The WHO house voices and the WHO glossary: three editorial profiles,
each read rule by rule out of one publication in this instance's
library, with every rule carrying the quote and the page it came
from. It is a subgraph of who-iris:
| part | where | graph typology |
|---|---|---|
| the voices | skills/voices/ | voices, declared from within by skills/skills.json |
| the glossary | glossary/ | glossary |
| this page | the who-iris-docs subgraph | docs |
| the rules for working on a voice | style-guide-agents.md | docs (not kept inside skills/, where any non-README .md is read as a skill) |
| voice | read from | rules |
|---|---|---|
who-editorial | WHO Editorial Style Manual (who-pub-tps-931, 1993) | spelling, capitalization, eponym form, reference layout, non-discriminatory language |
who-guideline-development | Handbook for guideline development, 2nd ed (9789241548960-eng, 2014) | how a recommendation is stated, graded and attributed |
who-publication-design | WPRO publication style guide (wpr-rdo-2020-003-eng, 2020) | logo integrity, exclusion zone, brand colour, typeface, photo provenance |
A voice is derived from a publication; it is not the publication. So the voices sit beside the library and not inside it, and no source text is copied into them.
Each source is cited in the bare { libraryId, sectionId } form, which resolves
against the citing instance. The cross-instance form
{ instance, libraryId, sectionId } is supported for a voice whose source lives
in another instance.
A voice names its sources once, in sources[], and each rule says which
source it uses and where in that source, so a source's details are never
written twice.
bun run cat check:voices
bun run cat check:glossary
check:voices finds every instance with a voices/ directory and resolves each
citation through the declaration of the instance that holds the source. It
reports four kinds of failure separately, because each needs a different fix:
unknown instance, no library graph, never ingested, and no such section.