diff --git a/Makefile b/Makefile
index 91e6dff..9cc73ad 100644
--- a/Makefile
+++ b/Makefile
@@ -58,7 +58,7 @@ sysexits_bin: src/sysexits.c
 sysexits: sysexits_bin
 	bin/sysexits >include/sysexits.h
 
-programs: echo false id lowercase nonzero roll simexec sleep streq str true
+programs: echo false id lowercase nonzero pscat roll simexec sleep streq str true
 
 calculate: libstr sysexits src/calculate.c
 	$(CC) $(CFLAGS) -o src/calculate.o -c src/calculate.c
@@ -103,6 +103,12 @@ roll.o: lib/libio.h src/roll.c sysexits
 roll: libio sysexits roll.o
 	$(CC) $(CFLAGS) -o bin/roll build/libio.o build/roll.o
 
+roll_stdio.o: lib/libio.h src/roll.c sysexits
+	$(CC) $(CFLAGS) -c -DUSE_STDIO -o build/roll_stdio.o src/roll.c
+
+roll_stdio: libio sysexits roll_stdio.o
+	$(CC) $(CFLAGS) -o bin/roll_stdio build/libio.o build/roll_stdio.o
+
 rldecode.o: sysexits src/runlength.c
 	$(CC) $(CFLAGS) -Df=decode -c -o build/rldecode.o src/runlength.c
 
diff --git a/src/roll.c b/src/roll.c
index 2c068f8..f9008b2 100644
--- a/src/roll.c
+++ b/src/roll.c
@@ -19,8 +19,9 @@ int main(int argc, char *argv[]){
 		fprintf(stderr,
 			"Usage: %s [dice...]\n"
 			"\tDice should be formatted [rolls]d[sides], e.g. 1d3, 5d6...\n",
-			argv0 == NULL? program_name : argv0
+			argv0 == NULL ? program_name : argv0
 		);
+		return EX_USAGE;
 	}
 
 	srand(time(NULL));
@@ -32,7 +33,7 @@ int main(int argc, char *argv[]){
 		if(!isdigit(**argv)){
 error:			fprintf(stderr,
 				"%s: %s: Improperly formatted die (should be"
-				" [rolls]d[sides]).",
+				" [rolls]d[sides]).\n",
 				argv0, argvc
 			);
 			return EX_USAGE;
@@ -48,8 +49,26 @@ error:			fprintf(stderr,
 		if(**argv != '\0')
 			goto error;
 
+		/* On 2022-08-19 I experimented with using fprintf(3) instead
+		 * of libio's fdputd(3) to see if stdio was faster, thinking
+		 * the buffering would make a difference. stdio flushes its
+		 * buffer after every newline so in this particular code it
+		 * cannot improve performance unless fdputd(3)'s implementation
+		 * really sucked - no beans!
+		 * It also is an issue where every benchmark reports 0.00 real
+		 * / user / sys times for execution because this program does
+		 * very little - but if you fire up strace(3) you will find the
+		 * system calls (which are nowadays usually the slowest part of
+		 * a given small UNIX utility) are the same either way.
+		 * I'm keeping this little option here in case someone wants to
+		 * peer review; no harm in it. */
 		while(--r >= 0)
-			fdputd(1, rand() % s + 1);
+#ifdef USE_STDIO
+			fprintf(stdout, "%d\n",
+#else
+			fdputd(1,
+#endif
+				rand() % s + 1);
 	}
 
 	return EX_OK;