diff --git a/Tests/expected/diagAlignOutside.out b/Tests/expected/diagAlignOutside.out
new file mode 100644
index 0000000..857dcbb
--- /dev/null
+++ b/Tests/expected/diagAlignOutside.out
@@ -0,0 +1,4 @@
+Error: #Align outside the Program or Data Segment.
+ There is nothing there for it to move along.
+ File: testPrograms/diagnostics/alignOutside.asm at line 6.
+[exit 1]
diff --git a/Tests/expected/diagBareAlign.out b/Tests/expected/diagBareAlign.out
new file mode 100644
index 0000000..80c3d43
--- /dev/null
+++ b/Tests/expected/diagBareAlign.out
@@ -0,0 +1,3 @@
+Error: #Align without a number.
+ File: testPrograms/diagnostics/bareAlign.asm at line 11.
+[exit 1]
diff --git a/Tests/expected/diagBareInclude.out b/Tests/expected/diagBareInclude.out
new file mode 100644
index 0000000..fd88371
--- /dev/null
+++ b/Tests/expected/diagBareInclude.out
@@ -0,0 +1,3 @@
+Error: #Include without a file name.
+ File: testPrograms/diagnostics/bareInclude.asm at line 11.
+[exit 1]
diff --git a/Tests/expected/diagBareSWI.out b/Tests/expected/diagBareSWI.out
new file mode 100644
index 0000000..3024eac
--- /dev/null
+++ b/Tests/expected/diagBareSWI.out
@@ -0,0 +1,4 @@
+Error: SWI without a vector to go to.
+File: testPrograms/diagnostics/bareSWI.asm at line 9.
+Token: SWI
+[exit 1]
diff --git a/Tests/expected/diagDuplicateLabel.out b/Tests/expected/diagDuplicateLabel.out
new file mode 100644
index 0000000..c60a749
--- /dev/null
+++ b/Tests/expected/diagDuplicateLabel.out
@@ -0,0 +1,3 @@
+Error: Label "twice" is defined more than once.
+File: testPrograms/diagnostics/duplicateLabel.asm at line 13.
+[exit 1]
diff --git a/Tests/expected/diagDuplicateVector.out b/Tests/expected/diagDuplicateVector.out
new file mode 100644
index 0000000..ce94367
--- /dev/null
+++ b/Tests/expected/diagDuplicateVector.out
@@ -0,0 +1,3 @@
+Error: That vector already has a handler.
+File: testPrograms/diagnostics/duplicateVector.asm at line 20.
+[exit 1]
diff --git a/Tests/expected/diagPinnedRange.out b/Tests/expected/diagPinnedRange.out
new file mode 100644
index 0000000..03146a4
--- /dev/null
+++ b/Tests/expected/diagPinnedRange.out
@@ -0,0 +1,6 @@
+Error: 100 is not a vector number that can be given.
+ Numbers below 16 belong to the machine, and 64 and up are
+ handed out by the assembler. A vector two programs have to
+ agree about takes one from 16 to 63.
+File: testPrograms/diagnostics/pinnedVectorRange.asm at line 19.
+[exit 1]
diff --git a/Tests/expected/diagPinnedTaken.out b/Tests/expected/diagPinnedTaken.out
new file mode 100644
index 0000000..c33b43c
--- /dev/null
+++ b/Tests/expected/diagPinnedTaken.out
@@ -0,0 +1,4 @@
+Error: Another vector already has that number.
+File: testPrograms/diagnostics/pinnedVectorTaken.asm at line 28.
+Token: second
+[exit 1]
diff --git a/Tests/expected/diagStringInProgram.out b/Tests/expected/diagStringInProgram.out
new file mode 100644
index 0000000..96ef743
--- /dev/null
+++ b/Tests/expected/diagStringInProgram.out
@@ -0,0 +1,8 @@
+Error: A string cannot go in the Program Segment.
+ Strings live in Data Memory, which is the only memory an instruction
+ can read. A string in Program Memory could not be reached even by the
+ program holding it, except through the memory controller.
+ File: testPrograms/diagnostics/stringInProgram.asm at line 19.
+ The string: "this is not where this goes"
+ Move it below a #Data line.
+[exit 1]
diff --git a/Tests/expected/diagUnbasedSegment.out b/Tests/expected/diagUnbasedSegment.out
new file mode 100644
index 0000000..5e91ded
--- /dev/null
+++ b/Tests/expected/diagUnbasedSegment.out
@@ -0,0 +1,6 @@
+Error: The Program Segment is based at 0x2000, but the Data Segment
+ has 1 byte at 0x0000 and was never given a #Base.
+ Half a program loaded at zero lands on whatever is already there.
+ File: testPrograms/diagnostics/unbasedSegment.asm
+ Say "#Base 0x0000" in the Data Segment if that is what you meant.
+[exit 1]
diff --git a/Tests/expected/diagUnknownVector.out b/Tests/expected/diagUnknownVector.out
new file mode 100644
index 0000000..f17402c
--- /dev/null
+++ b/Tests/expected/diagUnknownVector.out
@@ -0,0 +1,4 @@
+Error: "neverDeclared" is not a vector.
+ Names used with SWI have to be given a handler in a #Vectors section.
+File: testPrograms/diagnostics/unknownVector.asm at line 9.
+[exit 1]
diff --git a/Tests/expected/lib-print.out b/Tests/expected/lib-print.out
new file mode 100644
index 0000000..dc88bc3
--- /dev/null
+++ b/Tests/expected/lib-print.out
@@ -0,0 +1,3 @@
+Error: Undefined label "start".
