"""Independent fixed-witness audit; no producer/referee modules are used.""" import csv import hashlib import json import re from pathlib import Path from fractions import Fraction import numpy as np from scipy.optimize import isotonic_regression from pypdf import PdfReader from flint import arb, ctx HERE = Path(__file__).resolve().parent SRC = HERE / 'evidence' PRODUCER = HERE.parent.parent.parent / 'friends_new_victor_1104' ctx.prec = 384 def sha(p): return hashlib.sha256(p.read_bytes()).hexdigest() # Read the original supplementary PDF, first rate column, in exact tenths. rate_text = '\n'.join(p.extract_text() for p in PdfReader(SRC/'rudorf_s2.pdf').pages) rates = {} for line in rate_text.splitlines(): m = re.fullmatch(r'([ACGU]{3})\s+(\d+\.\d)\s+\d+\.\d\s+\d+\.\d\s+\d+\.\d', line.strip()) if m: rates[m[1].replace('U','T')] = int(Fraction(m[2])*10) assert len(rates) == 61 rows = list(csv.DictReader((SRC/'sequences.csv').open(encoding='utf-8-sig'))) assert len(rows) == 23 profiles = [] for row in rows: sequence = row["Sequence 5' → 3'"].strip().upper() assert len(sequence) % 3 == 0 codons = [sequence[k:k+3] for k in range(0,len(sequence),3)] assert codons[-1] in {'TAA','TAG','TGA'} assert all(c in rates for c in codons[:-1]) profiles.append([rates[c] for c in codons[:-1]]) assert {len(p) for p in profiles} == {387} labels = [r['Construct name'] for r in rows] # Candidate file has no labels: explicitly check its accompanying row manifest. assert labels == json.loads((PRODUCER/'results/sequence_labels.json').read_text()) pairs = [(i,j) for i in range(23) for j in range(i+1,23)] sq_input = [sum((a-b)**2 for a,b in zip(profiles[i],profiles[j])) for i,j in pairs] assert min(sq_input)>0 order = sorted(range(253), key=lambda k:sq_input[k]) ordered_q = [sq_input[k] for k in order] def ball(q): q = Fraction(q) return arb(q.numerator)/arb(q.denominator) def pava(y, weights=None): blocks=[] for k,value in enumerate(y): w = 1 if weights is None else weights[k] blocks.append([k,k+1,float(value)*w,w]) while len(blocks)>1 and blocks[-2][2]/blocks[-2][3] > blocks[-1][2]/blocks[-1][3]: right=blocks.pop(); left=blocks.pop() blocks.append([left[0],right[1],left[2]+right[2],left[3]+right[3]]) fit=np.empty(len(y)) for a,b,s,w in blocks: fit[a:b]=s/w return fit,blocks def evaluate(path): coords = list(csv.reader(path.open()))[1:] assert len(coords)==23 and all(len(r)==2 for r in coords) xy = [[Fraction(v) for v in row] for row in coords] sq = [sum((xy[i][d]-xy[j][d])**2 for d in range(2)) for i,j in pairs] y = np.array([float(sq[k])**0.5 for k in order]) fit,blocks = pava(y) assert np.max(np.abs(fit-isotonic_regression(y).x))<1e-10 stress = 100*np.sqrt(np.sum((y-fit)**2)/np.sum(y*y)) # Independently recompute distances and feasible disparities with Arb. distances = [ball(sq[k]).sqrt() for k in order] means = [sum(distances[a:b],arb(0))/(b-a) for a,b,_,_ in blocks] assert all(means[k]=0: kkt=False cert = 100*(residual/sum((ball(v) for v in sq),arb(0))).sqrt() assert cert