UAT ownership reset: structural surface conformance plus three-bucket pre-production review

[BryanHarrisScripts]

Developer Brief

Purpose

Move Human UAT back to its intended job: verify the story, not police PlotPickle's interface.

The current Visual Director/WebMCP stack already captures 30 governed Skin V1 surfaces and treats Dashboard as the locked canonical reference, but the comparison is still too token-oriented. A surface can use legal colours, borders and typography while still being visibly non-conforming through dead space, broken media, duplicate/stacked shells, abnormal composition or a legacy-looking structure.

At the same time, the built-in Afterglow UAT currently exposes seven tested-surface handoffs. For the Human pre-production acceptance pass, that is too broad. The Human should primarily prove that the correct persistent Afterglow story/project is loaded and that the same project/address reaches the current pre-production buckets.

Product decisions from Human review — September 18, 2026

1. Dashboard remains the sole locked visual authority.
2. Visual Director/UI Continuity should also use the 20+ governed current surfaces as a peer corpus so it can recognize the established PlotPickle surface family rather than checking tokens in isolation.
3. Automated visual QA owns detection of non-conforming services/surfaces. Human UAT should not be the first place we discover obvious layout or broken-asset regressions.
4. Human UAT for the current pre-production slice is intentionally simple:
   • confirm the Afterglow working story is loaded;
   • expose/verify the persistent project ID;
   • inspect the same project/address through the three current buckets: Outline, Storyboard and Previs.
5. The pre-production UAT bucket list is expected to grow from three to five later. The implementation must make that list explicit/data-driven rather than hard-coding assumptions throughout the runner.

Part A — structural Visual Director

Keep Dashboard as the canonical reference and add peer-corpus structural evidence from the complete governed WebMCP surface set.

The rendered profile must collect enough evidence to detect:

• broken/unloaded image/media elements;
• extreme vertical dead space / abnormally sparse composition;
• duplicate/stacked application-shell or landmark patterns where observable;
• anomalous root/content geometry relative to the established surface corpus;
• existing typography, palette, spacing, hierarchy, border and control-density deviations.

Rules:

• obvious broken media is a blocking defect;
• structural anomalies that can be legitimate for a specialist workspace begin as advisories unless deterministically unsafe;
• findings must name the rendered symptom and actionable correction;
• a surface never becomes canonical because peers agree with it; Dashboard remains the locked authority;
• historical per-surface screenshots remain regression evidence only.

Part B — Visual QA skill

Update the Visual QA skill so agents explicitly inspect:

• structural composition, not only token conformance;
• dead/unused regions;
• broken media;
• duplicate shells/navigation;
• whether the target fits both Dashboard authority and the established governed surface family.

Part C — simplify Human UAT

Change the built-in UAT inspection area from the current seven handoffs to the three current pre-production buckets:

• Outline
• Storyboard
• Previs

Expose the persistent working-copy project ID beside the title/story address so the Human can confirm that every bucket is viewing the same project.

The UAT copy should state that this is the current three-bucket pre-production acceptance set and is designed to expand as later pre-production buckets become canonical.

Do not remove Write, Scene Workspace or Production capabilities from PlotPickle. They simply stop being Human UAT bucket buttons in this bounded pre-production check.

Acceptance criteria

• Visual Director still compares every governed surface against Dashboard.
• Visual Director also derives peer-corpus structural norms from the governed current surface set.
• Broken rendered image/media evidence can produce a blocker.
• Abnormal dead-space/root composition can produce an actionable advisory.
• Structural evidence is emitted into the existing visual-director report; no parallel report system is created.
• Visual QA skill explicitly covers dead space, broken media, duplicate shells and peer-corpus continuity.
• UAT displays the persistent Afterglow project ID.
• UAT exposes exactly the current Outline, Storyboard and Previs bucket handoffs.
• Each bucket preserves Block 17 / Mini-Block 1 and the same profile-owned working project authority.
• The bucket configuration is explicit and extensible for the planned future five-bucket model.
• Focused regressions green.
• Architecture Verification/WebMCP/Visual Director green.
• Merge only when green so Human can rerun UAT from production/main.

Related

• UAT Human test completion: in-place runner plus conforming Write and Storyboard surfaces #2222 / PR #2222 finish UAT in-place flow and converge tested surfaces #2223 — in-place UAT and surface convergence
• Pre-UAT convergence: canonical routing fence, complete surface scan, and visible UAT Guide #2189 — canonical routed UAT surface coverage
• Matrix Reference Artifact Contract #2124 — Matrix Experience Surface Contract
• Make Dashboard the active visual authority for every Skin V1 surface #1883 — Dashboard Visual Director authority