diff --git a/docs/research/SIDON_SOFA_COLORING.md b/docs/research/SIDON_SOFA_COLORING.md
index ccda0993..a4cb7d4d 100644
--- a/docs/research/SIDON_SOFA_COLORING.md
+++ b/docs/research/SIDON_SOFA_COLORING.md
@@ -665,46 +665,6 @@ and the chromatic number of Γ_γ (combinatorics).
     Question: Does a longer braid word → higher chromatic number?
     Is there a braid invariant that bounds χ from below?
 
-### Direction F: Hardware Coloring Filter (OISC CMYK)
-
-The SilverSight infrastructure includes a **Blitter6502OISC** — a SUBLEQ
-one-instruction CPU on Tang Nano 9K FPGA with Q16_16 fixed-point LUT
-arithmetic, connected via **Tailscale mesh** to Provider-Nixos (EPYC
-bare metal). A 4-gate CMYK filter designed on this substrate maps
-directly onto the conflict graph chromatic number computation.
-
-**The mapping:**
-
-| CMYK Gate | SUBLEQ operation | What it computes |
-|-----------|-----------------|------------------|
-| C (Cyan) | SUBLEQ on x-coords | ‖γ(t_a)pᵢ − γ(t_b)pⱼ‖_x in Q16_16 |
-| M (Magenta) | SUBLEQ on y-coords | ‖γ(t_a)pᵢ − γ(t_b)pⱼ‖_y in Q16_16 |
-| Y (Yellow) | SUBLEQ on distance² | d² = dx² + dy² − 1 (conflict iff ≤ 0) |
-| K (Key) | SUBLEQ on color ID | assigns color class (1..χ) to t_a |
-
-Each gate is one SUBLEQ subtract-and-branch: subtract, branch if result
-≤ 0. This is exactly one unit-distance check in Q16_16 fixed-point:
-
-    Y-gate: SUBLEQ(Q16_16_MUL(dx,dx) + Q16_16_MUL(dy,dy), Q16_16_ONE)
-    → branch if ≤ 0 means d² ≤ 1 → CONFLICT → assign different color
-
-**4 gates = 4 colors.** A CMYK filter with χ = 4 tests the boundary
-below de Grey's lower bound (5 ≤ χ(ℝ²)). If the 4-gate filter rejects
-all colorings of Γ_γ^T, this gives a hardware proof that χ(Γ_γ^T) ≥ 5.
-
-**Hardware parallelism.** The FPGA evaluates all n(n−1)/2 pairwise
-distances simultaneously (one SUBLEQ per pair per time-step pair),
-giving O(1) latency per conflict graph evaluation at fixed (n, m).
-
-**Compute pipeline:**
-- FPGA (Tang Nano 9K): conflict graph construction + greedy coloring
-- Provider-Nixos (EPYC): shape optimization over A*(n, χ) landscape
-- Tailscale mesh: distribute (n, χ) parameter sweep across nodes
-
-    Question: Can the 4-gate CMYK filter on SUBLEQ hardware prove
-    χ(Γ_γ^T) ≥ 5 for Gerver's sofa at n = 13, m = 100?
-    Infrastructure: Blitter6502OISC + Q16_16 LUT + Tailscale mesh
-
 ---
 
 ## 8. Why This Problem Is Structurally Rich