The x86 survey compares against LLVM built at -O0, the level the hand-written backend corresponds to

TODO.org

@@ -1085,13 +1085,6 @@ Every program in the corpus that compiles, has a =main= and terminates agrees wi
 the LLVM build down to stderr. =docs/BUILT.md=, "The hand-written x86 backend, and
 the four measurements behind it", is the account.
 
-** NEXT Nothing pins the LLVM side at -O0 when the two backends are compared
-Decided 2026-09-25: the x86 parity survey compares against LLVM at -O0 only; -O2 is never its target. The acceptance suite's paired -O2/-O0 rows stay, being the check for undefined behaviour in emitted IR, which is a different question.
-The survey builds both sides at the default =-O2=, so a construct LLVM folds is
-compared as a constant rather than as a lowering. That is how the float =%= gap
-survived. Two things would close it: an =-O0= pass of the sweep, and something
-that walks the two backends' primitive match arms mechanically. Neither is queued.
-
 ** DONE Reading (uninit) before writing it is undefined behaviour
 CLOSED: [2026-09-25]
 Reading an =(uninit)= value before writing it is undefined behaviour, and the backends may differ on it. An exhausted match stays =ud2= on x86.

test/dune

@@ -161,7 +161,8 @@
 ; second backend still lowers the language: it refuses by name rather than
 ; miscompiling, so when another lane adds a primitive the backend says so
 ; loudly, and nothing was listening. Two such refusals sat in the tree for a
-; month. Now they fail a build somebody can run.
+; month. Now they fail a build somebody can run. The LLVM side is built at
+; -O0, the level this backend corresponds to.
 ;
 ;     dune build --root . @x86
 ;

test/survey-x86.sh

@@ -20,6 +20,11 @@
 # compared a checked build against an unchecked one would say nothing about
 # bounds.flan, which is the one program the two backends disagreed about.
 #
+# The LLVM side is built at -O0, never at the default -O2. This backend has no
+# optimiser, so -O0 is its counterpart: at -O2 LLVM folds a constant
+# expression before it is lowered, and the sweep then compares a lowering
+# against a constant. That is how the float % gap stayed hidden.
+#
 # Six outcomes, and the third is the progress meter:
 #
 #   MATCH     built both ways, same stdout, same stderr, same exit status
@@ -91,14 +96,12 @@ forever="dev-loop dev-watch dev-chatty agent-auto"
 # in the epilogue that every exit already went through. The five are in the
 # sweep now and they are five of the MATCHes.
 
-# The one whose whole point is a fault, and which therefore cannot be compared
+# The one whose whole point is a fault. dev-segv stores through a null
-# at this sweep's optimisation level. dev-segv stores through a null pointer,
+# pointer, which is undefined, so neither side's answer is a lowering to
-# which is undefined: what LLVM at -O2 does with it is its own business, and
+# compare. test_dev.ml builds it in a dev session, where the fault is the thing
-# this backend has no optimiser and exits 139. That is not a lowering
+# asserted. It also calls agent/start, so it leaves a socket in /tmp on both
-# disagreement. test_dev.ml builds it in a dev session,
+# runs, and under SURVEY_FLAGS=--dev it parks in the break loop instead of
-# where the fault is the thing asserted. It also calls agent/start, so it
+# dying.
-# leaves a socket in /tmp on both runs, and under SURVEY_FLAGS=--dev it parks
-# in the break loop instead of dying.
 faults="dev-segv"
 
 TIMEOUT=${TIMEOUT:-20}
@@ -126,7 +129,7 @@ for src in "$corpus"/test/programs/*.flan; do
 
   # LLVM first. A program that does not compile at all, or has no main, is not
   # this backend's business -- the frontend refused it either way.
-  if ! "$flan" build "$src" "${extra[@]}" -o "$out/$name.llvm" \
+  if ! "$flan" build "$src" -O0 "${extra[@]}" -o "$out/$name.llvm" \
        >"$out/$name.llvm.err" 2>&1; then
     # A package the build tree does not have is not the frontend refusing
     # the program: it is a sweep that did not look at it, and it is counted