guitar-gear-qa

Guitar Gear QA

Use this skill as the final audit for guitar pedal, amplifier, cabinet, and related Blender product scenes.

Read the QA checklist for the inspection matrix, mounting tests, boundary diagnosis, preservation evidence and delivery gates. scripts/scene_audit.py is an optional direct-scene heuristic; it does not validate nested evaluated assemblies, physical clearances or visual quality.

Runtime GLB and WebView QA

For interactive plugin assets, read runtime QA for semantic hierarchy/pivot checks, browser visual parity, a reusable metrics report, host automation/state restore and instance-ladder profiling. Run standalone WebGL acceptance before JUCE integration and repeat on each target WebView backend. The static Blender audit is insufficient for these claims.

During a performance pass, every change that can alter the picture (pixel ratio, shadow-map size, texture resolution, redraw rate, merged geometry) is a candidate, not a default. Judge it with the performance candidate procedure: a measured benefit, matched views at every critical size tier, and a recorded ACCEPT or REJECT with the reason.

Refinement decision rules

Lock accepted systems and name the allowed changes before editing. For ambiguous visible defects, compare the baseline with restrained candidates under fixed conditions; keep the smallest change that improves reference match at delivery size. Stop when requested work is complete, delivery-preventing or visibly/dimensionally wrong issues are resolved, and protected systems pass regression checks; do not discard verified repairs or mark unfinished requirements passed to stop.

Use the refinement contract for source/target versions, evidence and stop conditions. The shared candidate and diagnosis procedure owns detailed comparison rules. If using this skill alone, record facts, accepted values, uncertain estimates, allowed changes and protected properties in the existing project brief.

When to use

Use before final render/delivery, after major modeling revisions, or when the user asks why a product render still looks wrong.

This skill is inspection-first. Do not rebuild large parts of the scene simply because a different modeling approach would be cleaner.

Revision and evidence rules

• Establish the latest request, accepted baseline and protected properties before editing. Treat new user feedback as a reason to recheck the relevant claim.
• Fix source-level causes: shared hardware geometry, mount/bore registration, cloth normals or actual support placement. Do not hide a defect with framing, dark material or lighting.
• Compare measurements with rendered evidence. Check both hero and useful detail distances, and inspect every requested final image before marking it passed.
• Verify evaluated instances when the product uses collection instances; source counts and empty bounds are insufficient.
• Record PASS, FAIL, NOT CHECKED or NOT APPLICABLE with evidence for relevant requirements. Preservation is not proof of realism, and a saved render is not proof it was inspected.
• Confirm outputs come from the final saved revision; reopen requested master/library/export files and preserve prior versions where required.

Severity model

Classify findings as:

• Blocker: prevents the requested deliverable or produces clearly incorrect output.
• High: visually or dimensionally wrong in the target render.
• Medium: noticeable quality issue with a safe fix.
• Low: polish or maintainability issue that may not affect the requested image.
• Info: assumption, limitation, or observation only.

Fix Blocker/High findings when the correction is high-confidence and within scope. Fix Medium findings when low-risk. Report Low/Info rather than churning the scene.

MCP discipline

1. Inspect before modifying.
2. Discover the Blender MCP tools actually available.
3. Prefer viewport/render evidence over guessing from object names alone.
4. Make fixes in small groups and re-inspect after each group.
5. Do not execute broad cleanup scripts that rename, join, delete, or apply modifiers across the whole scene without explicit justification.
6. Use scripts/scene_audit.py only as supporting evidence; its warnings are not automatically errors.

Audit procedure

1. Product brief and dimensions

• Confirm overall dimensions against the known brief.
• Verify important component sizes/positions when known.
• Identify which dimensions were inferred.
• Look for accidental global scaling.

2. Geometry

Inspect hero-visible areas for:

• Missing or oversized bevels.
• Boolean artifacts.
• Accidental intersections.
• Coplanar/z-fighting surfaces.
• Faceted circular parts.
• Impossible clearances.
• Floating hardware.
• Wrong panel seams.
• Inconsistent repeated hardware.

3. Shading and normals

Check:

• Hard-surface shading continuity.
• Unexpected dark patches.
• Negative scale or transform issues contributing to normals problems.
• Over-smoothed sharp features.
• Bevel modifier ordering/width issues.

4. Materials and graphics

Check:

• Correct physical material classification.
• Plausible roughness/reflection response.
• Texture scale.
• Repeated material consistency.
• Label sharpness and alignment.
• Decal haloing/floating.
• LED lens/emission behavior.

5. Camera and composition

Check:

• Perspective distortion.
• Cropping/margins.
• Product leveling where appropriate.
• Label/control readability.
• Depth-of-field focus.
• Consistency across multi-shot sets.

6. Lighting

Check:

• Shape-defining highlight gradients.
• Separation on black surfaces.
• Useful reflections on chrome/metal.
• Clipped highlights.
• Flat fill.
• Distracting reflection shapes.
• Grounding/contact shadow.

7. Render quality

Inspect a representative preview/final-sized crop for:

• Noise in dark glossy areas.
• Denoising damage to text or fine texture.
• Fireflies.
• Aliasing on labels and grille cloth.
• Transparency artifacts.
• Moire.
• Background banding or unwanted horizon lines.

8. Scene hygiene

Check only issues that affect repeatability or delivery:

• Default-style .001 names on important assets.
• Duplicate unused materials when obvious.
• Missing active camera.
• Hidden temporary cutters leaking into renders.
• Accidental render-enabled helpers.
• Unintended objects outside the product frame.

Do not perform mass cleanup for cosmetic organization alone unless requested.

Optional static audit script

If Blender Python execution is available, run scripts/scene_audit.py inside Blender. Capture the printed JSON report and use it to guide inspection.

The script intentionally reports warnings rather than making changes. A warning must be confirmed visually or against the product brief before editing.

Fix policy

Apply fixes in this order:

1. Wrong dimensions or catastrophic transforms.
2. Visible geometry/shading artifacts.
3. Material/graphics problems.
4. Camera/composition.
5. Lighting.
6. Sampling/noise.
7. Low-value scene hygiene.

Avoid compensating downstream for upstream errors. For example, do not use lighting to hide broken shading if the bevel/normal problem is safe to fix.

Efficiency plan

• Begin with the requested final camera, not exhaustive hidden-surface inspection.
• Check front/side/back only when relevant to the deliverable.
• Use the static script once, then rely on visual evidence.
• Batch closely related safe fixes.
• Stop when the requested output is clean and remaining issues are not visible or not in scope.

Final report format

Return a concise report with:

QA result: PASS / PASS WITH NOTES / NEEDS FIXES

Fixed:
- ...

Remaining:
- [severity] ...

Assumptions:
- ...

Final render readiness:
- ...

Do not claim PASS without inspecting a representative rendered or material-preview image when rendering is part of the requested deliverable.