diff --git a/exact_quantum_factoring_v2.txt b/exact_quantum_factoring_v2.txt
index 1265ff3..8f354d0 100644
--- a/exact_quantum_factoring_v2.txt
+++ b/exact_quantum_factoring_v2.txt
@@ -28,7 +28,7 @@ uses multiplicative orders: if $a$ has order $r$ modulo $m$ and $r$ is
 even, then $\gcd(a^{r/2}-1, m)$ can split $m$. The splitting argument is
 proved directly below.
 
-./shor-1997-factoring-discrete-log.pdf
+./pdf/shor-1997-factoring-discrete-log.pdf
 
 Three obstacles stand between "bounded error" and "probability 1":
 
@@ -62,10 +62,10 @@ universality cannot replace an exact gate analysis. The construction
 below uses only the fixed gate set $G$, receives only $N$, and accepts
 no advice.
 
-./miller-1976-riemann-hypothesis-primality.pdf
-./mosca-zalka-2004-exact-qft-dlog.pdf
-./imran-2022-exact-order-finding.pdf
-./nishimura-ozawa-2009-perfect-equivalence.pdf
+./pdf/miller-1976-riemann-hypothesis-primality.pdf
+./pdf/mosca-zalka-2004-exact-qft-dlog.pdf
+./pdf/imran-2022-exact-order-finding.pdf
+./pdf/nishimura-ozawa-2009-perfect-equivalence.pdf
 
 Recent resource-oriented algorithms by Oded Regev (arXiv:2308.06572)
 and by Aurelien Pilatte (arXiv:2404.16450) also factor in polynomial
@@ -73,8 +73,8 @@ time; early correctness required number-theoretic hypotheses, later
 removed, and the guarantees remain probabilistic. The present axis is
 different: certainty and worst-case bounds in an exact model.
 
-./regev-2024-efficient-quantum-factoring.pdf
-./pilatte-2026-unconditional-quantum-factoring.pdf
+./pdf/regev-2024-efficient-quantum-factoring.pdf
+./pdf/pilatte-2026-unconditional-quantum-factoring.pdf
 
 Amplitude amplification (Lov K. Grover, Proc. 28th STOC (1996) 212-219;
 Gilles Brassard, Peter Hoyer, Michele Mosca, Alain Tapp, Contemp. Math.
@@ -85,9 +85,9 @@ ordinary amplification step. The factoring preparation makes
 success probability exactly 1/4 using fixed gates and dyadic
 arithmetic; the standard one-step reflection then makes it 1.
 
-./grover-1996-database-search.pdf
-./brassard-hoyer-mosca-tapp-2002-amplitude-amplification.pdf
-./brassard-hoyer-1997-exact-simon.pdf
+./pdf/grover-1996-database-search.pdf
+./pdf/brassard-hoyer-mosca-tapp-2002-amplitude-amplification.pdf
+./pdf/brassard-hoyer-1997-exact-simon.pdf
 
 The proof has two main ingredients, beyond the order-to-gcd reduction.
 
@@ -105,7 +105,7 @@ Ekert, Chiara Macchiavello, and Michele Mosca (Proc. R. Soc. Lond. A
 state, not a Fourier eigenstate. Testing the auxiliary return yields
 the interpolated matrix with its actual, unnormalized probability.
 
-./cleve-ekert-macchiavello-mosca-1998-quantum-algorithms.pdf
+./pdf/cleve-ekert-macchiavello-mosca-1998-quantum-algorithms.pdf
 
 Felix M. Lev's four-phase approximation to the order-sampling
 transform (arXiv:quant-ph/0109103, Sections 3-4) gives a related route
@@ -120,7 +120,7 @@ finite polynomial pieces give exact dyadic sums. Two small correction
 branches make the true-order mass exact; neither the true order nor its
 factorization is required to run the trial.
 
-./lev-2001-integral-shor.pdf
+./pdf/lev-2001-integral-shor.pdf
 
 Second, a factoring run also generates the prime factorizations of
 $p-1$ for its prime factors $p > 2$, recursively. This auxiliary
@@ -134,8 +134,8 @@ polynomial-congruence method of Manindra Agrawal, Neeraj Kayal, and
 Nitin Saxena (Ann. of Math. 160 (2004) 781-793), with the constants
 and auxiliary-prime choice established directly.
 
-./pratt-1975-via-kleinberg-tardos-08intractability-slides.pdf
-./agrawal-kayal-saxena-2004-primes-in-p.pdf
+./pdf/pratt-1975-via-kleinberg-tardos-08intractability-slides.pdf
+./pdf/agrawal-kayal-saxena-2004-primes-in-p.pdf
 
 Once a run has completed and its data have been verified, those data
 determine the exact probabilities needed to test earlier outputs.
@@ -159,7 +159,7 @@ verification of the auxiliary data and the all-good generation
 property. A second dyadic correction makes the full preparation
 success probability exactly 1/4.
 
-./ozols-roetteler-roland-2012-quantum-rejection-sampling.pdf
+./pdf/ozols-roetteler-roland-2012-quantum-rejection-sampling.pdf
 
 As a first computation, take $N = 15$ and $a = 2$, a unit mod 15. The
 powers of 2 mod 15 are $2, 4, 8, 1, 2, \ldots$, so