a composed corpus still sees two nodes, and wiki query with schema:sameAs returns only the triples each page states itself.
If you must keep schema:sameAs (it is what search engines read), either write both, or consolidate in the query without relying on inference:
wiki query --no-inference 'PREFIX schema: <https://schema.org/>
SELECT ?p ?o WHERE {
{ <https://me.example/Ethan_Davidson> ?p ?o }
UNION { <https://me.example/Ethan_Davidson> schema:sameAs ?other . ?other ?p ?o }
}'
The @id alternative
A page can set its own subject with @id instead of taking one from its filename. Giving the generated page an external identifier and pointing owl:sameAs at it also merges:
# acm-memory/wiki/Ethan_T_Davidson.md
---
'@id': https://acm.example/#Ethan_T_Davidson
'@type': schema:Person
schema:name: Ethan Davidson
schema:jobTitle: Officer
---
The tradeoff is rendering. wiki query --format markdown links only subjects under the wiki's base IRI to their page; a subject with an external @id prints as a bare IRI:
| s | o |
| [Ethan_Davidson](Ethan_Davidson.md) | https://acm.example/#Ethan_T_Davidson |
| https://acm.example/#Ethan_T_Davidson | [Ethan_Davidson](Ethan_Davidson.md) |
Prefer filename-derived subjects plus owl:sameAs unless the identifier must be stable outside the wiki, such as a Wikidata entity.
Recommendations
- Give the generated record its own filename (an identifier-keyed name works well) so it never collides with a curated page.
- Reconcile curated and generated records with
owl:sameAson the curated page. - Leave
@idoverrides for identifiers that outlive the wiki. - Run
wiki check --strictafter composing; a same-basename overlap is an error, not a silent merge.