# KD-Tree aware-rupture Clifford attractor
# Perturb a,b,c,d ONLY when local density < THRESH (neighbors within R)
import numpy as np, matplotlib.pyplot as plt
from scipy.spatial import cKDTree

# ---- knobs (qualia) ----
SEED = 369
N = 120_000              # total steps
CADENCE = 300            # how often to check density / maybe mutate
WINDOW = 5000            # recent points used to estimate local density
R = 0.03                 # neighbor radius (smaller => rarer ruptures)
THRESH = 5               # mutate if neighbors < THRESH (gap)
HARD = np.array([.05,.05,.02,.02])  # mutation gains for (a,b,c,d)
MEM = 0.015              # breath memory (coherence)

# ---- init ----
rng = np.random.default_rng(SEED)
a,b,c,d = 1.22,-1.37,1.61,0.92
x=y=0.0
mu = np.zeros(2)
pts = np.empty((N,2), dtype=np.float32)

for i in range(N):
    xn = np.sin(a*y) + c*np.cos(a*x)
    yn = np.sin(b*x) + d*np.cos(b*y)
    pts[i] = (xn,yn)
    mu = (1-MEM)*mu + MEM*pts[i]
    x,y = xn,yn

    # every CADENCE steps, decide whether to mutate based on local density
    if i>WINDOW and i % CADENCE == 0:
        cloud = pts[i-WINDOW:i]                 # recent geometry only
        tree = cKDTree(cloud)
        # count neighbors of current point within radius R (exclude itself by using k>1)
        k = len(tree.query_ball_point(pts[i], r=R))
        if k < THRESH:
            # GAP => aware rupture guided by breath direction + a little noise
            m = np.linalg.norm(mu) + 1e-9
            dmu = mu / m
            jig = (rng.random(4)-0.5)
            a += HARD[0]*(dmu[0] + jig[0])
            b += HARD[1]*(dmu[1] + jig[1])
            c += HARD[2]*((dmu[0]-dmu[1]) + jig[2])
            d += HARD[3]*((dmu[1]-dmu[0]) + jig[3])
            mu *= 0.2  # partial forgetting so it stays alive

# ---- render ----
plt.figure(figsize=(6.6,6.6))
plt.axis('off'); plt.gca().set_aspect('equal')
plt.plot(pts[:,0], pts[:,1], ",k", alpha=0.15)
plt.show()