diff --git a/Makefile b/Makefile
index 85469f9b..5901f87c 100644
--- a/Makefile
+++ b/Makefile
@@ -263,6 +263,7 @@ TESTS	= $(BIN)/test_smoke $(BIN)/test_lex $(BIN)/test_parse $(BIN)/test_check \
 	  $(BIN)/test_globstructret_run \
 	  $(BIN)/test_globstructwiden_run \
 	  $(BIN)/test_globtagwiden_run \
+	  $(BIN)/test_globfloatstructarg_run \
 	  $(BIN)/test_arr_ptr_global \
 	  $(BIN)/test_def_arr_infer_len \
 	  $(BIN)/test_def_arr_len \
@@ -804,6 +805,16 @@ $(BIN)/test_globtagwiden_run: test/wcc/989_globtagwiden_run.c \
 		$(LIB)/libwwrt.a | $(BIN)
 	$(CC) $(CFLAGS) -o $@ $<
 
+# 989_globfloatstructarg_run (F8-c9, #31, the EXCEPTION): a module-global
+# float-bearing struct passed by value must drain its float eightbytes into
+# the SSE arg regs. align-UP — structfloatclass keyed off the global tnode.
+$(BIN)/test_globfloatstructarg_run: test/wcc/989_globfloatstructarg_run.c \
+		$(BIN)/ww $(BIN)/ww_ww \
+		$(BIN)/w6c $(BIN)/w6a $(BIN)/w6l \
+		$(BIN)/w6c_ww $(BIN)/w6a_ww $(BIN)/w6l_ww \
+		$(LIB)/libwwrt.a | $(BIN)
+	$(CC) $(CFLAGS) -o $@ $<
+
 $(BIN)/test_let_global: test/wcc/630_let_global.c $(BIN)/ww $(BIN)/w6c \
 		$(BIN)/w6a $(BIN)/w6l $(LIB)/libwwrt.a | $(BIN)
 	$(CC) $(CFLAGS) -o $@ $<
diff --git a/selfhost/cmd/w6c/main.combined.ww b/selfhost/cmd/w6c/main.combined.ww
index 0801f1d4..8279700c 100644
--- a/selfhost/cmd/w6c/main.combined.ww
+++ b/selfhost/cmd/w6c/main.combined.ww
@@ -30607,7 +30607,17 @@ fn cgcall(c: *cgen, n: *node) void = {
 			let stfc: i32 = 0;
 			if (a.kind == nkind.N_IDENT) {
 				let lc: *local = localfindnode(c, a.str);
-				if (lc != nil) { stfc = structfloatclass(c, lc.tnode); };
+				if (lc != nil) { stfc = structfloatclass(c, lc.tnode); }
+				else {
+					// #31 EXCEPTION (align-UP): a module-global float-bearing
+					// struct arg — lc is nil, so the local-only stfc stayed 0 and
+					// the drain fell to all-GP (X0 never loaded). Key the float-
+					// class off the global's declared tnode (letvartnode), the same
+					// SysV classification the local path uses. cstage keys
+					// structfloatclass off the operand TYPE, not a local slot.
+					let gtn: *node = letvartnode(c, a.str);
+					if (gtn != nil) { stfc = structfloatclass(c, gtn); };
+				};
 			};
 			if (stfc != 0) {
 				// #165: float-bearing struct arg — drain by SysV
diff --git a/selfhost/cmd/wcc/cgenexpr.ww b/selfhost/cmd/wcc/cgenexpr.ww
index 3ba94c28..8a8dc8f2 100644
--- a/selfhost/cmd/wcc/cgenexpr.ww
+++ b/selfhost/cmd/wcc/cgenexpr.ww
@@ -7750,7 +7750,17 @@ fn cgcall(c: *cgen, n: *node) void = {
 			let stfc: i32 = 0;
 			if (a.kind == nkind.N_IDENT) {
 				let lc: *local = localfindnode(c, a.str);
-				if (lc != nil) { stfc = structfloatclass(c, lc.tnode); };
+				if (lc != nil) { stfc = structfloatclass(c, lc.tnode); }
+				else {
+					// #31 EXCEPTION (align-UP): a module-global float-bearing
+					// struct arg — lc is nil, so the local-only stfc stayed 0 and
+					// the drain fell to all-GP (X0 never loaded). Key the float-
+					// class off the global's declared tnode (letvartnode), the same
+					// SysV classification the local path uses. cstage keys
+					// structfloatclass off the operand TYPE, not a local slot.
