diff --git a/src/ov_MAIN_012/ov_MAIN_012.c b/src/ov_MAIN_012/ov_MAIN_012.c
index 57a8aa6ff..5b5a5bb73 100644
--- a/src/ov_MAIN_012/ov_MAIN_012.c
+++ b/src/ov_MAIN_012/ov_MAIN_012.c
@@ -10495,7 +10495,29 @@ DEFINE_func_801685EC()  /* dedup: shared engine-core @0x801685ec (src/shared) */
 DEFINE_func_80168640()  /* dedup: shared engine-core @0x80168640 (src/shared) */
 
 
-INCLUDE_ASM("asm/ov_MAIN_012/nonmatchings/ov_MAIN_012", func_80168664);
+
+
+extern Mat32 D_800AE620;
+extern void RotMatrixX(int, void*);
+extern void func_80048EAC(void *a0, void *a1);
+extern s32 func_801670E4(s32 a0, s32 a1, s32 a2, s32 a3);
+
+void func_80168664(void *arg0) {
+    s32 iVar3;
+    void *blk;
+    iVar3 = *(s32 *)((u8 *)arg0 + 0x34);
+    *(s32 *)((u8 *)arg0 + 0x1c) = 0x20;
+    *(s16 *)((u8 *)arg0 + 0x10) = 0x80;
+    *(s16 *)((u8 *)arg0 + 0x12) = 0;
+    *(s32 *)((u8 *)arg0 + 0x30) = 0x1800;
+    *(Blk20 *)((u8 *)arg0 + 0x38) = (*(Blk20 *)&D_800AE620);
+    blk = (u8 *)arg0 + 0x38;
+    ((s32 (*)(s32, void *))RotMatrixX)(0x400, blk);
+    func_80048EAC((void *)(*(s32 *)((u8 *)iVar3 + 0x20) + 0x34), blk);
+    func_801670E4((s32)arg0, -6, -0x44, -0x18);
+    *(s16 *)((u8 *)arg0 + 2) = *(s16 *)((u8 *)arg0 + 2) + 1;
+}
+
 
 
 
@@ -10677,7 +10699,54 @@ INCLUDE_ASM("asm/ov_MAIN_012/nonmatchings/ov_MAIN_012", func_80168D94);
 DEFINE_func_80168EC4()  /* dedup: shared engine-core @0x80168ec4 (src/shared) */
 
 
-INCLUDE_ASM("asm/ov_MAIN_012/nonmatchings/ov_MAIN_012", func_80168F40);
+
+
+typedef struct {
+    u16 f00;
+    u16 f02;
+    u16 f04;
+    u16 pad06;
+    u16 f08;
+    u16 f0A;
+    u16 f0C;
+    u16 pad0E;
+    u8  f10;
+    u8  f11;
+    u8  f12;
+    u8  pad13;
+    u8  f14;
+    u8  f15;
+    u8  f16;
+    u8  pad17;
+    u32 f18;
+} S1_80168F40;
+
+extern s32 func_80016A5C();
+extern Mat32 D_800AE620;
+
+void func_80168F40(void *a0)
+{
+    S1_80168F40 s1;
+    Blk20 s2;
+    s32 pad[2];
+
+    s1.f00 = *(u16 *)((s32)a0 + 0x10);
+    s1.f02 = *(u16 *)((s32)a0 + 0x14);
+    s1.f04 = *(u16 *)((s32)a0 + 0x18);
+    s1.f08 = *(u16 *)((s32)a0 + 0x6);
+    s1.f0A = *(u16 *)((s32)a0 + 0xA);
+    s1.f0C = *(u16 *)((s32)a0 + 0xE);
+    s1.f10 = 8;
+    s1.f14 = 0xC0;
+    s1.f12 = 0;
+    s1.f11 = 0;
+    s1.f16 = 0x80;
+    s1.f15 = 0x80;
+    s2 = (*(Blk20 *)&D_800AE620);
+    s1.f18 = 0x50000000;
+    func_80016A5C(&s1, &s2);
+}
+
 
 
 
diff --git a/src/ov_SC02_037/ov_SC02_037.c b/src/ov_SC02_037/ov_SC02_037.c
index d748786fd..54c6b092f 100644
--- a/src/ov_SC02_037/ov_SC02_037.c
+++ b/src/ov_SC02_037/ov_SC02_037.c
@@ -461,7 +461,47 @@ DEFINE_func_8012B70C()  /* dedup: shared engine-core @0x8012b70c (src/shared) */
 DEFINE_func_8012B744()  /* dedup: shared engine-core @0x8012b744 (src/shared) */
 
