Maddox: Architectural Audit

Existing audit utilities like Project Wallace and CSS Stats output raw, uncontextualized metrics that require active technical literacy to parse into business priorities. They tell you how many colors exist. They don't tell you which ones are breaking the system.

Maddox operates at the translation layer. Instead of counting discrete color values, the engine identifies Token Forking: clustering near-miss hex codes using programmatic threshold matching to isolate exactly where palette integrity collapses. The output is a structural verdict with specific, scoped remediation targets, not a raw statistics dump.

The v2.5 engine evaluates codebases against three architectural pillars: Color Efficiency, Typography Integrity, and Structural Debt. Color clusters are computed using RGB Euclidean distance (d <= 25). Near-miss hex codes within this threshold are flagged as candidate Token Forks for human review.

Typography scoring benchmarks against an ideal architectural scale of 8 to 12 discrete steps. Structural scoring maps z-index sprawl and spacing entropy. Each pillar is weighted (40/30/30) into a composite score with a deterministic A-F grade. The clusters are candidate fragmentation groups, not absolute violations. The manifest is an engineering roadmap, not an automated rewrite.

During the high-velocity deployment of our own studio site (grafikui.com), we ran Maddox against our production build to baseline the engine. The verdict: 79% (B), Structural Drift. The audit exposed 7 distinct variants of near-black hex codes and 32 non-standard font sizes embedded within our infrastructure.

Rather than ignoring the telemetry, we used the Maddox manifest to refactor our @theme source of truth, consolidating 105 fragmented utility declarations into a single, zero-drift atomic system. We built this utility because we needed it.