+					let gtn: *node = letvartnode(c, a.str);
+					if (gtn != nil) { stfc = structfloatclass(c, gtn); };
+				};
 			};
 			if (stfc != 0) {
 				// #165: float-bearing struct arg — drain by SysV
diff --git a/selfhost/cmd/wwdump/main.combined.ww b/selfhost/cmd/wwdump/main.combined.ww
index 2a0a16d4..98bf1f78 100644
--- a/selfhost/cmd/wwdump/main.combined.ww
+++ b/selfhost/cmd/wwdump/main.combined.ww
@@ -30607,7 +30607,17 @@ fn cgcall(c: *cgen, n: *node) void = {
 			let stfc: i32 = 0;
 			if (a.kind == nkind.N_IDENT) {
 				let lc: *local = localfindnode(c, a.str);
-				if (lc != nil) { stfc = structfloatclass(c, lc.tnode); };
+				if (lc != nil) { stfc = structfloatclass(c, lc.tnode); }
+				else {
+					// #31 EXCEPTION (align-UP): a module-global float-bearing
+					// struct arg — lc is nil, so the local-only stfc stayed 0 and
+					// the drain fell to all-GP (X0 never loaded). Key the float-
+					// class off the global's declared tnode (letvartnode), the same
+					// SysV classification the local path uses. cstage keys
+					// structfloatclass off the operand TYPE, not a local slot.
+					let gtn: *node = letvartnode(c, a.str);
+					if (gtn != nil) { stfc = structfloatclass(c, gtn); };
+				};
 			};
 			if (stfc != 0) {
 				// #165: float-bearing struct arg — drain by SysV
diff --git a/test/wcc/989_globfloatstructarg_run.c b/test/wcc/989_globfloatstructarg_run.c
new file mode 100644
index 00000000..6bd02de4
--- /dev/null
+++ b/test/wcc/989_globfloatstructarg_run.c
@@ -0,0 +1,161 @@
+/*
+ * 989_globfloatstructarg_run — F8-c9 (report-item #31, the EXCEPTION):
+ * passing a module-GLOBAL float-bearing struct by value must drain its float
+ * eightbytes into the SSE arg regs (X0..), not all-GP.
+ *
+ * THE BUG (cat-A silent miscompile, align-UP): cgcall's per-arg drain decides
+ * the SysV eightbyte class via structfloatclass, but read it from the local
+ * slot's tnode ONLY (`if (lc != nil) stfc = structfloatclass(lc.tnode)`). A
+ * module-global struct arg (lc==nil) kept stfc=0, so the float eightbytes
+ * drained as integers (POPQ into DI/SI) and X0 was never loaded — the callee
+ * read its f64 fields from the wrong registers. cstage classifies off the
+ * operand TYPE, so it loaded X0 and ran correct — the cat-A divergence.
+ *
+ * THIS IS THE EXCEPTION among the F8 members: the bug is a float-CLASS
+ * decision keyed to a local-only slot, NOT an address-routing fallback — an
+ * F7-flavoured stamp/type-keying fix. THE FIX: for a global ident, key
+ * structfloatclass off the global's declared tnode (letvartnode), the same
+ * classification the local path uses. align ww UP; the .s is byte-identical.
