diff --git a/lib/prelude.ml b/lib/prelude.ml
index 1e4e72a..988914f 100644
--- a/lib/prelude.ml
+++ b/lib/prelude.ml
@@ -637,11 +637,22 @@ let source = {flan|
 ;; to_upper both allocate a new string (core/strings/conversion.odin), which
 ;; is not available here; the obvious substitute — lowering a [u8] in place —
 ;; is a trap, and it is worth saying why rather than shipping it. A string
-;; literal is emitted `private unnamed_addr constant` (emit.ml), so it lives
-;; in read-only memory, and (bytes "Hello") is a [u8] pointing straight at it.
-;; An in-place lower-ascii! would type check against that slice and segfault
-;; on the store. Given a byte function, a caller that really does own its
-;; buffer writes the two-line loop itself and can see what it is writing to.
+;; literal is emitted `private unnamed_addr constant` (emit.ml), so (bytes
+;; "Hello") is a [u8] pointing straight into read-only memory. An in-place
+;; lower-ascii! type checks against that slice, and what happens next depends
+;; on the optimiser — which is the worst of the available answers. Measured,
+;; with (set (at (bytes "Hi") 0) \h):
+;;
+;;   -O0   the store is emitted against the constant and the program takes
+;;         SIGSEGV.
+;;   -O2   LLVM deletes the store as undefined behaviour and the program
+;;         carries on and prints "Hi".
+;;
+;; So the same source either dies or silently does nothing depending on a
+;; flag, and the -O2 half is the quiet-wrongness class this file keeps
+;; refusing elsewhere. Given a byte function instead, a caller that really
+;; does own its buffer writes the two-line loop itself over storage it can
+;; see the declaration of.
 ;;
 ;; ASCII only, and only the 26 letters: case outside ASCII is not a byte
 ;; operation at all — it is per-code-point, it is not length-preserving (ß
diff --git a/test/programs/utf8.flan b/test/programs/utf8.flan
index db5b269..a45d2d2 100644
--- a/test/programs/utf8.flan
+++ b/test/programs/utf8.flan
@@ -190,6 +190,19 @@
   (show-opt (encode-rune! (slice scratch 0 1) 0x41))       ; 1, exactly enough
   (newline)
 
+  ;; "Nothing is written when it refuses" is a claim about the buffer, not
+  ;; about the return value, and the None cases above do not test it: an
+  ;; encoder that lays down the lead byte and only then notices the buffer is
+  ;; short returns None exactly as this one does, and every line above still
+  ;; passes. So put a known byte in scratch, ask for an encoding that must be
+  ;; refused, and read the byte back.
+  (show-i32 (round-trip 0x41))                             ; 65, scratch[0] = A
+  (show-opt (encode-rune! (slice scratch 0 2) 0x65e5))     ; -1, needs 3 bytes
+  (show-i32 (i32 (at scratch 0)))                          ; 65 still
+  (show-opt (encode-rune! (slice scratch 0 4) 0xd800))     ; -1, surrogate
+  (show-i32 (i32 (at scratch 0)))                          ; 65 still
+  (newline)
+
   ;; Splitting. n separators give n+1 fields, always: an interior empty field
   ;; survives, a leading and a trailing one do too, and an input with no
   ;; separator at all is one field rather than none. The empty input is the
diff --git a/test/test_acceptance.ml b/test/test_acceptance.ml
index 63e5ea8..e47dc10 100644
--- a/test/test_acceptance.ml
+++ b/test/test_acceptance.ml
@@ -913,6 +913,7 @@ ERR@7 unexpected token: not the kind the caller was reading
        -1 1 1 2 2 3 3 -1 -1 3 3 4 4 -1 \n\
        0 65 127 128 2047 2048 65535 65536 1114111 \n\
        -1 -1 -1 -1 -1 1 \n\
+       65 -1 65 -1 65 \n\
        [a][b][c] [a][][b] [abc] [] [][] [][a] [a][] \n\
        60\n\
        97 122 97 64 91 65 90 65 96 123 53 \n\