diff --git a/10.01_base/2_src/pru1/pru1_main_unibus.c b/10.01_base/2_src/pru1/pru1_main_unibus.c
index 09d5fab..43ac8ab 100644
--- a/10.01_base/2_src/pru1/pru1_main_unibus.c
+++ b/10.01_base/2_src/pru1/pru1_main_unibus.c
@@ -159,9 +159,11 @@ void main(void) {
 					// "set" is inverting!
 					cpu_grant_mask = buslatches_getbyte(0) & PRIORITY_ARBITRATION_BIT_MASK; // read GRANT IN
 					// forward un-requested GRANT IN to GRANT OUT for other cards on neighbor slots
-					uint8_t grant_out_mask = cpu_grant_mask & ~sm_arb.device_request_mask ; 
-					buslatches_setbits(0, PRIORITY_ARBITRATION_BIT_MASK, ~grant_out_mask)
-				;
+					sm_arb.device_forwarded_grant_mask = cpu_grant_mask
+							& ~sm_arb.device_request_signalled_mask;
+					buslatches_setbits(0, PRIORITY_ARBITRATION_BIT_MASK, ~sm_arb.device_forwarded_grant_mask)
+					;
+					// "A device may not accept a grant (assert SACK) after it passes the grant"
 				}
 				// handle GRANT/SACK/BBSY for emulated devices
 				uint8_t granted_request = sm_device_arb_worker(cpu_grant_mask); // devices process GRANT
@@ -293,7 +295,7 @@ void main(void) {
 				mailbox.arm2pru_req = ARM2PRU_NONE; // ACK: done
 				break;
 			case ARM2PRU_BUSLATCH_SET: { // set a mux register
-                // and read back
+				// and read back
 				// don't feed "volatile" vars into buslatch_macros !!!
 				uint8_t reg_sel = mailbox.buslatch.addr & 7;
 				uint8_t bitmask = mailbox.buslatch.bitmask;
diff --git a/10.01_base/2_src/pru1/pru1_statemachine_arbitration.c b/10.01_base/2_src/pru1/pru1_statemachine_arbitration.c
index 3c55511..2c009d6 100644
--- a/10.01_base/2_src/pru1/pru1_statemachine_arbitration.c
+++ b/10.01_base/2_src/pru1/pru1_statemachine_arbitration.c
@@ -100,8 +100,11 @@ void sm_arb_reset() {
 	// cleanup: clear all REQUESTS and SACK
 	buslatches_setbits(1, PRIORITY_ARBITRATION_BIT_MASK | BIT(5), 0);
 	sm_arb.device_request_mask = 0;
+	sm_arb.device_forwarded_grant_mask = 0;
+	sm_arb.device_request_signalled_mask = 0;
+
 	sm_arb.bbsy_wait_grant_mask = 0;
-	sm_arb.cpu_request = 0 ;
+	sm_arb.cpu_request = 0;
 	sm_arb.arbitrator_grant_mask = 0;
 	timeout_cleanup(TIMEOUT_SACK);
 }
@@ -171,6 +174,8 @@ uint8_t sm_arb_worker_device(uint8_t grant_mask) {
 	// Always update UNIBUS BR/NPR lines, are ORed with requests from other devices.
 	buslatches_setbits(1, PRIORITY_ARBITRATION_BIT_MASK, sm_arb.device_request_mask)
 	;
+	// now relevant for GRANt forwarding
+	sm_arb.device_request_signalled_mask = sm_arb.device_request_mask;
 
 	// read GRANT IN lines from CPU (Arbitrator). 
 	// Only one nit ion cpu_grant_mask at a time may be active, else arbitrator malfunction.
@@ -179,7 +184,8 @@ uint8_t sm_arb_worker_device(uint8_t grant_mask) {
 	if (sm_arb.bbsy_wait_grant_mask == 0) {
 		// State 1: Wait For GRANT:
 		// process the requested grant for an open requests.
-		grant_mask &= sm_arb.device_request_mask;
+		// "A device may not accept a grant (assert SACK) after it passes the grant"
+		grant_mask &= (sm_arb.device_request_mask & ~sm_arb.device_forwarded_grant_mask);
 		if (grant_mask) {
 			// one of our requests was granted:
 			// set SACK
diff --git a/10.01_base/2_src/pru1/pru1_statemachine_arbitration.h b/10.01_base/2_src/pru1/pru1_statemachine_arbitration.h
index 5f0c927..3d74666 100644
--- a/10.01_base/2_src/pru1/pru1_statemachine_arbitration.h
+++ b/10.01_base/2_src/pru1/pru1_statemachine_arbitration.h
@@ -42,6 +42,12 @@ typedef struct {
 	// BR/NPR lines = set of _PRIORITY_ARBITRATION_BIT_*
 	uint8_t device_request_mask;
 
+	// device_request_mask when actually on BR/NR lines
+	uint8_t device_request_signalled_mask;
+
+	// forwarded to GRANT OUT, not accepted as response to device_request_signalled_mask 
+	uint8_t device_forwarded_grant_mask;
+
 	// sm_arb has 2 states: State 1 "Wait for GRANT" and State 2 "wait for BBSY"
 	// When arbitrator GRANts a request, we set SACK, GRAMT is cleared and we wait
 	// for BBSY clear. 
@@ -51,11 +57,11 @@ typedef struct {
 
 	/*** master ****/
 	// CPU is requesting memory access via PRU2ARM_DMA/mailbox.dma
-	uint8_t cpu_request ; 
-	
+	uint8_t cpu_request;
+
 	uint8_t arbitrator_grant_mask; // single GRANT line set by master
 
-	// uint8_t	dummy[3] ;
+	uint8_t dummy[2]; // make it dword-sized
 
 } statemachine_arbitration_t;