Gay-MCP Skill: Deterministic Color Generation

Status: ✅ Production Ready Trit: +1 (PLUS - optimistic/generative) Principle: Same seed → Same colors (SPI guarantee) Implementation: Gay.jl (Julia) + SplitMixTernary (Ruby)

Overview

Gay-MCP provides deterministic color generation via SplitMix64 + golden angle. Every invocation with the same seed produces identical colors, enabling:

Parallel computation: Fork generators, get same results
Reproducibility: Colors are functions of (seed, index)
GF(3) trits: Each color maps to {-1, 0, +1}

Core Algorithm

SplitMix64:
  state = (state + γ) mod 2⁶⁴
  z = state
  z = (z ⊕ (z >> 30)) × 0xBF58476D1CE4E5B9
  z = (z ⊕ (z >> 27)) × 0x94D049BB133111EB
  return z ⊕ (z >> 31)

Color Generation:
  L = 10 + random() × 85    # Lightness: 10-95
  C = random() × 100        # Chroma: 0-100
  H = random() × 360        # Hue: 0-360
  trit = hue_to_trit(H)     # GF(3) mapping

Constants

GOLDEN = 0x9E3779B97F4A7C15  # φ⁻¹ × 2⁶⁴
MIX1   = 0xBF58476D1CE4E5B9
MIX2   = 0x94D049BB133111EB
MASK64 = 0xFFFFFFFFFFFFFFFF

MCP Server

The Gay MCP server provides these tools:

Tool	Description
`color_at`	Get color at specific index
`palette`	Generate N-color palette
`golden_thread`	Golden angle spiral
`reafference`	Self-recognition loop
`loopy_strange`	Generator ≡ Observer

Commands

# Start MCP server
julia --project=@gay -e "using Gay; Gay.serve_mcp()"

# Generate palette
just gay-palette seed=1069 n=12

# Test determinism
just gay-test

API (Ruby)

require 'splitmix_ternary'

# Create generator
gen = SplitMixTernary.new(1069)

# Get color at index
color = gen.color_at(42)
# => { L: 45.2, C: 67.8, H: 234.5, trit: -1, index: 42 }

# Generate trits
gen.next_trit  # => -1, 0, or +1

# Split for parallelism
child = gen.split(7)  # Independent child generator

API (Julia)

using Gay

# Set seed
Gay.gay_seed(1069)

# Get color
color = Gay.color_at(42)

# Generate palette
palette = Gay.palette(12)

# Golden thread
colors = Gay.golden_thread(steps=10)

API (Python)

from gay import SplitMixTernary, TripartiteStreams

# Create generator
gen = SplitMixTernary(seed=1069)

# Get color at index (returns dict with L, C, H, trit, hex)
color = gen.color_at(42)
# => {'L': 45.2, 'C': 67.8, 'H': 234.5, 'trit': -1, 'index': 42, 'hex': '#2E5FA3'}

# Generate hex color directly
hex_color = gen.hex_at(42)  # => '#2E5FA3'

# Generate palette as hex list
palette = gen.palette_hex(n=12)
# => ['#D8267F', '#2CD826', '#4FD826', ...]

# Generate trits
trit = gen.next_trit()  # => -1, 0, or +1

# Split for parallelism (deterministic child)
child = gen.split(offset=7)

# Tripartite streams (GF(3) = 0 guaranteed)
streams = TripartiteStreams(seed=1069)
triplet = streams.next_triplet()
# => {'minus': -1, 'ergodic': 0, 'plus': 1, 'gf3_sum': 0, 'conserved': True}

Hex Color Output (#RRGGBB)

Convert OkLCH to hex for CSS/web usage:

def oklch_to_hex(L: float, C: float, H: float) -> str:
    """Convert OkLCH to #RRGGBB hex string."""
    import math
    
    # OkLCH -> OkLab
    a = C * math.cos(math.radians(H))
    b = C * math.sin(math.radians(H))
    
    # OkLab -> Linear RGB (simplified)
    l_ = L/100 + 0.3963377774 * a + 0.2158037573 * b
    m_ = L/100 - 0.1055613458 * a - 0.0638541728 * b
    s_ = L/100 - 0.0894841775 * a - 1.2914855480 * b
    
    l, m, s = l_**3, m_**3, s_**3
    
    r = +4.0767416621 * l - 3.3077115913 * m + 0.2309699292 * s
    g = -1.2684380046 * l + 2.6097574011 * m - 0.3413193965 * s
    b = -0.0041960863 * l - 0.7034186147 * m + 1.7076147010 * s
    
