Color coordination rules

Skin-tone palettes, scoring formula, per-item flag thresholds. Implemented in color_rules.json and color_tool.score_outfit_colors().

R1 — Skin-tone palettes

color_rules.json defines three skin tones today, each with three palette lists:

Skin tone	Best colors	Good colors	Avoid colors
warm olive	terracotta, burnt orange, camel, olive, burgundy, warm white, gold, deep teal, mustard, chocolate brown	black, navy, blush, cream, sage green, rust	cool grey, icy pink, lavender, silver
cool fair	navy, icy blue, lavender, cool grey, silver, emerald, royal blue, raspberry	black, white, pink, plum, burgundy	orange, warm brown, gold-heavy palettes
deep warm	cobalt, emerald, bright red, gold, orange, white, camel, purple	black, navy, burgundy, teal	muted beige, dusty pink, light grey

Each tone also declares a metal_preference (gold / silver) and a notes line that ships into the reasoning trail.

R2 — Scoring formula

score = 60   # base

for each item in outfit:
    color = item.color (lowercased)
    if color matches any best_color:   score += 8
    elif color matches any good_color: score += 4
    elif color matches any avoid_color: score -= 10

score = clamp(0, 100)

R3 — Match function

Match is bidirectional substring: target in palette_color OR palette_color in target.

This catches: - "warm white" matches palette "white". - "dark indigo" matches palette "navy" indirectly via downstream rules (currently dark indigo just doesn't match — Phase 4 can refine).

The substring approach is permissive on purpose. Tight matching would underscore the system; loose matching surfaces in the reasoning trail.

R4 — Per-item notes & flags

For every item, the agent appends to the reasoning trail one of:

• "✓ {name} ({color}) — excellent color for your skin tone." (best)
• "✓ {name} ({color}) — works well for your skin tone." (good)
• "⚠ {name} ({color}) — this color is less flattering for your skin tone." (avoid)

The avoid-line is the only place where Wearly says a color is less flattering. It never says "bad" or "wrong." The note is a guidance, not a verdict.

R5 — Unknown skin tone fallback

If the user's skin tone isn't in color_rules.json, fall back to a universal palette: ["black", "navy", "white", "camel"] as best, no good list, no avoid list. The reasoning trail explains the fallback.

R6 — Score color (visual)

The Streamlit color card uses a warm-palette-only mapping for the score number:

Score	Color
≥ 80	#9F5A36 (deep terracotta)
60–79	#C17F5A (accent terracotta)
< 60	#8A4A20 (warm warning brown)

No greens, no cool tones. The score-color system stays inside the brand palette.

R7 — Cross-item color harmony (future)

The current scoring is per-item only. A future Phase 4 enhancement scores the outfit as a whole — checking that the items don't visually clash with each other. Hue clusters, neutral counts, and accent management would all factor in. The per-item score remains for transparency.

---

Source basis

Primary category: Color harmony (docs/evidence-and-references.md §3.3). Current basis: Color-theory conventions (warm / cool framework, complementary / analogous principles). The named-color palettes are design-informed; they do not claim empirical proof that these specific colors are "objectively best" for a given skin tone. The honest claim is "these colors tend to read well under warm/cool color-theory conventions." Verified citations: Itten, J. (1961). Kunst der Farbe / The Elements of Color (English ed. 1970, Van Nostrand Reinhold / Wiley); Munsell, A. H. (1905). A Color Notation (public domain via Project Gutenberg #26054); ISO 11664-4:2008 Colorimetry — Part 4: CIE 1976 Lab Colour space. These are the foundational frameworks Wearly's named palettes draw on as background — Wearly does not compute CIE distances at runtime. See docs/evidence-and-references.md §3.3 for full citations and the relationship to Wearly's rule. *Pending verification:** Any peer-reviewed perceptual-psychology study on skin-tone × garment-color preference would further strengthen this rule pack.