Why

Level data currently crosses several validation boundaries with slightly different rules. createGame() infers object kinds and accepts legacy automaton direction strings, while the CommonJS loader validates a related shape and normalizes it later. Data-only sharing has another copy of the same normalization. This makes a level valid through one path but invalid through another, and makes future mechanics expensive to add safely.

What Changes

• Define one pure, host-owned level normalization and validation boundary for authored level data.
• Make direct engine creation, trusted CommonJS loading, and data-only sharing use that boundary before their own context-specific error adapters.
• Preserve the current legacy level shorthand during a documented migration: inferred kind, case-insensitive word tokens, and U/D/L/R/W automaton steps are accepted at input and materialized as canonical data.
• Add parity fixtures proving valid data, invalid data, legacy normalization, and CommonJS/share round trips agree before play or installation.

Capabilities

New Capabilities

• level-schema-boundary: canonical level normalization, validation, and adapters for engine, CommonJS, and data-only sharing.

Modified Capabilities

• commonjs-mods: use the canonical level boundary instead of maintaining a second level validator.
• level-sharing: validate and normalize shared levels through the same boundary used by local CommonJS previews and direct engine creation.

Impact

The implementation will add a small data/schema module and reduce duplicated level checks in src/engine.js, src/mods.js, and src/share.js. It will add focused tests for direct creation, local module loading, shared payloads, and export/import. Existing normalized pack fields, level IDs, rules semantics, and trusted-local CommonJS execution remain unchanged.