Policies (policies in config)

The config schema includes optional policies so sync, quality, and generate --resume can respect copy and exclusion rules without ad-hoc flags.

Who applies what (vs exclude)

knob	Applied by	Effect
exclude (scan config)	Scanner / validate / cleanup pipelines	Limits which source files participate in discovery.
policies.preserve	sync, merges, cleanup, and resume flows that respect preserve	Keys/prefixes that stay in targets, hydrate from source, and/or skip machine translation.
policies.parity	quality, review, generate --resume heuristics	Excludes expected matches from parity-style checks.

policies.preserve

Field	Sync	Cleanup	generate --resume (when respectPreserve)
copyKeys	Exact paths stay in the sync template; hydrate from source; retain on prune	Not removed as unused	Verbatim — skip MT (copy source)
copyPrefixes	Same for any path under the prefix (e.g. navigation.)	Same	Verbatim — skip MT
uncertainPrefixes	Declared roots without scan anchors — hydrate + retain like auto uncertain prefixes	Protected when uncertain-key policy is protect / warn_only	Shape — allowed to translate (not treated as verbatim preserve)

Trailing dots on prefixes (e.g. navigation.) are fine; sync normalizes them for keep matching.

Split semantics (no rename)

• Verbatim: copyKeys / copyPrefixes — never machine-translate; sync keeps/hydrates from source.
• Shape: uncertainPrefixes (and auto-detected uncertain prefixes from the scan) — keep trees in locale JSON through sync; generate --resume may still fill them when policy allows.

Prefer declaring uncertainPrefixes (or copyPrefixes when copy must stay source-language) instead of scan-only anchor files once you are on ≥ 1.0.3.

policies.parity

• excludeKeys
• excludePrefixes
• excludeValues

Why init --rich shows this block

i18nprune init --rich includes a policies section so teams see where preserve/parity rules belong. You can keep it empty until needed.