diff --git a/paper/diff.pdf b/paper/diff.pdf
index 74e748f..a128bfb 100644
Binary files a/paper/diff.pdf and b/paper/diff.pdf differ
diff --git a/paper/diff.tex b/paper/diff.tex
index dd2a161..f124048 100644
--- a/paper/diff.tex
+++ b/paper/diff.tex
@@ -1,7 +1,7 @@
 \documentclass[english]{article}
 %DIF LATEXDIFF DIFFERENCE FILE
 %DIF DEL old.tex    Wed Aug 11 10:03:33 2021
-%DIF ADD main.tex   Wed Aug 11 10:21:36 2021
+%DIF ADD main.tex   Thu Aug 12 15:01:10 2021
 \usepackage{graphicx}
 \usepackage{amsmath}
 \usepackage{hyperref}
@@ -854,7 +854,7 @@ \subsection*{Cognitively relevant dynamic high-order correlations in
 and orders) with the average accuracies across all of the
 kernel parameters we examined.  Using Figure~\ref{fig:decoding}c as a
 template, the best-matching kernel was a Laplace kernel with a width
-of 50 (Fig.~\ref{fig:kernels}d; also see Fig.~\pca).  We used this kernel to compute a
+of 50 (\DIFdelbegin \DIFdel{Fig.~\ref{fig:kernels}d; also see }\DIFdelend \DIFaddbegin \DIFadd{see Kernel-based approach for computing dynamic correlations and }\DIFaddend Fig.~\pca).  We used this kernel to compute a
 single $K$ by $K$ $n^\mathrm{th}$-order DISFC matrix for each
 experimental condition.  We then used Neurosynth~\DIFdelbegin \DIFdel{\mbox{%DIFAUXCMD
 \citep{RubiEtal17} }\hspace{0pt}%DIFAUXCMD
@@ -1945,8 +1945,8 @@ \subsubsection*{Reverse inference}
 }\href{https://github.com/ContextLab/timecorr-paper/releases/tag/v0.4}{github.com/ContextLab/timecorr-paper/releases/tag/v0.4}
 \DIFadd{and has been deposited in the Zenodo database under accession code
 }\href{https://doi.org/10.5281/zenodo.5165253}{https://doi.org/10.5281/zenodo.5165253}\DIFadd{.
-The source data underlying Figs. 2-6 and Supplementary Figs. S1-S9 are
-provided as a Source Data file. Source Data are provided with the
+The source data underlying Figures~2--6 and Supplementary Figures~S1--S9 are
+provided as Source Data files. Source Data are provided with the
 manuscript. The raw fMRI data are protected and are not available due to data
 privacy laws. The processed fMRI dataset collected by
 \mbox{%DIFAUXCMD
diff --git a/paper/main.pdf b/paper/main.pdf
index 0f317ec..909708d 100644
Binary files a/paper/main.pdf and b/paper/main.pdf differ
diff --git a/paper/main.tex b/paper/main.tex
index e2e289a..cd554f5 100644
--- a/paper/main.tex
+++ b/paper/main.tex
@@ -608,7 +608,7 @@ \subsection*{Cognitively relevant dynamic high-order correlations in
 and orders) with the average accuracies across all of the
 kernel parameters we examined.  Using Figure~\ref{fig:decoding}c as a
 template, the best-matching kernel was a Laplace kernel with a width
-of 50 (Fig.~\ref{fig:kernels}d; also see Fig.~\pca).  We used this kernel to compute a
+of 50 (see Kernel-based approach for computing dynamic correlations and Fig.~\pca).  We used this kernel to compute a
 single $K$ by $K$ $n^\mathrm{th}$-order DISFC matrix for each
 experimental condition.  We then used Neurosynth~\cite{RubiEtal17} to
 compute the terms most highly associated with the most strongly
@@ -1476,8 +1476,8 @@ \section*{Data Availability}
 \href{https://github.com/ContextLab/timecorr-paper/releases/tag/v0.4}{github.com/ContextLab/timecorr-paper/releases/tag/v0.4}
 and has been deposited in the Zenodo database under accession code
 \href{https://doi.org/10.5281/zenodo.5165253}{https://doi.org/10.5281/zenodo.5165253}.
-The source data underlying Figs. 2-6 and Supplementary Figs. S1-S9 are
-provided as a Source Data file. Source Data are provided with the
+The source data underlying Figures~2--6 and Supplementary Figures~S1--S9 are
+provided as Source Data files. Source Data are provided with the
 manuscript. The raw fMRI data are protected and are not available due to data
 privacy laws. The processed fMRI dataset collected by
 \cite{SimoEtal16} has been made publicly available \cite{SimoEtal16b} at