"""Deterministic analytic certificates and synthetic implementation checks. Run: PYTHONDONTWRITEBYTECODE=1 python3 verify.py No original-field simulation or empirical power estimation. """ from pathlib import Path from fractions import Fraction as F from decimal import Decimal as D, localcontext import itertools, json, math, random from rational_cox import (exp_negative, log_void, add, sub, mul, scale, certificate as cox_certificate, enclose, outward) from affine_certificate import probe_certificate from shared_profile import demonstration, rules, bounds, solve from rational_profile import certificate as profile_rational_certificate from nonzero_overlap import deterministic_boundary, minimum_radius, stable_population, stable_count_certificate from rational_nonzero_overlap import certificate as nonzero_rational_certificate from log_core import fit_finite_mixture from paired_core import serializable HERE=Path(__file__).resolve().parent def interval_exp(a): # Monotonicity, with exact Fraction endpoints. return exp_negative(a[1],terms=120)[0],exp_negative(a[0],terms=120)[1] def cooperative_logs(A,C,rho=F(1,10)): logs=[] for j in (1,2): vv=[] for i in (1,2,3): a=rho*i eA=interval_exp(scale(A,a*j)) eC=interval_exp(scale(C,a)) term=sub((F(2),F(2)),eC) if j==1 else sub((F(4),F(4)),scale(eC,F(3))) vv.append(mul(eA,term)) logs.append({i:sub(log_void(vv[i-1],terms=180),log_void(vv[0],terms=180)) for i in (2,3)}) a,b=logs L=sub(mul(b[3],a[2]),mul(a[3],b[2])) return logs,L def enum_dimer(a,b,old): """Enumerate finite Poisson-occurrence cone for dimer events (not sampling).""" total=old+1 streams=[] for row in range(1,total+1): # Ancestor width total-row+1, sites -(total-row),...,0. for origin in range(-(total-row)-1,1): streams.append((row,origin,a if row<=old else b)) q=D(0) for bits in itertools.product((0,1),repeat=len(streams)): weight=D(1); rows=[set() for _ in range(total+1)] for bit,(row,origin,rate) in zip(bits,streams): v=(-rate).exp();weight*=1-v if bit else v if bit:rows[row].update((origin,origin+1)) H=set() for row in range(1,total+1): H=(H & {x+1 for x in H})|rows[row] if 0 in H:q+=weight return 1-q def counts_from_marginals(vs,n=10**12): """Deterministic nested five-bit fixture; no random draws.""" cuts=sorted(set([D(0),D(1)]+[1-v for v in vs])) cells={} for lo,hi in zip(cuts,cuts[1:]): if hi<=lo:continue midpoint=(lo+hi)/2 key=''.join('1' if midpoint<1-v else '0' for v in vs) cells[key]=cells.get(key,0)+int((hi-lo)*n) residual=n-sum(cells.values()) cells['00000']=cells.get('00000',0)+residual assert sum(cells.values())==n and min(cells.values())>=0 return {'outcomes':cells} def law_void(N,phi1=D(1),x=D('.5'),Z=D('.8'),step=D('.25')): rates=[phi1,phi1+2*x,phi1+2,phi1-2*step,phi1+2*step] return [Z*(-a*N).exp() for a in rates] def run(): findings={} checks={} logs,L=cooperative_logs((F(2),F(2)),(F(1),F(1))) checks['dimer_rational_sign']=F(2675298677,10**12)=2*epsilon*u1*u2. u1,u2=D('.2'),D('.4');popL=D('.02') checks['ordered_calibration_blind_region']=all(popL<=2*eps*u1*u2 for eps in [D('.5'),D('.8'),D(1)]) findings['rank_population_margin']={'L':str(popL),'epsilon_threshold':'0.125','margin_at_epsilon_0':str(popL),'margin_at_epsilon_0_125':'0'} (HERE/'fixture_rank_synthetic.json').write_text(json.dumps(rank_data,indent=2)) # A cubic signal with unbalanced u1=P^2,u2=P can retain a fixed # positive ratio-error budget. This checks the balanced-regime qualifier. asymP=D('.1');unequal=[] for uj,xj in [(asymP**2,D('.5')),(asymP,D('.25'))]: vb=D('.9')*(-uj/2).exp();probe=D('.25')*uj unequal.append(counts_from_marginals([vb,vb*(-xj*uj).exp(),vb*(-uj).exp(),vb*probe.exp(),vb*(-probe).exp()],n=10**14)) ur=probe_certificate({'alpha':'.05','contract':{'step':'.25','ratio_error':'.1'},'groups':unequal}) checks['unbalanced_cubic_signal_fixed_ratio_budget']=ur['rank_exclusion'] and not ur['shape_exclusion'] findings['unbalanced_rank_result']=serializable(ur) # Same safe probe data, but the main3 arm is always covered. unsafe=[] for group in coop: counts={} for key,val in group['outcomes'].items(): k=key[:2]+'1'+key[3:];counts[k]=counts.get(k,0)+val unsafe.append({'outcomes':counts}) r=probe_certificate({'alpha':'.05','contract':{'step':'.25','ratio_error':'0'},'groups':unsafe}) checks['safe_shape_despite_unsafe_main']=r['shape_exclusion'] and