Skip to main content

0010: Declare Native Visual Review Capability Without Mirroring Content

Status: Capability-manifest decision superseded by ADR 0026; subject-ownership and operability observations retained

Historical note: the Visual Almanac Capability Declaration wording below records the original decision. ADR 0026 retired that unconsumed manifest, schemas, templates, validator, and compatibility promise. Current projects record a persistent native route in their ordinary project map and operating docs. Detached-export terminology and consequences are superseded by ADR 0025.

Context​

After making the engine-native Visual Almanac primary, STAGE still needed a credible way to describe whether a project could actually support human and agent inspection. A menu path or editor window proves that a human can find a surface; it does not prove that an agent can select a subject, wait for readiness, inspect a named state, reset it, or tear it down.

The first external Visual Rehearsal trial also copied subject identities into STAGE-owned files. That created a second visual-content catalog beside the game's real catalogs and made placeholder coverage easier to produce than useful inspection of authored content.

Decision​

  • Add a versioned Visual Almanac Capability Declaration describing the native inspection surface, its canonical subject sources, human and agent entry points, truth modes, presentation authority, readiness, reset, teardown, decision authority, limitations, and optional Evidence Export.
  • Keep subject inventories in the game engine's canonical catalogs, scenes, registries, or authoring assets. The STAGE declaration records how subjects are discovered; it does not mirror them.
  • Record human and agent entry points separately. A human menu or window does not satisfy agent operability.
  • Call an agent entry point available only when it can select and open a subject, read readiness, reset state, and perform teardown through a stable project operation.
  • Label each inspection path as authored state seek, deterministic simulation, composed runtime, or live runtime. Presentation media must not inflate the truth mode that produced it.
  • Keep camera, lighting, background, rendering, UI, procedural, and temporal presentation under human authority. Generated defaults may not overwrite human edits.
  • Keep Evidence Export optional and separately declared.

Consequences​

  • STAGE can distinguish a useful human-operated gallery from a genuinely agent-operable review surface without dismissing either one.
  • Projects can adopt the capability declaration before building an agent driver; deferred operability remains visible rather than implied.
  • Engine-owned content remains authoritative and cannot silently drift from an external STAGE subject list.
  • Player-facing codices, developer Almanacs, and export packages may share implementation, but their audiences and guarantees remain distinct.
  • The declaration adds one small artifact and validator, but avoids a much larger parallel visual-content model.
Last updated on