Skip to content

MAX-786/claude-3d-harness / library/cc/reference-analysis-validator

reference-analysis-validator

Verified

Measure and validate supplied reference images, wireframes, texture atlases, and Blender renders before declaring a reconstruction 1:1. Use when an asset must match a template, when visual feedback says the output is off, when part counts must be exact, or before exporting a brand mascot/logo reconstruction. Pairs with reference-to-3d, contour-to-mesh, orthographic-registration, atlas-uv-fitting, and Blender MCP.

references/metrics-and-thresholds.md

View on GitHub
# Metrics and thresholds for 1:1 reconstruction

## Mask IoU

`intersection(mask_ref, mask_render) / union(mask_ref, mask_render)`. Good for silhouette agreement. It punishes missing/extra visible area.

Recommended gates:

- >= 0.90: final flat/front-logo match.
- >= 0.85: acceptable mascot/soft-surface match if detail landmarks also pass.
- < 0.80: do not export as final.

## Centroid drift

Compare binary mask centroids from image moments. Use absolute pixels and percent of image diagonal.

Recommended gate: <= 12 px at 1024 px validation render, or <= 1.2% of width.

## Bounding box drift

Compare x/y center, width, height of reference and render masks.

Recommended gate: center <= 12 px; size <= 3% of image size.

## SSIM

Use as a perceptual diagnostic, not the only gate. SSIM can be unfair when materials/lighting differ; use it after binary masks pass.

Recommended gate: >= 0.75 for early textured validation; >= 0.88 for near-final flat-color overlay.

## Part-count gate

Part count is not a soft metric. If the manifest declares N structural components, the scene/export must contain exactly those N structural meshes. Decorative arcs, glow/context nodes, shadows, guide marks, and background rings must not be counted as structural components unless the manifest says so.

## Overlay image

Use two overlays:

1. reference as cyan/magenta, render as yellow/green, difference as red.
2. semi-transparent original reference behind front orthographic model render.

The final report should include metric JSON plus at least one visual overlay.
reverse-engineer-anythingmorluto/reaReverse engineer native, managed, Electron/JavaScript, packaged, firmware, and browser targets with REA, and build agent environments (AgentEnv bundles) of web applications from a live instance or a capture, and of Linux desktop apps with a task from a request. Use shipped-artifact or requested runtime evidence to explain features, compare versions, decompile code, or guide a reconstruction. Skip REA for ordinary source-repository architecture analysis.Stars: 471kUpdated Claude CodeCodexVerifiedInstall
diagnosing-superpowersobra/superpowersUse when a superpowers session went wrong and your human partner wants to know why — repeated work, ignored plans, stumbles, poor results, a skill that didn't fire, "it took too long", "why is it so expensive", "what is it doing" — or wants to build a bug report for the superpowers maintainers, for the current session or a past one identified by id or path, on any harness.Stars: 322kUpdated Claude CodeCodexVerifiedInstall
dispatching-parallel-agentsobra/superpowersUse when facing 2+ independent tasks that can be worked on without shared state or sequential dependenciesStars: 322kUpdated Claude CodeCodexVerifiedInstall
executing-plansobra/superpowersUse when executing an implementation plan in the current session as the implementer yourself — your human partner chose inline execution, or no subagent tool is availableStars: 322kUpdated Claude CodeCodexVerifiedInstall