diff --git a/docs/papers/asymmetric-kv-compression.md b/docs/papers/asymmetric-kv-compression.md
index 2e8ac24..dc5ee9c 100644
--- a/docs/papers/asymmetric-kv-compression.md
+++ b/docs/papers/asymmetric-kv-compression.md
@@ -265,6 +265,13 @@ These findings have been independently confirmed by multiple researchers:
 - APEX Mini (12.2 GB) + TurboQuant = 35B MoE at 8K context on a 16GB consumer GPU
 - Independently confirms boundary layer sensitivity for weights: "first and last 5 transformer layers are 10x more sensitive than the middle," consistent with our Boundary V finding for KV cache
 
+**vLLM TurboQuant PR** — [vllm-project/vllm#38479](https://github.com/vllm-project/vllm/pull/38479) (2026-04-03):
+- TurboQuant KV cache compression being integrated into vLLM (major production inference framework)
+- PR author (@vibhavagarwal5) explicitly cites turboquant_plus for boundary layer implementation and asymmetric KV cache
+- @Alberto-Codes independently diagnosed the 0% gsm8k quality failure as the symmetric K/V compression problem documented in our asymmetric paper
+- @varjoranta validated turbo4-resurrection findings via ablation on A100 with Qwen3-8B, confirming our paper's predictions about compression quality tradeoffs
+- Community testing across H100, A100, A4000, DGX Spark (GB10) confirms our findings that symmetric turbo3 on sensitive models produces garbage, asymmetric fixes it
+
 ---
 
 ## 8. Limitations
diff --git a/docs/papers/layer-aware-v-compression.md b/docs/papers/layer-aware-v-compression.md
index c7737a1..c0ed083 100644
--- a/docs/papers/layer-aware-v-compression.md
+++ b/docs/papers/layer-aware-v-compression.md
@@ -371,6 +371,7 @@ The boundary layer sensitivity finding has been independently confirmed:
 - **@HyperionMS2040** — RTX 3090 (2026-03-30): Qwen3.5-4B asymmetric data shows clean quality ladder with no catastrophic failure when K precision is maintained, consistent with boundary layer sensitivity.
 - **@Corianas_** — Independent validation of Boundary V on NanoGPT (X collaborator, 2026-03-29).
 - **@mudler (Ettore Di Giacinto)** — [APEX](https://github.com/mudler/apex-quant) MoE weight quantization, [LocalAI](https://github.com/mudler/LocalAI) (44.7k stars) (2026-04-01): Independently discovered that "first and last 5 transformer layers are 10x more sensitive than the middle" for weight quantization on Qwen3.5-35B-A3B across 25+ experiments. Same boundary sensitivity principle as Boundary V, applied to weights instead of KV cache. APEX assigns higher precision to edge layers, lower to middle layers. Directly validates the boundary layer hypothesis from the weight quantization direction.
+- **vLLM TurboQuant PR** — [vllm-project/vllm#38479](https://github.com/vllm-project/vllm/pull/38479) (2026-04-03): PR author (@vibhavagarwal5) implementing TQ_boundary_layers in vLLM, explicitly "inspired from https://github.com/TheTom/turboquant_plus". Boundary layer protection being adopted by a major production inference framework.
 
 ---