Scenario — steady-state authoring

A localized install (Spanish active) evolves: a way is added or edited. Once an adopter is in localized mode, maintenance settles into a stable two-part shape: the English root + its one localization, kept together.

How it plays out

flowchart LR
    A[author / edit a way] --> E[English .md root<br/>source of truth]
    E --> L[regenerate its Spanish alias<br/>translate vs root · tune]
    L --> C[multi corpus rebuilt<br/>English anchor + Spanish layer]
    C --> G{ways tune locale clean?}
    G -->|no| L
    G -->|yes| D[done — both layers current]

    classDef authoring fill:#7c3aed,stroke:#4a5568,color:#ffffff
    classDef root fill:#2d8e5e,stroke:#4a5568,color:#ffffff
    classDef process fill:#2d7d9a,stroke:#4a5568,color:#ffffff
    classDef gate fill:#fbbf24,stroke:#4a5568,color:#1a1a1a
    classDef done fill:#2d8e5e,stroke:#4a5568,color:#ffffff

    class A authoring
    class E root
    class L,C process
    class G gate
    class D done

What each move is doing

• The English root is authored normally. Writing or editing a way means writing English description+vocabulary — exactly as on an English install. The 17× all-languages stub tax is gone (ADR-139, Phase A); the root is just English.
• One localization, not eighteen. Because ways' language is es, the steady state is the root plus Spanish — the single active language — regenerated for the new or changed way. Not a matrix of every language the project ever shipped; only what this adopter actually uses.
• Re-localization is incremental and root-anchored. The changed way's Spanish alias is re-translated against the (possibly updated) English root and re-tuned. A drifting English root pulls its Spanish alias back into alignment — the root leads, the localization follows.
• Tuning gates the change. ways tune locale re-checks the touched alias: does it still align to the English root, and does it still avoid colliding with another way? Clean → done; flagged → re-author the stub (see 01.013.E).

The point

Localized maintenance is not a perpetual translation marathon — it is "write English, keep one shadow in sync." The cost scales with one language and with what changed, not with a language matrix or the whole corpus. This is the steady state the adopter-run model converges to, and it is sustainable for a single maintainer precisely because the English root does the anchoring, not a committee of equal translations.