Update STATUS.md + MEMORY.md: single-shot inference verified

STATUS.md

@@ -23,11 +23,14 @@
 
 ## Stage E Checklist (from ROADMAP/NEXT_PRIORITIES_PART_2)
 
-- [x] **E1:** Wire `LayerCacheHandle` → `gather_compressed_kv`, `gather_all_compressed_kv`, `gather_swa_kv`, `num_query_heads`, `head_dim` ✅
-- [x] **E2:** End-to-end smoke test through one full layer ✅ (SWA + CSA + HCA)
-- [x] **E3:** Top-level `model/dsv4.py` ✅
-- [x] **E4:** Delete `torch.cuda.synchronize()` from fast path ✅
-- [x] **E5:** Fold batch loop into kernel grid ✅ (single launch for batched decode)
+- [x] **E1:** Wire `LayerCacheHandle` → gather methods ✅
+- [x] **E2:** E2E smoke tests (SWA + CSA + HCA) ✅
+- [x] **E3:** DSV4Model class ✅
+- [x] **E4:** Removed `torch.cuda.synchronize` ✅
+- [x] **E5:** Batch loop folded into kernel grid ✅
+- [x] **Single-shot inference:** Full 61-layer pipeline runs on B200 ✅
+  - FMHA kernel verified: hd=512, 128 query heads, all layers correct
+  - Garbage output expected without mHC/MoE/KV-cache (architecture gaps, not kernel)
 - [ ] **E6:** FP4 output fusion for FMHA → wo_a
 - [ ] **E7:** Lightning indexer FP4 tensor-core scoring
 - [ ] **E8:** Multi-CTA grid for prefill