 
-INCLUDE_ASM("asm/ov_SC02_037/nonmatchings/ov_SC02_037", func_8012B77C);
+extern s32 ratan2(s32 a0, s32 a1);
+extern s32 func_80047948(s32 a0);
+extern s32 func_8004787C(s32 a0);
+
+s32 func_8012B77C(s32 out, s32 a1, s32 a2)
+{
+    register u32 packed __asm__("$20");
+    s32 dz;
+    s32 dx;
+    s32 dy;
+    register s32 t __asm__("$16");
+    s32 sn;
+    s32 cs;
+    s32 ang;
+
+    packed &= 0xFFFF;
+
+    dz = *(s16 *)((s32)a2 + 0xA) - *(s16 *)((s32)a1 + 0xA);
+    dx = *(s16 *)((s32)a2 + 0x2) - *(s16 *)((s32)a1 + 0x2);
+    dy = *(s16 *)((s32)a2 + 0x6) - *(s16 *)((s32)a1 + 0x6);
+
+    ang = ratan2(-dz, dx);
+    t = ang - 0x400;
+    t &= 0xFFF;
+    packed |= (u32)(t << 16);
+
+    sn = func_80047948(t);
+    cs = func_8004787C(t);
+
+    dz = (dz * sn + dx * cs) >> 12;
+    ang = ratan2(dy, -dz);
+    packed &= 0xFFFF0000;
+    packed |= (u32)(ang & 0xFFFF);
+
+    {
+        u32 addr;
+        __asm__("addu %0,%1,$zero" : "=r"(addr) : "r"(out));
+        *(u32 *)addr = packed;
+    }
+}
+
 
 DEFINE_func_8012B864()  /* dedup: shared engine-core @0x8012b864 (src/shared) */
 
@@ -10829,7 +10869,29 @@ DEFINE_func_801685EC()  /* dedup: shared engine-core @0x801685ec (src/shared) */
 DEFINE_func_80168640()  /* dedup: shared engine-core @0x80168640 (src/shared) */
 
 
-INCLUDE_ASM("asm/ov_SC02_037/nonmatchings/ov_SC02_037", func_80168664);
+
+
+extern void func_80048EAC(void *a0, void *a1);
+extern s32  func_801670E4(s32 arg0, s32 arg1, s32 arg2, s32 arg3);
+extern void RotMatrixX(int, void*);
+extern Mat32 D_800AE620;
+
+void func_80168664(s32 param_1)
+{
+    s32 node;
+
+    node = *(s32 *)(param_1 + 0x34);
+    *(s32 *)(param_1 + 0x1c) = 0x20;
+    *(u16 *)(param_1 + 0x10) = 0x80;
+    *(u16 *)(param_1 + 0x12) = 0;
+    *(s32 *)(param_1 + 0x30) = 0x1800;
+    *(Blk20 *)(param_1 + 0x38) = (*(Blk20 *)&D_800AE620);
+    ((s32 (*)(s32, void *))RotMatrixX)(0x400, (void *)(param_1 + 0x38));
+    func_80048EAC((void *)(*(s32 *)(node + 0x20) + 0x34), (void *)(param_1 + 0x38));
+    func_801670E4(param_1, -6, -0x44, -0x18);
+    *(s16 *)(param_1 + 2) = *(s16 *)(param_1 + 2) + 1;
+}
+
 
 
 
diff --git a/src/ov_SC03_107/ov_SC03_107.c b/src/ov_SC03_107/ov_SC03_107.c
index fb48fce09..b9a2e2947 100644
--- a/src/ov_SC03_107/ov_SC03_107.c
+++ b/src/ov_SC03_107/ov_SC03_107.c
@@ -10491,7 +10491,28 @@ DEFINE_func_801685EC()  /* dedup: shared engine-core @0x801685ec (src/shared) */
 DEFINE_func_80168640()  /* dedup: shared engine-core @0x80168640 (src/shared) */
 
 
-INCLUDE_ASM("asm/ov_SC03_107/nonmatchings/ov_SC03_107", func_80168664);
+
+
+extern void RotMatrixX(int, void*);
+extern void func_80048EAC(void *a0, void *a1);
+extern s32 func_801670E4(s32 arg0, s32 arg1, s32 arg2, s32 arg3);
+extern Mat32 D_800AE620;
+
+void func_80168664(s32 a0) {
+    s32 p;
+
+    p = *(s32 *)(a0 + 0x34);
+    *(s32 *)(a0 + 0x1C) = 0x20;
+    *(u16 *)(a0 + 0x10) = 0x80;
+    *(u16 *)(a0 + 0x12) = 0;
+    *(s32 *)(a0 + 0x30) = 0x1800;
+    *(Blk20 *)(a0 + 0x38) = (*(Blk20 *)&D_800AE620);
+    ((s32 (*)(s32, void *))RotMatrixX)(0x400, (void *)(a0 + 0x38));
+    func_80048EAC((void *)(*(s32 *)(p + 0x20) + 0x34), (void *)(a0 + 0x38));
+    func_801670E4(a0, -6, -0x44, -0x18);
+    *(u16 *)(a0 + 2) = *(u16 *)(a0 + 2) + 1;
+}
+
 
 
 
@@ -10673,7 +10694,54 @@ INCLUDE_ASM("asm/ov_SC03_107/nonmatchings/ov_SC03_107", func_80168D94);
 DEFINE_func_80168EC4()  /* dedup: shared engine-core @0x80168ec4 (src/shared) */
 
