diff --git a/justfile b/justfile
index f3784be..48fada1 100644
--- a/justfile
+++ b/justfile
@@ -82,7 +82,7 @@ fuzz target pkg fuzztime="60s":
 
 # The cmd/link GOOBJ parity gate, opt-in: it costs minutes no pipeline can afford.
 link-parity:
-    systemd-run --user --scope -p MemoryMax={{memlimit}} -p MemorySwapMax=0 env GASM_LINK_PARITY=1 go test -count=1 -timeout 0 -run 'TestGOOBJLinkRegression' ./verify/
+    systemd-run --user --scope -p MemoryMax={{memlimit}} -p MemorySwapMax=0 env GASM_LINK_PARITY=1 go test -count=1 -timeout 0 -run 'TestGOOBJLink' ./verify/
 
 # Benchmarks. On an idle machine only, and deliberately unfenced.
 bench pkgs=packages:
diff --git a/verify/goobj_link_vector_test.go b/verify/goobj_link_vector_test.go
new file mode 100644
index 0000000..8063fb2
--- /dev/null
+++ b/verify/goobj_link_vector_test.go
@@ -0,0 +1,56 @@
+// Copyright (c) 2026 Petr Balvín <opensource@petrbalvin.org> (https://petrbalvin.org)
+// SPDX-License-Identifier: BSD-3-Clause
+
+package verify
+
+import (
+	"os"
+	"os/exec"
+	"testing"
+)
+
+// goobjVectorMainSrc is the Go side of the vector regression.  The kernel
+// computes a signed byte checksum through the RVV load/store and arithmetic
+// families; main derives the same value in Go and panics on any divergence
+// before printing the deterministic line the baseline and the gasm-linked
+// binaries must agree on.
+const goobjVectorMainSrc = `package main
+
+func checksum(buf []byte, seed int64) int64
+
+func main() {
+	buf := []byte("gasm link parity vector kernel!")
+	total := int64(0)
+	for _, b := range buf {
+		total += int64(int8(b))
+	}
+	want := total*0x0101010101010101 ^ 0x5eed
+	if got := checksum(buf, 0x5eed); got != want {
+		panic("checksum")
+	}
+	println("ok", checksum(buf, 0x5eed))
+}
+`
+
+// TestGOOBJLinkRVVKernel proves the riscv64 vector families merged into the
+// encoder corpus are proven end to end: the kernel assembles by gasm into a
+// GOOBJ object, substitutes byte-wise into a real go build archive, relinks
+// with cmd/link and the linked binary computes the same checksum as the
+// toolchain-built one under qemu-user.  The reduction reads the live vl, so
+// the checksum holds for any hardware VLEN.
+func TestGOOBJLinkRVVKernel(t *testing.T) {
+	if testing.Short() {
+		t.Skip("builds the gasm binary and links Go programs")
+	}
+	if os.Getenv("GASM_LINK_PARITY") == "" {
+		t.Skip("deliberate verification: set GASM_LINK_PARITY=1 (just link-parity)")
+	}
+	goBin, err := exec.LookPath("go")
+	if err != nil {
+		t.Skip("no Go toolchain available")
+	}
+	gasmBin := buildLinkParityGasm(t, goBin)
+	t.Run("riscv64", func(t *testing.T) {
+		goobjLinkKernel(t, goBin, gasmBin, "riscv64", "linkvec_riscv64.s", goobjVectorMainSrc)
+	})
+}
diff --git a/verify/testdata/linkvec_riscv64.s b/verify/testdata/linkvec_riscv64.s
new file mode 100644
index 0000000..5079d2d
--- /dev/null
+++ b/verify/testdata/linkvec_riscv64.s
@@ -0,0 +1,65 @@
+// Copyright (c) 2026 Petr Balvín <opensource@petrbalvin.org> (https://petrbalvin.org)
+// SPDX-License-Identifier: BSD-3-Clause
+
+#include "textflag.h"
+
+// The vector link-regression kernel for riscv64: the RVV load/store and
+// arithmetic families computing a checksum end to end.  bytetotal
+// strip-mines a byte buffer with VSETVLI, loads each chunk with VLE8V,
+// widens it to E64 lanes with VSEXTVF8 and reduces the chunk with
+// VREDSUMVS, so the result never depends on the hardware VLEN.  checksum
+// passes the slice to bytetotal through the ABI0 stack slots and mixes the
+// reduction with the seed through the intra-file mix call.  mix multiplies
+// by the odd golden-ratio constant and XORs the seed, the same arithmetic
+// the Go side of the regression mirrors bit for bit.
+
+// func bytetotal(buf []byte) int64
+TEXT ·bytetotal(SB), NOSPLIT, $0-32
+	MOV       buf+0(FP), X10
+	MOV       buf_len+8(FP), X12
+	MOV       $0, X13
+	BEQZ      X12, empty
+
+loop:
+	VSETVLI   X12, E64, M1, TA, MA, X11
+	VLE8V     (X10), V16
+	VSEXTVF8  V16, V9
+	VMVVI     $0, V10
+	VREDSUMVS V10, V9, V11
+	VMVXS     V11, X14
+	ADD       X14, X13, X13
+	ADD       X11, X10, X10
+	SUB       X11, X12, X12
+	BNE       X12, X0, loop
+
+empty:
+	MOV       X13, ret+24(FP)
+	RET
+
+// func mix(x, seed int64) int64
+TEXT ·mix(SB), NOSPLIT, $0-24
+	MOV x+0(FP), X10
+	MOV seed+8(FP), X11
+	MOV $0x0101010101010101, X12
+	MUL X12, X10, X10
+	XOR X11, X10, X10
+	MOV X10, ret+16(FP)
+	RET
+
+// func checksum(buf []byte, seed int64) int64
+TEXT ·checksum(SB), NOSPLIT, $40-40
+	MOV  buf+0(FP), X10
+	MOV  buf+8(FP), X11
+	MOV  buf+16(FP), X12
+	MOV  X10, 8(SP)
+	MOV  X11, 16(SP)
+	MOV  X12, 24(SP)
+	CALL ·bytetotal(SB)
+	MOV  32(SP), X10
+	MOV  seed+24(FP), X11
+	MOV  X10, 8(SP)
+	MOV  X11, 16(SP)
+	CALL ·mix(SB)
+	MOV  24(SP), X10
+	MOV  X10, ret+32(FP)
+	RET