Function gdn_decode_wy4_woa 

pub fn gdn_decode_wy4_woa(
    gpu: &dyn GpuBackend,
    kernel: KernelHandle,
    h_table: DevicePtr,
    query: DevicePtr,
    key: DevicePtr,
    value: DevicePtr,
    gate: DevicePtr,
    beta: DevicePtr,
    output: DevicePtr,
    stash: DevicePtr,
    batch_size: u32,
    num_k_heads: u32,
    num_v_heads: u32,
    k_dim: u32,
    v_dim: u32,
    qk_stride: u32,
    v_stride: u32,
    gb_stride: u32,
    stash_seq_floats: u32,
    engaged_flag: DevicePtr,
    stream: u64,
) -> Result<()>

Write-on-accept K=4 verify (gated_delta_rule_wy4_woa): byte-identical output to gdn_decode_wy4, writes NO state, stashes the per-row update terms for gdn_wy4_fold. Table form only. Grid (num_v_heads, batch), 128.