Skip to main content

0040: Require Decision-Bearing Artifact Consumers

Status: Accepted

Date: 2026-07-22

Context​

ADR 0038 put galleries and almanacs back in the ordinary project backlog, and the visual-review skill made live native inspection the default. A qualified dogfood probe using the installed 0.5.6 plugin still found a semantic escape hatch. It rejected a universal capture matrix, used the project's existing Almanac, then proposed an external contact-sheet generator and "thin export harness" for generic human and AI review.

The probe followed the words "named consumer" while violating their intent. "Human", "AI", "reviewer", "recurring inspection", and "standing capability" name audiences or activities. They do not name the decision, operation, portability need, retention requirement, owner, or retirement condition that would make a persistent artifact useful.

Decision​

  • Require a decision-bearing artifact consumer before retaining visual capture or export infrastructure.
  • The consumer must identify a concrete decision or operation and why the live native surface cannot serve it.
  • Valid needs include remote access without the engine, stable revision or platform comparison, objective CI, audit, release evidence, and durable retention of an accepted decision.
  • Generic human review, AI review, repeated inspection, or a desired standing capability is insufficient by itself.
  • After rejecting a detached workflow, do not reintroduce it as an optional approval matrix, contact-sheet generator, capture manifest, or thin adapter.
  • Decompose broad inspection requests into bounded visual questions and use the existing native project route unless the artifact gate is independently met.
  • An existing project gallery or Almanac does not imply a corresponding export layer. The project feature and artifact transport have separate consumers.

Consequences​

  • Agents must explain the decision value of persistence instead of merely naming who might look at an artifact.
  • Co-located owner review remains complete without a package, even when it is frequent.
  • Legitimate remote, comparison, CI, audit, release, and retention workflows remain available, but their minimum outputs and owners become explicit.
  • A request for comprehensive visual quality is narrowed into questions rather than converted into a fixed capture matrix by default.

Evidence Boundary​

This decision comes from one behavioral probe against the same-owner Lanternworks dogfood project. It demonstrates an instruction ambiguity, not the general cost or value of every capture pipeline. Future projects may justify exports through concrete consumers; this ADR defines the gate they must pass, not a ban on visual artifacts.

Last updated on