diff --git a/benchmarks/ovm-build.sh b/benchmarks/ovm-build.sh
@@ -53,15 +53,18 @@ extract-other() {
  time for f in $TAR_DIR/*gz; do
    tar -x --directory $TAR_DIR --file $f 
  done
-  ls -l $TAR_DIR
 }

 extract-oil() {
  local target=_release/oil.tar
  rm -f -v $target
  make $target
  tar -x --directory $TAR_DIR --file $target
-  ls -l $TAR_DIR
+}
+
+extract() {
+  extract-oil
+  extract-other
 }

 # NOTE: build/test.sh measures the time already.
@@ -153,10 +156,13 @@ build-task() {

  case $action in
    configure)
+      "${TIME_PREFIX[@]}" -- ./configure
+
      # Cleaning here relies on the ORDER of tasks.txt.  configure happens
      # before build.  The Clang build shouldn't reuse GCC objects!
+      # It has to be done after configure, because the Makefile must exist!
+
      make clean
-      "${TIME_PREFIX[@]}" -- ./configure
      ;;
    make)
      "${TIME_PREFIX[@]}" -- make CC=$compiler_path
@@ -230,7 +236,7 @@ measure() {
  local prefix=${name%.compiler-provenance.txt}  # strip suffix

  local times_out="$raw_dir/$prefix.times.csv"
-  mkdir -p $BASE_DIR/{raw,stage1}
+  mkdir -p $BASE_DIR/{raw,stage1} $raw_dir

  # TODO: the $times_out calculation is duplicated in build-task()0

@@ -250,4 +256,31 @@ measure() {
  cp -v $provenance $raw_dir
 }

+#
+# Data Preparation and Analysis
+#
+
+stage1() {
+  local raw_dir=${1:-$BASE_DIR/raw}
+
+  local out=$BASE_DIR/stage1
+  mkdir -p $out
+
+  local times_csv=$out/times.csv
+  # Globs are in lexicographical order, which works for our dates.
+  local -a a=($raw_dir/flanders.*.times.csv)
+  local -a b=($raw_dir/lisa.*.times.csv)
+  csv-concat ${a[-1]} ${b[-1]} > $times_csv
+
+  # Construct a one-column CSV file
+  local raw_data_csv=$out/raw-data.csv
+  { echo 'path'
+    echo ${a[-1]}
+    echo ${b[-1]}
+  } > $raw_data_csv
+
+  head $out/*
+  wc -l $out/*
+}
+
 "$@"
diff --git a/build/actions.sh b/build/actions.sh
@@ -183,7 +183,10 @@ clean-repo() {
  clean-pyc
 }

-# 'make clean'.  NOTE: There are no object files written now.
+# 'make clean'.
+# NOTE:
+# - There are no object files written now.
+# - We're not cleaning _build/detect-config.* ?
 clean-source-tarball-build() {
  rm -f -v _bin/oil.* 
  rm -f -v _build/oil/ovm _build/oil/ovm-dbg
diff --git a/build/codegen.sh b/build/codegen.sh
@@ -51,6 +51,16 @@ install-re2c() {

 re2c() { _deps/re2c-1.0.3/re2c "$@"; }

+download-clang() {
+  wget --directory _deps \
+    http://releases.llvm.org/5.0.1/clang+llvm-5.0.1-x86_64-linux-gnu-ubuntu-16.04.tar.xz
+}
+
+extract-clang() {
+  cd _deps
+  time tar -x --xz < clang*.tar.xz
+}
+
 types-gen() {
  PYTHONPATH=. osh/ast_gen.py "$@" > _devbuild/gen/osh-types.h
 }
diff --git a/build/common.sh b/build/common.sh
@@ -7,7 +7,7 @@ set -o errexit
 # TODO: This changes depending on the version.  Maybe there should be a 'clang'
 # function for things that really require clang, like code coverage and so
 # forth.
-readonly CLANG_DIR=~/install/clang+llvm-5.0.1-x86_64-linux-gnu-ubuntu-16.04
+readonly CLANG_DIR=$PWD/_deps/clang+llvm-5.0.1-x86_64-linux-gnu-ubuntu-16.04
 readonly CLANG=$CLANG_DIR/bin/clang
 readonly CLANGXX=$CLANG_DIR/bin/clang++

diff --git a/core/id_kind_gen.py b/core/id_kind_gen.py
@@ -12,7 +12,7 @@

 import sys

-from asdl.gen_cpp import FormatLines
+from asdl.visitor import FormatLines
 from osh.meta import Id, Kind, LookupKind, ID_SPEC