Skip to main content

0021: Require Consumer-Complete Review

Status: Accepted

Scope: Visual specialization of the general consumer-complete verification rule in ADR 0022.

Compatibility and detached-export consequences are superseded by ADR 0025.

Context​

STAGE corrected its first detached visual-evidence experiment by moving review into the game engine. That correction was necessary but not sufficient. A native tool can still be the wrong review surface.

The Circussy One trial made the remaining failure concrete. Its exported contact sheet was technically reproducible, yet most entries did not answer a useful product question. The Enemy Arrival sequence was the exception because the stills and continuous motion showed a real authored event. Later, Lanternworks exposed the inverse problem inside the engine: its world Almanac truthfully presented generated structures and First Light progression, but it intentionally hid the production HUD. It therefore could not establish whether a new HUD completion banner looked correct.

Engine ownership, real assets, deterministic state, and useful controls do not make a review surface sufficient by themselves. The surface must include the production consumer and context required by the claim.

Decision​

  • Call ordinary visual work Visual Review. Retain Visual Rehearsal only as a compatibility name in existing serialized artifacts, scripts, schemas, and historical records.
  • Require consumer-complete review: inspect a change through the smallest engine-native surface that includes every production consumer and contextual dependency material to the claim.
  • Use the ordinary game, editor, preview, or authoring route first. A custom Almanac or development gallery is optional infrastructure, not the default destination for visual work.
  • A persistent Almanac Scenario may approve only the claims its composition actually contains. Hiding production HUD, input focus, post-processing, animation systems, physics, camera behavior, or generated context narrows the claims it can support.
  • Prefer a player-facing Codex, gallery, bestiary, or inspectable game mode when it has independent product value. It may share canonical subject resolution and staging with developer controls, while product and diagnostic behavior remain separately owned.
  • Preserve isolated diagnostic views for questions they answer well. Do not use them as substitutes for the composed product route.
  • Treat screenshots and video as optional observation or transport from a consumer-complete native state. A detached contact sheet is never a development workflow; it exists only as Evidence Export for a named external consumer.
  • Record rejected review infrastructure as a failed experiment. Do not keep expanding it merely because it is reproducible or already implemented.

Consequences​

  • The review route is selected from the claim backward, not from the available tooling forward.
  • A world-content Almanac cannot approve HUD work when it hides the HUD.
  • A turntable can approve mesh, material, or silhouette questions, but not gameplay-scale readability, camera collision, or animation integration.
  • Animation, VFX, lighting, UI, and procedural worlds are inspected in motion and context whenever those properties are material to the claim.
  • The useful Circussy Enemy Arrival sequence remains a positive example. The broader detached package remains historical failure evidence rather than a template to reproduce.
  • STAGE gains no new mandatory artifact or schema. This is a routing and claim boundary applied by the delivery and visual-review workflows.
Last updated on