"""Independent finite construction checks. Standard library only; no model training.
Run: python this_file.py. Stdout contains the two figure datasets.
"""
import json, math
from itertools import product, combinations
from fractions import Fraction as F
figures=[]
def cross(p,q):
 if not 0<p<1 or not 0<q<1:raise ValueError("probability out of domain")
 return -p*math.log2(q)-(1-p)*math.log2(1-q)
p=.9;qs=[.1,.2,.5,.8,.9];h=cross(p,p);cost=[cross(p,q)for q in qs]
for q,c in zip(qs,cost):assert math.isclose(c-h,p*math.log2(p/q)+(1-p)*math.log2((1-p)/(1-q)),abs_tol=1e-12)
for bad in [0,1,-.1,1.1]:
 try:cross(p,bad)
 except ValueError:pass
 else:raise AssertionError("invalid probability accepted")
figures.append(dict(id="entropy-mismatch",x=qs,series=[dict(name="Cross-entropy, source p=0.9",y=cost),dict(name="Source entropy",y=[h]*5)]))
ns=[8,16,32,64,128];assert next(n for n in range(1,1000)if n*h+16<n)==31
figures.append(dict(id="side-information-crossover",x=ns,series=[dict(name="Uniform model",y=ns),dict(name="Matched model + 16 side bits",y=[n*h+16 for n in ns])]))
assert len(figures)==2
print(json.dumps({"status":"PASS","scope":"Constructed analytic examples only; not neural training, a benchmark, or measured hardware performance.","figures":figures},indent=2))