diff --git a/Platform/Apple/tools/PackPartitions/src/org/badvision/A2PackPartitions.groovy b/Platform/Apple/tools/PackPartitions/src/org/badvision/A2PackPartitions.groovy
index 3279d904..473ca996 100644
--- a/Platform/Apple/tools/PackPartitions/src/org/badvision/A2PackPartitions.groovy
+++ b/Platform/Apple/tools/PackPartitions/src/org/badvision/A2PackPartitions.groovy
@@ -1657,9 +1657,10 @@ class A2PackPartitions
         def compressedLen = compressedData.length
         assert compressedLen > 0
 
-        // As a check, verify that decompression works with only a 3-byte underlap
+        // As a check, verify that decompression works with only a 5-byte underlap
+        // Note: used to think 3 bytes was sufficient, then gen_flags.b blew that out of the water.
         if (debugCompression && (uncompressedLen - compressedLen) > 0) {
-            def underlap = 3
+            def underlap = 5
             def checkData = new byte[uncompressedLen+underlap]
             def initialOffset = uncompressedLen - compressedLen + underlap
             System.arraycopy(compressedData, 0, checkData, initialOffset, compressedLen)
diff --git a/Platform/Apple/tools/PackPartitions/src/org/badvision/Lx47Algorithm.java b/Platform/Apple/tools/PackPartitions/src/org/badvision/Lx47Algorithm.java
index 9ec79426..2f743a10 100644
--- a/Platform/Apple/tools/PackPartitions/src/org/badvision/Lx47Algorithm.java
+++ b/Platform/Apple/tools/PackPartitions/src/org/badvision/Lx47Algorithm.java
@@ -178,6 +178,7 @@ public class Lx47Algorithm
         }
 
         void writeByte(int value) {
+            //System.out.format("byte: pos=%d val=$%x\n", outPos, (byte)(value & 0xFF));
             buf[outPos++] = (byte)(value & 0xFF);
             bitPos += 8;
         }
@@ -189,6 +190,7 @@ public class Lx47Algorithm
                 writeByte(0);
                 bitPos -= 8;
             }
+            //System.out.format("bit: pos=%d mask=$%x bit=%d\n", bitIndex, mask, value > 0 ? 1 : 0);
             if (value > 0)
                 buf[bitIndex] |= mask;
             mask >>= 1;
@@ -344,24 +346,30 @@ public class Lx47Algorithm
         public byte[] buf;
         public int inPos;
         private int indexByte;
+        private int inStart;
+        private int indexPos;
         private int mask;
 
         Lx47Reader(byte[] inBuf, int inStart) {
+            this.inStart = inStart;
             buf = inBuf;
             mask = 0;
             inPos = inStart;
         }
 
         int readByte() {
+            //System.out.format("byte: pos=%d val=$%x\n", inPos - inStart, buf[inPos] & 0xFF);
             return buf[inPos++] & 0xFF;
         }
 
         int readBit() {
             if (mask == 0) {
                 mask = 128;
+                indexPos = inPos - inStart;
                 indexByte = readByte();
             }
             int ret = ((indexByte & mask) != 0) ? 1 : 0;
+            //System.out.format("bit: pos=%d mask=$%x bit=%d\n", indexPos, mask, ret);
             mask >>= 1;
             return ret;
         }
@@ -400,7 +408,7 @@ public class Lx47Algorithm
         String expect = debugs.removeFirst();
         assert expect.equals(toCheck) : 
             String.format("Expecting '%s', got '%s'", expect, toCheck);
-        //System.out.format("OK [%d]: %s\n", debugs.size(), expect);
+        System.out.format("OK [%d]: %s\n", debugs.size(), expect);
     }
 
     public void decompress(byte[] input_data, int inStart, byte[] output_data, int outStart, int outLen)
diff --git a/Platform/Apple/virtual/src/core/mem.s b/Platform/Apple/virtual/src/core/mem.s
index 214eb6b4..ffc0574a 100644
--- a/Platform/Apple/virtual/src/core/mem.s
+++ b/Platform/Apple/virtual/src/core/mem.s
@@ -29,6 +29,12 @@ MAX_SEGS	= 96
 
 DEBUG		= 0
 
+; We overlap the compressed and uncompressed as much as possible, e.g.:
+;   DDDDDDDDDDDDDD
+;        SSSSSSSSSsssss  ; sssss is the 5-byte 'underlap'
+; Note: used to think this was 3, then found a case (gen_flags.b) requiring 5.
+UNDERLAP	= 5
+
 ; Zero page temporary variables.
 ; Don't move these - they overlap in clever ways with ProRWTS shadows (see below)
 tmp		= $2	; len 2
@@ -2124,12 +2130,12 @@ disk_queueLoad: !zone
 	lda resNum
 	cmp #1
 	bne +
-	; Take $BFFD - size. Why $BFFD and not $C000? Because decomp temporarily overwrites 3-byte "unnderlap" after.
-	lda #$FD
+	; Conservative end-of-mem, because decomp temporarily overwrites 5-byte "underlap" after.
+	lda #<($C000-UNDERLAP)
 	sec
 	sbc reqLen
 	sta targetAddr
-	lda #$BF
+	lda #>($C000-UNDERLAP)
 	sbc reqLen+1
 	sta targetAddr+1
 +	jsr shared_alloc	; reserve memory for this resource (main or aux as appropriate)
@@ -2304,7 +2310,7 @@ disk_finishLoad: !zone
 	; Calculate end of uncompressed data, and start of compressed data.
 	; We overlap the compressed and uncompressed as much as possible, e.g.:
 	;   DDDDDDDDDDDDDD
-	;        SSSSSSSSSsss  ; sss is the 3-byte 'underlap'
+	;        SSSSSSSSSsssss  ; sssss is the 5-byte 'underlap'
 	sta pSrc
 	sty pSrc+1
 	clc
@@ -2316,7 +2322,7 @@ disk_finishLoad: !zone
 	sta pEnd+1
 	tay
 	txa
-	adc #3			; this is the max "underlap" required for decompressing overlapped buffers
+	adc #UNDERLAP		; this is the max "underlap" required for decompressing overlapped buffers
 	bcc +
 	iny
 +	sec
@@ -2325,8 +2331,8 @@ disk_finishLoad: !zone
 	tya
 	sbc reqLen+1
 	sta pDst+1
-	; save the 3 byte underlap bytes because we're going to read over them
-	ldy #2
+	; save the 5 underlap bytes because we're going to read over them
+	ldy #UNDERLAP-1
 	ldx isAuxCmd
 	sta clrAuxRd,x		; from aux mem if appropriate
 -	lda (pEnd),y
@@ -2349,7 +2355,7 @@ disk_finishLoad: !zone
 -	pla
 	sta (pEnd),y
 	iny
-	cpy #3
+	cpy #UNDERLAP
 	bne -
 	sta clrAuxRd
 	sta clrAuxWr