r['base_log_interval']['lower']==D('-Infinity') findings['unsafe_main_result']=serializable(r) # Valid arbitrary nonmonotone coupling; finite signed intervals remain valid. signed={'alpha':'.05','contract':{'step':'.25','ratio_error':'0'}, 'groups':[{'outcomes':{'11111':200000,'10000':1000,'00000':798000,'01111':1000}}]*2} r=probe_certificate(signed) checks['signed_negative_responses_accepted']=True # No logarithm is manufactured from all-covered data. r=probe_certificate({'alpha':'.05','contract':{'step':'.25','ratio_error':'0'}, 'groups':[{'outcomes':{'11111':10000}}]*2}) checks['unsafe_zero_void_no_certificate']=not r['affine_probe_null_exclusion'] triples=[['.01','.015','.09'],['.2','.3','.4'],['.001','.003','.009']] fit=fit_finite_mixture(triples) # inspect helper field names explicitly entries=fit.get('labels',fit.get('mixtures',fit.get('groups',[]))) findings['mixture_reconstruction']=serializable(fit) # Actual reconstruction comparison, independent of the helper's own residual field. items=fit['labels'] err=D(0) for original,item in zip(triples,items): rec=item['reconstructed_voids'] err=max(err,max(abs(D(1)-D(q)-v) for q,v in zip(original,rec))) checks['common_calibration_finite_mixture']=err=K u=-h(D(1)) tilted_mu=sum(p*a*(-D('.7')*a).exp() for p,a in zip(w,y))/sum(p*(-D('.7')*a).exp() for p,a in zip(w,y)) assert tilted_mu<=u+u.exp()*K/2+D('1e-65') checks['thousand_finite_law_variance_and_sharper_M_checks']=True sharp=demonstration(counts_from_marginals) findings['shared_profile']=serializable({k:v for k,v in sharp.items() if k!='fixture'}) rule=sharp['rules'];mom=sharp['moments'] checks['sharp_quadrature_matches_three_moments']=all( abs(sum(w*r**k for w,r in zip(rule[family+'_weights'],rule[family+'_nodes']))-mom[k-1])sharp['profiles'][1][1] checks['heterogeneous_gap_exceeds_twice_0_1']=sharp['gap']>D('.2') checks['heterogeneous_old_variance_test_inconclusive']=not sharp['old_certificate']['affine_probe_null_exclusion'] and sharp['L']0 and deltaD('.04214052') checks['nonzero_overlap_detects_beyond_variance_allowance']=not nz_old['affine_probe_null_exclusion'] and nz_count['overlap_exclusion'] checks['nonzero_overlap_finite_count_margin']=nz_count['margin_lower']>D('.03') # Expanded profiles pass because the two generating U values both lie in # [.125,.375], independently of the weak-source mass. from shared_profile import profile ips=[profile(mm,w) for w in ww] checks['expanded_profiles_pass_new_heterogeneous_alternative']=all(i[0]<=band[1]+D('1e-60') and i[1]+D('1e-60')>=band[0] for i in ips) commonvals=[beta*mm[0],beta*sum((D('1.5')*a.ln()).exp() for a in rb)/2,beta*mm[2],beta,beta*mm[1]] control={'alpha':'.05','contract':{'step':'.5','ratio_error':'.125'},'groups':[counts_from_marginals(commonvals)]*2} checks['compatible_common_environment_not_rejected_by_overlap']=not stable_count_certificate(control)['overlap_exclusion'] rational_nz=nonzero_rational_certificate(nz_data) checks['rational_nonzero_boundary_and_count_certificates']=all(rational_nz['assertions'].values()) checks['explicit_full_band_linear_count_certificate']=D(rational_nz['finite_linear_certificate']['margin']['lower_decimal'])>D('.0154') findings['nonzero_overlap']={'minimum_radius_example':serializable(critical),'population':serializable(nz_pop), 'count_certificate':serializable(nz_count),'old_certificate':serializable(nz_old),'profiles':serializable(ips)} findings['rational_nonzero_overlap']=rational_nz (HERE/'fixture_nonzero_overlap_synthetic.json').write_text(json.dumps(nz_data,indent=2)) (HERE/'nonzero_overlap_rational_certificate_output.json').write_text(json.dumps(rational_nz,indent=2)) (HERE/'fixture_null.json').write_text(json.dumps(null_data,indent=2)) (HERE/'fixture_cooperative_synthetic.json').write_text(json.dumps(coop_data,indent=2)) assert all(checks.values()),checks result={'identity':'analytic certificates and synthetic fixtures; no field simulation or empirical power', 'checks':checks,'findings':findings} (HERE/'verification_results.json').write_text(json.dumps(result,indent=2)) print(json.dumps({'checks':checks,'all_passed':True,'output':'verification_results.json'},indent=2)) if __name__=='__main__':run()