diff --git a/debug_translate b/debug_translate
new file mode 100755
index 00000000..39075aac
--- /dev/null
+++ b/debug_translate
@@ -0,0 +1,26 @@
+#!/bin/sh
+# Translate SubX files with debug information.
+#
+# Mu's core tooling has a gap:
+#   1. 'translate' can generate debug information, but syntax sugar passes
+#   (sigils and calls) have no C++ versions and take several minutes to run
+#   emulated.
+#   2. 'ntranslate' is fast, but you get no trace for runs and zero error-checking
+#   on the code emitted by sigils and calls. Which could still be buggy.
+#
+# This script is a hack to get the best of both worlds. We run natively what
+# we must, and leverage as much debug information as possible. This arrangement
+# is snappy but requires Linux just like 'ntranslate'. You also are on your
+# own to mentally map desugared instructions in traces and error messages back
+# to the original sources.
+
+set -e
+
+echo "  calls"
+cat $*          |apps/calls           > a.calls
+echo "  sigils"
+cat a.calls     |apps/sigils          > a.sigils
+
+subx --debug translate a.sigils -o a.elf
+
+chmod +x a.elf
diff --git a/ntranslate b/ntranslate
index 94b7e7f1..4cc8e955 100755
--- a/ntranslate
+++ b/ntranslate
@@ -2,13 +2,17 @@
 # Translate SubX using the self-hosted translator.
 #
 # Possible knobs:
-#   Whether to run a phase directly or emulated.
+#   Whether to run a phase natively or in emulated mode.
 #     This script is for running natively.
 #   Whether to stop after a phase.
-#     Just always run all phases, but print out phases so it's clear where an error happens.
-#   Whether to trace a phase. Whether to always trace or rerun with tracing enabled after an error.
-#     Leave tracing to other scripts. We save intermediate files so it's easy to rerun a single phase afterwards.
-#   Whether to run a phase with debug information. (Need to juggle multiple sets of debug files.)
+#     This script assumes inputs are already working so doesn't easily show
+#     which phase an error happens in.
+#   Whether to trace a phase. Whether to always trace or rerun with tracing
+#   enabled after an error.
+#     Leave tracing to other scripts. We save intermediate files so it's easy
+#     to rerun a single phase afterwards.
+#   Whether to run a phase with debug information. (Need to juggle multiple
+#   sets of debug files.)
 #     Again, that's for subsequent scripts.
 
 set -e
diff --git a/translate b/translate
index 4546e15d..caa3b375 100755
--- a/translate
+++ b/translate
@@ -2,13 +2,17 @@
 # Translate SubX using the self-hosted translator.
 #
 # Possible knobs:
-#   Whether to run a phase directly or emulated.
+#   Whether to run a phase natively or in emulated mode.
 #     Just always emulate for now since we debug on non-Linux.
 #   Whether to stop after a phase.
-#     Just always run all phases, but print out phases so it's clear where an error happens.
-#   Whether to trace a phase. Whether to always trace or rerun with tracing enabled after an error.
-#     Leave tracing to other scripts. We save intermediate files so it's easy to rerun a single phase afterwards.
-#   Whether to run a phase with debug information. (Need to juggle multiple sets of debug files.)
+#     Just always run all phases, but print out phases so it's clear where an
+#     error happens.
+#   Whether to trace a phase. Whether to always trace or rerun with tracing
+#   enabled after an error.
+#     Leave tracing to other scripts. We save intermediate files so it's easy
+#     to rerun a single phase afterwards.
+#   Whether to run a phase with debug information. (Need to juggle multiple
+#   sets of debug files.)
 #     Again, that's for subsequent scripts.
 
 set -e