Skip to main content

0035: Retire Unconsumed Method-Level Visual Taxonomy

Status: Accepted

Date: 2026-07-22

Context​

The detached visual-evidence and early Visual Almanac experiments introduced a large cross-project vocabulary: Visual Subject, Visual Truth Mode, Temporal Presentation Mode, Fixed View, Seeded Exploratory View, Visual Comparison, AI Visual Review, Human Visual Review, and Visual Baseline. Those terms helped describe schemas, manifests, contact sheets, and gallery controls that STAGE later retired or made project-specific.

The active glossary, skill, project map, and review guide kept much of that taxonomy after its owning interfaces disappeared. As a result, an ordinary request to inspect animation or lighting could appear to require cataloging a subject, declaring two modes, preserving coordinates, and packaging a review. The terminology recreated the workflow pressure that the implementation had already rejected.

One semantic lesson remains important: a still image or a sequence of hard cuts does not prove meaningful frame-to-frame production motion. That lesson does not require a method-wide type system.

Decision​

Keep the active STAGE visual contract small:

  • state one visual question;
  • identify the production consumer and inspect the smallest native route that contains its material context;
  • use real project content and actual production playback;
  • do not claim motion from stills or discrete state cuts;
  • use human judgment for visual acceptance; and
  • retain captures, baselines, comparison coordinates, or project-specific labels only when a named project consumer needs them.

Remove the retired visual terms from the active STAGE glossary, self-map, and ordinary review procedure. Projects may use their own names for persistent galleries, regression systems, animation states, fixtures, cameras, and baselines. Historical ADRs and case studies keep the old terminology where it is necessary to explain the experiment and its failure.

This decision does not retire visual regression testing or project-owned galleries. It removes STAGE's implication that those optional capabilities share one universal vocabulary or declaration protocol.

Consequences​

  • Ordinary visual review requires less ceremony and fewer invented entities.
  • The useful still-versus-motion distinction remains explicit in plain language.
  • A project-owned regression or gallery feature owns its own terminology and compatibility contract.
  • Historical evidence remains understandable without promoting its taxonomy into current agent guidance.
  • New method terms need an active interface, recurring decision, or cross-project ambiguity to justify their cognitive cost.

Evidence Boundary​

This correction follows the same-owner Circussy and Lanternworks visual experiments. It demonstrates that these STAGE terms lacked a current consumer; it does not show that formal visual taxonomies are useless in projects with real regression, review, or content-browser systems.

Last updated on