+File: Libraries/print.asm at line 8.
+[exit 1]
diff --git a/Tests/expected/printDecimalTest.out b/Tests/expected/printDecimalTest.out
new file mode 100644
index 0000000..29a4848
--- /dev/null
+++ b/Tests/expected/printDecimalTest.out
@@ -0,0 +1,3 @@
+Error: Undefined label "printByteDecimal".
+File: testPrograms/printDecimalTest.asm at line 9.
+[exit 1]
diff --git a/Tests/expected/printDigitTest.out b/Tests/expected/printDigitTest.out
new file mode 100644
index 0000000..d07301e
--- /dev/null
+++ b/Tests/expected/printDigitTest.out
@@ -0,0 +1,3 @@
+Error: Undefined label "printDecimalDigit".
+File: testPrograms/printDigitTest.asm at line 6.
+[exit 1]
diff --git a/Tests/expected/printHexTest.out b/Tests/expected/printHexTest.out
new file mode 100644
index 0000000..17d91fc
--- /dev/null
+++ b/Tests/expected/printHexTest.out
@@ -0,0 +1,3 @@
+Error: Undefined label "printByteHex".
+File: testPrograms/printHexTest.asm at line 8.
+[exit 1]
diff --git a/Tests/expected/shiftTest.out b/Tests/expected/shiftTest.out
new file mode 100644
index 0000000..46544ce
--- /dev/null
+++ b/Tests/expected/shiftTest.out
@@ -0,0 +1,3 @@
+Error: Undefined label "printDecimal".
+File: testPrograms/shiftTest.asm at line 12.
+[exit 1]
diff --git a/Tests/manifest b/Tests/manifest
index 3325838..e167c95 100644
--- a/Tests/manifest
+++ b/Tests/manifest
@@ -12,8 +12,10 @@
 # Modes:
 #   run       Assemble, execute, compare all output against Tests/expected/<name>.out
 #   assemble  Assemble only. For library files that have no entry point to run
-#   xfail     Assembly is expected to fail. Records a known breakage so that
-#             fixing one is noticed, and so an accidental new one is too
+#   xfail     Assembly is expected to fail, and the assembler's diagnostic is recorded
+#             and compared like any other output. Catches a known breakage that stops
+#             being detected, an accidental new one, and a message that changes
+#             without anybody meaning it to
 #
 # limit is a cycle count, for programs that never halt on their own. It is passed
 # to the emulator as --cycles, which bounds the run by cycles rather than by wall
diff --git a/Tests/run.sh b/Tests/run.sh
index a89dcfc..3b41637 100755
--- a/Tests/run.sh
+++ b/Tests/run.sh
@@ -141,13 +141,27 @@ assemble() {
     # test programs outside it include.
     local name="$1" src="$2"
     local bin="$BUILD/$name.bin"
-    if ( cd "$PROGRAMS" && "$ASSEMBLER" -I Libraries -I CosmOS/Source -o "$bin" "$src" ) >"$BUILD/.assemble.log" 2>&1; then
+    ( cd "$PROGRAMS" && "$ASSEMBLER" -I Libraries -I CosmOS/Source -o "$bin" "$src" ) \
+        >"$BUILD/.assemble.log" 2>&1
+    # WRITTEN TO A FILE RATHER THAN A VARIABLE, because this function is called inside a
+    # command substitution below and that runs it in a subshell, where anything it set
+    # would be thrown away. A file is the one channel that survives either call site.
+    echo $? > "$BUILD/.assemble.status"
+    if [ "$(cat "$BUILD/.assemble.status")" -eq 0 ]; then
         echo "$bin"
         return 0
     fi
     return 1
 }
 
+# Takes the colour out of a diagnostic, in place. The assembler paints its errors whether
+# or not anything is a terminal, so a recorded refusal would otherwise carry escape
+# sequences - and a recorded result full of them is unreadable in a diff, which is the one
+# moment it has to be read.
+uncolour() {
+    sed -i -E 's/\x1b\[[0-9;]*m//g' "$1"
+}
+
 while IFS='|' read -r name src mode stdin limit disk; do
     name="$(trim "$name")"
     [ -z "$name" ] && continue
@@ -163,10 +177,20 @@ while IFS='|' read -r name src mode stdin limit disk; do
         if assemble "$name" "$src" >/dev/null; then
             FAIL=$((FAIL + 1)); FAILED_NAMES+=("$name")
             report "FAIL" "$name" "expected assembly to fail, but it succeeded"
-        else
-            PASS=$((PASS + 1))
-            report "ok" "$name" "fails as recorded: $(head -1 "$BUILD/.assemble.log" | tr -d '\033' | sed 's/\[[0-9;]*m//g')"
+            continue
         fi
+        # WHAT IT SAID, not merely that it said something. Checking only that the assembler
+        # exited non zero passed for the intended error, for an unrelated error, for a
+        # changed message, and for the assembler falling over on its way to the point - all
+        # four look identical from outside. So the diagnostic is recorded and compared like
+        # any other output, which is what the Test Manual has always claimed happened here.
+        OUT="$BUILD/.out"
+        cp "$BUILD/.assemble.log" "$OUT"
+        uncolour "$OUT"
+        # The same last line every other recorded result carries, and for the same reason:
+        # which non zero status a refusal exits with is part of what it does.
+        printf '[exit %d]\n' "$(cat "$BUILD/.assemble.status")" >> "$OUT"
+        check "$name" "$OUT"
         continue
     fi