diff --git a/main.c b/main.c
index 39a7b67..e5c6b0c 100644
--- a/main.c
+++ b/main.c
@@ -5,7 +5,7 @@
 #include "src/vec_match.h"
 
 #define DATA_LEN 16
-#define REF_LEN 8
+#define REF_LEN 10
 
 static float reference[DATA_LEN] = {
 	//0, 10, 20, 30, 40, 50, 40, 30, 20, 10, 0
@@ -14,7 +14,7 @@ static float reference[DATA_LEN] = {
 
 static float data[DATA_LEN] = {
 	//0, 10, 20, 30, 40, 50, 50, 35, 15, 15, 0
-	0, 15.7, 0, 0, 0.1, 0.2, 0.1, 10, 24.242, 0, 0, 2, 0.2, 0.4, 0.5, 0
+	0, 15.7, 0, 0, 0.1, 0.2, 0.1, 10, 24, 0, 2, 2, 0.2, 0.4, 0.5, 0
 };
 
 static float ref_p[REF_LEN];
@@ -108,7 +108,7 @@ int main(void)
 	printf("Error rate: ENV %.2f, ABS %.2f\n", env_e, abs_e);
 
 	float thr;
-	int ref_pack_len = vec_pack_fit(ref_p, REF_LEN, reference, DATA_LEN, &thr);
+	int ref_pack_len = vec_pack_auto(ref_p, REF_LEN, reference, DATA_LEN, &thr);
 
 	printf("Reference packed with zero threshold %.1f to %d items.\n", thr, ref_pack_len);
 	printf("REF packed:  ");
diff --git a/main.d b/main.d
index eadfba7..266d428 100644
--- a/main.d
+++ b/main.d
@@ -24,4 +24,4 @@ main.out: src/vec_match.c /usr/include/stdc-predef.h \
  /usr/include/bits/huge_valf.h /usr/include/bits/huge_vall.h \
  /usr/include/bits/inf.h /usr/include/bits/nan.h \
  /usr/include/bits/mathdef.h /usr/include/bits/mathcalls.h \
- src/vec_match.h
+ /usr/include/string.h /usr/include/xlocale.h src/vec_match.h
diff --git a/src/vec_match.c b/src/vec_match.c
index 7b1dfcf..0b969f8 100644
--- a/src/vec_match.c
+++ b/src/vec_match.c
@@ -4,6 +4,7 @@
 #include <stdio.h>
 #include <float.h>
 #include <math.h>
+#include <string.h>
 
 #include "vec_match.h"
 
@@ -202,23 +203,43 @@ uint32_t vec_pack(float *result, uint32_t result_capacity,
 
 
 uint32_t vec_pack_auto(float *result, uint32_t result_capacity,
-				  const float *data, uint32_t data_length, float *threshold_p)
+					   const float *data, uint32_t data_length, float *threshold_p)
 {
-	float thr = 0;
-	uint32_t ref_pack_len;
+	// lossless case, all fits
+	if (data_length <= result_capacity) {
+		*threshold_p = 0;
+		memcpy(result, data, data_length * sizeof(float));
+		return data_length;
+	}
 
-	// TODO use smarter algorithm
 
-	while(true) {
+	float thr = 0.1;
+	uint32_t ref_pack_len;
+
+	while (true) {
 		ref_pack_len = vec_pack(result, result_capacity, data, data_length, thr);
-		printf("try %f -> %d\n", thr, ref_pack_len);//FIXME remove
+
+		//printf("try %f -> %d\n", thr, ref_pack_len);
 
 		if (ref_pack_len <= result_capacity) {
 			if (threshold_p != NULL) *threshold_p = thr;
-			return result_capacity;
+			return ref_pack_len;
+		} else {
+			float r = (result_capacity / (float)ref_pack_len);
+
+			// experimental values, adjust to best fit your use case
+			if (r < 0.5) {
+				thr += 0.8;
+			} else if (r < 0.6) {
+				thr += 0.5;
+			} else if (r < 0.75) {
+				thr += 0.4;
+			} else if (r < 0.90) {
+				thr += 0.25;
+			} else {
+				thr += 0.1;
+			}
 		}
-
-		thr += 0.1f;
 	}
 }
 
diff --git a/src/vec_match.h b/src/vec_match.h
index 965ce14..26d80c6 100644
--- a/src/vec_match.h
+++ b/src/vec_match.h
@@ -114,5 +114,5 @@ uint32_t vec_unpack(float *result, uint32_t result_capacity,
  * @param threshold_p field to store the used threshold, can be NULL
  * @return real result size
  */
-uint32_t vec_pack_fit(float *result, uint32_t result_capacity,
+uint32_t vec_pack_auto(float *result, uint32_t result_capacity,
 					  const float *data, uint32_t data_length, float *threshold_p);