Methodology · open-source

The editorial coherence score computed.

The score is a single number between 0 and 100. It is computed locally in the visitor's browser. Nothing is sent to a server. The formula is published below.

The formula.

score = 100
      − (palette_entropy   × 30)   
      − (typeface_variance × 25)   
      − (image_density_outlier × 15)
      − (saturation_outlier × 20)  
      + hue_bucket_modifier        

      clamped to 0–100

Each variable, plainly.

Palette entropy: The extracted colours are binned into 12 hue ranges of 30 degrees each. Shannon entropy is computed on the distribution. Normalised 0 to 1. More entropy = colours scattered across the spectrum = weaker visual cohesion.
Typeface variance: One or two type families = 0. Three = 0.5. Four or more = 1. The penalty assumes editorially mature brands work within disciplined type pairs.
Image density outlier: Currently set to 0 in this version. Reserved for a future iteration that measures inconsistent image density across the visited landing page.
Saturation outlier: Measured as the gap between the most saturated extracted colour and the mean saturation across all colours. A gap of 50 saturation points scores 1.0. Penalises a single hyper-saturated colour amid an otherwise muted palette.
Hue bucket modifier: The matched bucket (12 total) brings its own signal — some palette choices sit higher than others in editorial defensibility. Range: −12 to +5.

Reading the result.

75 and above · coherent editorial system
50 to 74 · fragmented at two or three points
25 to 49 · inconsistent · clear opportunity
below 25 · no detectable editorial system

The score is one signal among others. It does not replace judgment. The reading paragraph beside it is doing most of the work.

Source code reference: see src/lib/score.ts and src/content/readings.json in the public repository.