+ *
+ * Rows (build+run on cstage `ww` and wwstage `ww_ww`; rule-10 — agree+hit;
+ * each take() returns 0 iff every field round-trips through the call):
+ *   row             | shape                            | want
+ *   ----------------+----------------------------------+-----
+ *   float_first     | gp:{f:f64,i:i64}={1.5,9}; take    | 0 [#31: f→X0]
+ *   int_first       | gp:{i:i64,f:f64}={9,2.5}; take    | 0 [int eb→GP, f→X0]
+ *   both_float      | gp:{a:f64,b:f64}={1.5,2.5}; take  | 0 [two SSE eightbytes]
+ */
+#include <stdio.h>
+#include <stdlib.h>
+#include <string.h>
+#include <unistd.h>
+#include <sys/stat.h>
+#include <sys/wait.h>
+
+static int
+runwait(const char *cmd)
+{
+	int rc = system(cmd);
+	if (rc == -1) return -1;
+	if (WIFEXITED(rc)) return WEXITSTATUS(rc);
+	return -1;
+}
+
+struct row {
+	const char *label;
+	const char *src;
+	int want_exit;
+};
+
+static const struct row rows[] = {
+	{ "float_first",
+	  "package main;\n"
+	  "type pair = struct { f: f64, i: i64 };\n"
+	  "let gp: pair = pair { f = 1.5, i = 9 };\n"
+	  "fn take(p: pair) int = { if (p.f == 1.5 && p.i == 9) { return 0; }; return 2; };\n"
+	  "export fn main() int = { return take(gp); };\n",
+	  0 },
+
+	{ "int_first",
+	  "package main;\n"
+	  "type pair = struct { i: i64, f: f64 };\n"
+	  "let gp: pair = pair { i = 9, f = 2.5 };\n"
+	  "fn take(p: pair) int = { if (p.i == 9 && p.f == 2.5) { return 0; }; return 2; };\n"
+	  "export fn main() int = { return take(gp); };\n",
+	  0 },
+
+	{ "both_float",
+	  "package main;\n"
+	  "type pp = struct { a: f64, b: f64 };\n"
+	  "let gp: pp = pp { a = 1.5, b = 2.5 };\n"
+	  "fn take(p: pp) int = { if (p.a == 1.5 && p.b == 2.5) { return 0; }; return 2; };\n"
+	  "export fn main() int = { return take(gp); };\n",
+	  0 },
+};
+
+/* run_build — build+run `src` via `driver`; returns the binary's exit
+ * code, or -1 on a build failure. */
+static int
+run_build(const char *driver, const struct row *r, int i)
+{
+	char src[64], tmpdir[64], cmd[1024];
+	snprintf(src, sizeof src, "/tmp/gfsa_%d_%d.ww", getpid(), i);
+	snprintf(tmpdir, sizeof tmpdir, "/tmp/gfsa_%d_d_%d", getpid(), i);
+
+	FILE *f = fopen(src, "wb");
+	if (!f) return -2;
+	fputs(r->src, f);
+	fclose(f);
+
+	mkdir(tmpdir, 0755);
+	snprintf(cmd, sizeof cmd, "cd %s && %s build %s 2>/dev/null",
+	    tmpdir, driver, src);
+	int brc = runwait(cmd);
+
+	const char *base = strrchr(src, '/');
+	base = base ? base + 1 : src;
+	char outbin[128];
+	snprintf(outbin, sizeof outbin, "%s/%s", tmpdir, base);
+	char *dot = strrchr(outbin, '.');
+	if (dot && strcmp(dot, ".ww") == 0) *dot = '\0';
+
+	int got = -1;
+	if (brc == 0) got = runwait(outbin);
+
+	unlink(src); unlink(outbin); rmdir(tmpdir);
+	return brc == 0 ? got : -1;
+}
+
+int
+main(void)
+{
+	const char *bin = getenv("BIN");
+	if (!bin) bin = "out/bin";
+	char absbin[1024];
+	if (bin[0] != '/') {
+		char cwd[1024];
+		if (getcwd(cwd, sizeof cwd) == NULL) return 1;
+		snprintf(absbin, sizeof absbin, "%s/%s", cwd, bin);
+		bin = absbin;
+	}
+
+	char cdrv[1024], wdrv[1024];
+	snprintf(cdrv, sizeof cdrv, "%s/ww", bin);
+	snprintf(wdrv, sizeof wdrv, "%s/ww_ww", bin);
+
+	struct { const char *name; const char *drv; int gated; }
+	drivers[] = {
+		{ "cstage",  cdrv, 0 },
+		{ "wwstage", wdrv, 1 },
+		{ NULL, NULL, 0 },
+	};
+
+	int n = (int)(sizeof rows / sizeof rows[0]);
+	int total = 0, fail = 0;
+
+	for (int d = 0; drivers[d].name; d++) {
+		if (drivers[d].gated && access(drivers[d].drv, X_OK) != 0) {
+			fprintf(stderr, "globfloatstructarg_run: skip %s (no %s)\n",
+			    drivers[d].name, drivers[d].drv);
+			continue;
+		}
+		for (int i = 0; i < n; i++) {
+			total++;
+			int got = run_build(drivers[d].drv, &rows[i], i);
+			if (got != rows[i].want_exit) {
+				fprintf(stderr, "globfloatstructarg_run[%s][%s]: "
+				    "exit=%d want=%d\n", drivers[d].name,
+				    rows[i].label, got, rows[i].want_exit);
+				fail++;
+			}
+		}
+	}
+
+	if (fail) {
+		fprintf(stderr, "globfloatstructarg_run: %d/%d fixtures failed\n",
+		    fail, total);
+		return 1;
+	}
+	printf("globfloatstructarg_run: %d/%d ok\n", total, total);
+	return 0;
+}