"""Independent exact early-exit fixture and ideal lockstep calculation.
Standard library only; no neural training, timing or benchmark data.
"""
import json,math
conf=[.99,.98,.95,.9,.85,.8,.75,.7,.6,.55]
early=[1,1,0,1,1,0,1,0,0,1];deep=[1,1,1,1,0,1,1,1,0,1]
thresholds=[.5,.7,.8,.9,.96,1.01];cost=[];accuracy=[];exits=[]
for t in thresholds:
 chosen=[a if c>=t else b for c,a,b in zip(conf,early,deep)]
 cost.append(sum(1 if c>=t else 5 for c in conf)/10)
 accuracy.append(sum(chosen)/10);exits.append(sum(c>=t for c in conf))
assert exits==[10,8,6,4,2,0]and accuracy==[.6,.6,.7,.7,.8,.8]and cost==[1,1.8,2.6,3.4,4.2,5]
assert any(a>b for a,b in zip(early,deep))and any(a<b for a,b in zip(early,deep))
es=[0,.25,.5,.75,.9,1];serial=[1+4*(1-e)for e in es];batch=[1+4*(1-e**8)for e in es]
assert serial[0]==batch[0]==5 and serial[-1]==batch[-1]==1
assert all(a<=b for a,b in zip(serial,batch))
figures=[dict(id='finite-exit-frontier',x=cost,y=accuracy),dict(id='lockstep-batch-cost',x=es,series=[dict(name='Serial per input',y=serial),dict(name='Lockstep completion of batch B=8',y=batch)])]
print(json.dumps(dict(status='PASS',scope='Author-stipulated ten-input fixture and independent fixed-cost batch model; not a neural benchmark or hardware timing.',figures=figures),indent=2))