 
-INCLUDE_ASM("asm/ov_SC03_107/nonmatchings/ov_SC03_107", func_80168F40);
+
+
+typedef struct {
+    u16 f00;
+    u16 f02;
+    u16 f04;
+    u16 pad06;
+    u16 f08;
+    u16 f0A;
+    u16 f0C;
+    u16 pad0E;
+    u8  f10;
+    u8  f11;
+    u8  f12;
+    u8  pad13;
+    u8  f14;
+    u8  f15;
+    u8  f16;
+    u8  pad17;
+    u32 f18;
+} S1_80168F40;
+
+extern s32 func_80016A5C();
+extern Mat32 D_800AE620;
+
+void func_80168F40(void *a0)
+{
+    S1_80168F40 s1;
+    Blk20 s2;
+    s32 pad[2];
+
+    s1.f00 = *(u16 *)((s32)a0 + 0x10);
+    s1.f02 = *(u16 *)((s32)a0 + 0x14);
+    s1.f04 = *(u16 *)((s32)a0 + 0x18);
+    s1.f08 = *(u16 *)((s32)a0 + 0x6);
+    s1.f0A = *(u16 *)((s32)a0 + 0xA);
+    s1.f0C = *(u16 *)((s32)a0 + 0xE);
+    s1.f10 = 8;
+    s1.f14 = 0xC0;
+    s1.f12 = 0;
+    s1.f11 = 0;
+    s1.f16 = 0x80;
+    s1.f15 = 0x80;
+    s2 = (*(Blk20 *)&D_800AE620);
+    s1.f18 = 0x50000000;
+    func_80016A5C(&s1, &s2);
+}
+
 
 
 
diff --git a/tools/dedup_extend.py b/tools/dedup_extend.py
index c95df38e2..971eed9d9 100644
--- a/tools/dedup_extend.py
+++ b/tools/dedup_extend.py
@@ -188,16 +188,37 @@ def write_drafts(binary, plan):
     return d
 
 
-def gate(binary, drafts_dir, chunk):
-    """The EXISTING whole-binary byte-gate is the sole arbiter (G3/P9). Returns the banked fn set."""
+def gate(binary, drafts_dir, chunk, ladder=True):
+    """The whole-binary byte-gate is the sole arbiter (G3/P9). Returns the banked fn set.
+
+    LADDER (S46): this called `harvest_verify` VERBATIM, which is the bare gate with no recovery.
+    Measured cost of that: 0 of 142 extendable groups banked, and the classified reasons were 118
+    PLUMBING / 21 CC1-FAIL / 3 DIFF — i.e. ~1 in 50 was a real byte divergence and everything else
+    was a declaration conflict in the TARGET TU (`conflicting types for memcpy / ApplyMatrixSV /
+    D_800AE620 / func_8014C568`). `gate_stage` wraps the same gate in exactly the reconcile ladder
+    those need (canon_resident_calls -> cast_call_sites -> sig_unify -> harvest_verify), and it
+    carries the §156 reconcile LEDGER so a failed candidate cannot leave a fleet-shared edit behind.
+    Measured on the same failure class in the S46 wave residue: 6 of 19 PLUMBING recovered (~32%).
+
+    GATE_NO_ARITY=1 is set for the child: gate_stage's arity pre-pass writes the fleet-shared
+    engine_core.h BEFORE the gate, and on a failing draft that edit can survive — the F1 defect that
+    broke 141 of 213 binaries in S45 and made every later gate report `near`. The ladder's other
+    rungs are draft-local and safe.
+    """
     good = open(os.path.join(REPO, f"config/check.{binary}.sha")).read().split()[0]
     vout = f".run/extend_verified.{binary}.txt"
     fout = f".run/extend_failed.{binary}.txt"
-    cmd = [PY, "tools/harvest_verify.py", "--binary", binary,
-           "--out", f"build/{binary}/{binary}", "--good-sha", good,
-           "--drafts", os.path.relpath(drafts_dir, REPO), "--chunk", str(chunk),
-           "--verified-out", vout, "--failed-out", fout]
-    subprocess.run(cmd, cwd=REPO, timeout=7200)
+    if ladder:
+        cmd = [PY, "tools/gate_stage.py", "--binary", binary,
+               "--drafts", os.path.relpath(drafts_dir, REPO), "--no-propagate",
+               "--verified-out", vout, "--failed-out", fout]
+    else:
+        cmd = [PY, "tools/harvest_verify.py", "--binary", binary,
+               "--out", f"build/{binary}/{binary}", "--good-sha", good,
+               "--drafts", os.path.relpath(drafts_dir, REPO), "--chunk", str(chunk),
+               "--verified-out", vout, "--failed-out", fout]
+    env = dict(os.environ, GATE_NO_ARITY="1")
+    subprocess.run(cmd, cwd=REPO, timeout=7200, env=env)
     p = os.path.join(REPO, vout)
     return set(open(p).read().split()) if os.path.exists(p) else set()