diff --git a/bin/checksum_delta_diff_test.py b/bin/checksum_delta_diff_test.py
index e86db98..6076262 100644
--- a/bin/checksum_delta_diff_test.py
+++ b/bin/checksum_delta_diff_test.py
@@ -13,16 +13,20 @@ def checksum_scapy(pkt):
     if len(pkt) % 2 == 1:
         pkt += b"\0"
 
+    # P4: ABOVE not needed (always even)
+
     # array: create an array of 16 bit values from the input
     # and then sum it up -> this might be sligthly/much higher
     # than 16 bit (additions!)
     s = sum(array.array("H", pkt))
+    # P4: summing manually into 32bit
 
     # add the (right shift 16) and the 16 right bits
     # basically: assuming 32 bit (?): add the two 16 bit "words"
     # together
     # This might still exceed 16 bit!
     s = (s >> 16) + (s & 0xffff)
+    # P4:
 
     # right shift 16 -> zero least significant bits,
     # and add the upper bits to it
diff --git a/doc/plan.org b/doc/plan.org
index 216df0a..59f10f1 100644
--- a/doc/plan.org
+++ b/doc/plan.org
@@ -5245,6 +5245,16 @@ likely this is the cause for the main error:
 
 **** go through all steps again and try to understand why it (silently) fails later
 *** 2019-07-13: fix overflow error
+*** 2019-07-15: fix off-by-one-sometimes
+    - scapy shift code / return does not work
+**** scapy code results
+Using     meta.v6sum = (bit<16>) ((((tmp>>8) & 0xff)|tmp<<8) & 0xffff)
+;
+
+we get on the wire:
+
+sum 0x324a (incorrect -> 0x5429),
+
 ** The NetPFGA saga
 Problems encountered:
 - The logfile for a compile run is 10k+ lines
diff --git a/p4src/actions_delta_checksum.p4 b/p4src/actions_delta_checksum.p4
index 42291ff..008ae29 100644
--- a/p4src/actions_delta_checksum.p4
+++ b/p4src/actions_delta_checksum.p4
@@ -13,13 +13,12 @@ action v4sum() {
     tmp = tmp + (bit<32>) hdr.ipv4.totalLen -20;                // 16 bit
     tmp = tmp + (bit<32>) hdr.ipv4.protocol;                    // 8 bit
 
-    carryover = tmp >> 16;            /* maximum is 6*(2**16)  >> 16 == 6 */
-    tmp = (tmp & 0xffff) + carryover; /* Now tmp contains at maximum 65541 */
-    carryover = tmp >> 16;            /* Now carryover contains at maximum 1 */
-    tmp = tmp + carryover;            /* No overrun possible anymore */
-
     /* filtering code copied from scapy */
-    meta.v4sum = (bit<16>) ((((tmp>>8) & 0xff)|tmp<<8) & 0xffff) ;
+    tmp = (tmp >> 16) + (tmp & 0xffff);
+    tmp = tmp + (tmp >> 16);
+    tmp = ~tmp;
+    meta.v4sum = (bit<16>) ((((tmp>>8) & 0xff)|tmp<<8) & 0xffff);
+
 }
 
 action v6sum() {
@@ -47,15 +46,14 @@ action v6sum() {
     tmp = tmp + (bit<32>) hdr.ipv6.payload_length;              // 16 bit
     tmp = tmp + (bit<32>) hdr.ipv6.next_header;                 // 8 bit
 
-    carryover = tmp >> 16;            /* maximum is 18*(2**16)  >> 16 == 18 */
-    tmp = (tmp & 0xffff) + carryover; /* Now tmp contains at maximum 65554*/
-    carryover = tmp >> 16;            /* Now carryover contains at maximum 1 */
-    tmp = tmp + carryover;            /* No overrun possible anymore */
-
     /* filtering code copied from scapy */
+    tmp = (tmp >> 16) + (tmp & 0xffff);
+    tmp = tmp + (tmp >> 16);
+    tmp = ~tmp;
     meta.v6sum = (bit<16>) ((((tmp>>8) & 0xff)|tmp<<8) & 0xffff) ;
 }
 
+
 action delta_prepare()
 {
     v4sum();
@@ -70,13 +68,13 @@ action delta_prepare()
 action delta_udp_from_v4_to_v6()
 {
     delta_prepare();
-    hdr.udp.checksum = hdr.udp.checksum + ~meta.headerdiff;
+    hdr.udp.checksum = hdr.udp.checksum + meta.headerdiff;
 }
 
 action delta_tcp_from_v4_to_v6()
 {
     delta_prepare();
-    hdr.tcp.checksum = hdr.tcp.checksum + ~meta.headerdiff;
+    hdr.tcp.checksum = hdr.tcp.checksum + meta.headerdiff;
 }
 
 action delta_ipv4_from_v6_to_v4()