    # Clamp and convert to 0-255
    def to_byte(x): return max(0, min(255, int(x * 255)))
    
    return f"#{to_byte(r):02X}{to_byte(g):02X}{to_byte(b):02X}"

Integration with discrete_backprop

Color learning via gradient-free optimization:

from gay import SplitMixTernary
from discrete_backprop import DiscreteBackprop

class ColorLearner:
    """Learn optimal color sequences via trit-based backprop."""
    
    def __init__(self, seed: int, target_palette: list):
        self.gen = SplitMixTernary(seed)
        self.target = target_palette
        self.backprop = DiscreteBackprop(dims=3)  # L, C, H
    
    def loss(self, index: int) -> float:
        """Compute color distance to target."""
        color = self.gen.color_at(index)
        target = self.target[index % len(self.target)]
        return sum((color[k] - target[k])**2 for k in ['L', 'C', 'H'])
    
    def step(self, indices: list) -> dict:
        """Discrete gradient step via trit perturbation."""
        losses = [self.loss(i) for i in indices]
        
        # Compute trit-based gradient (Δtrit per dimension)
        gradients = self.backprop.discrete_gradient(
            params=[self.gen.state],
            losses=losses,
            perturbation='trit'  # Use {-1, 0, +1} perturbations
        )
        
        # Chain seed based on gradient direction
        direction_trit = sum(g for g in gradients) % 3 - 1  # Map to {-1, 0, +1}
        self.gen.chain_seed(direction_trit)
        
        return {
            'loss': sum(losses) / len(losses),
            'gradient_trit': direction_trit,
            'new_seed': hex(self.gen.seed)
        }

# Usage
learner = ColorLearner(seed=0x42D, target_palette=[
    {'L': 50, 'C': 80, 'H': 30},   # Target warm
    {'L': 70, 'C': 60, 'H': 150},  # Target neutral
    {'L': 40, 'C': 90, 'H': 240},  # Target cold
])

for epoch in range(100):
    result = learner.step(indices=[0, 1, 2])
    if result['loss'] < 0.01:
        break

Tripartite Streams

Three independent streams with GF(3) = 0:

streams = SplitMixTernary::TripartiteStreams.new(seed)

triplet = streams.next_triplet
# => { minus: -1, ergodic: 0, plus: 1, gf3_sum: 0, conserved: true }

Trit Mapping

Hue 0-60°, 300-360° → +1 (PLUS, warm)
Hue 60-180°         →  0 (ERGODIC, neutral)
Hue 180-300°        → -1 (MINUS, cold)

Out-of-Order Proof

proof = SplitMixTernary.prove_out_of_order(seed)
# => { 
#      ordered_equals_reversed: true,
#      ordered_equals_shuffled: true,
#      proof: "QED: Math is doable out of order"
#    }

Integration with Langevin Dynamics (NEW)

Track which colors affect which noise calls in Langevin training:

# Instrument Langevin noise via color tracking
function instrument_langevin_noise(sde, step_id)
    color = color_at(rng, step_id)
    noise = randn_from_color(color)
    return (color, noise, step_id)
end

# Export audit trail showing cause-effect
audit_log = export_color_trace(
    trajectory=sde_solution,
    seed=base_seed
)
# Shows: step_47 → color_0xD8267F → noise_0.342 → parameter_update

# Verify GF(3) conservation across trajectory
gf3_check(color_sequence, balance_threshold=0.1)

Integration with Unworld

Colors are derived, not temporal:

# Seed chaining
next_seed = Unworld.chain_seed(current_seed, color[:trit])

# Derive color
color = Unworld.derive_color(seed, index)

Example Output

╔═══════════════════════════════════════════════════════════════════╗
║  GAY.JL: Deterministic Color Generation                          ║
╚═══════════════════════════════════════════════════════════════════╝

Seed: 0x42D

─── Palette (12 colors) ───
  1: #D8267F (trit=+1)
  2: #2CD826 (trit=0)
  3: #4FD826 (trit=0)
  ...

─── Out-of-Order Proof ───
  Indices: [1, 5, 10, 20, 50]
  Ordered = Reversed: true
  Ordered = Shuffled: true
  QED: Math is doable out of order

Skill Name: gay-mcp Type: Deterministic Color Generation Trit: +1 (PLUS) GF(3): Conserved via tripartite streams SPI: Guaranteed (same seed → same output)

gay-mcp