Each wiki process builds the wiki RDF graph once (unless you pass --reload) and reuses it for:

• every Wiki Subcommand query in that process
• every SPARQL block in Wiki Subcommand render
• wiki build --render when rendering before static output

OWL-RL expansion runs when inference is enabled (default for most commands; use --no-inference on query / render when debugging asserted triples only).

Cross-process reuse

• By default, a new shell still starts cold.
• Pass --cache on Wiki Subcommand query, Wiki Subcommand render, or wiki build --render to persist a warm graph under .wiki/cache/.
• The persisted graph is reused only when the wiki fingerprint still matches.
• --reload forces a fresh build and refreshes the current cache entry.

Long-lived workflows

Wiki Subcommand serve with --watch keeps one process alive. On file changes it rebuilds the graph, re-runs SPARQL rendering, and reloads the browser.

Tradeoffs

• In-process reuse is still the default and simplest model.
• --cache helps repeated one-shot commands across fresh shells.
• The disk cache is invalidated automatically on wiki or config changes.
• .wiki/cache/ is excluded from wiki fingerprinting so cache artifacts do not invalidate themselves.

CI tips

wiki render --check            # fail if any inline block is stale
wiki render --cache            # repeated one-shot render loop across shells
wiki build --render --cache    # repeated one-shot render + build loop across shells
wiki query "..."              # same graph as a prior render in one script if you chain in one shell

Related

• Wiki Subcommand query
• Wiki Subcommand render
• Wiki Subcommand serve — long-lived preview and optional SPARQL endpoint