D-RMAP-DIR-1 round 2: unique board entry id; 65,000-user figures

.claude/board/entries/2026-10-20-d-rmap-dir-1-round-2-focus-geometry.md

@@ -1,4 +1,4 @@
-# D-RMAP-DIR-1 round 2: when 256-slot focus geometry beats compiled closure (2026-10-20)
+# D-RMAP-DIR-1-R2: round 2, when 256-slot focus geometry beats compiled closure (2026-10-20)
 
 **Status:** MEASURED, lab only; nothing promoted. **Basis:** `crates/dir-rmap-probe` (`tests/equivalence.rs` 11/11; `examples/bench.rs`, 4 runs at 8,000 users plus 1 at 65,000). Plan `d-rmap-dir-1-membership-closure-focus-v1.md` §9.
 

.claude/board/entries/README.md

@@ -39,7 +39,7 @@ exactly one of them, never both.
 | 2026-10-20 | `D-TGA-SLOT-AXIS` | thoughts are 64 slots, and the slot number is CE64's witness pointer | [2026-10-20-d-tga-slot-axis-witness-pointer.md](2026-10-20-d-tga-slot-axis-witness-pointer.md) |
 | 2026-10-20 | `D-RPR` |  | [2026-10-20-d-rpr-consolidation-1624.md](2026-10-20-d-rpr-consolidation-1624.md) |
 | 2026-10-20 | `D-RPR-1` |  | [2026-10-20-d-rpr-1-handoff-lab-demo.md](2026-10-20-d-rpr-1-handoff-lab-demo.md) |
-| 2026-10-20 | `D-RMAP-DIR-1` |  | [2026-10-20-d-rmap-dir-1-round-2-focus-geometry.md](2026-10-20-d-rmap-dir-1-round-2-focus-geometry.md) |
+| 2026-10-20 | `D-RMAP-DIR-1-R2` |  | [2026-10-20-d-rmap-dir-1-round-2-focus-geometry.md](2026-10-20-d-rmap-dir-1-round-2-focus-geometry.md) |
 | 2026-10-20 | `D-RMAP-DIR-1` |  | [2026-10-20-d-rmap-dir-1-membership-closure-focus.md](2026-10-20-d-rmap-dir-1-membership-closure-focus.md) |
 | 2026-10-20 | `D-RMAP-2` |  | [2026-10-20-d-rmap-2-tile-gemm-cell.md](2026-10-20-d-rmap-2-tile-gemm-cell.md) |
 | 2026-10-20 | `D-RMAP-2-AMX` |  | [2026-10-20-d-rmap-2-amx-tier.md](2026-10-20-d-rmap-2-amx-tier.md) |

.claude/plans/d-rmap-dir-1-membership-closure-focus-v1.md

@@ -245,6 +245,13 @@ Per evaluation, warm, ns, 8,000 users:
 - Composite membership is a cheap question. F's gain over B+ stays at 1.7–2.9× across every focus size, because the number of questions is at most 256.
 - Conjunctions multiply questions as the focus grows. F gains 2.4–5.4× up to 64 groups, then the question loop dominates, and past L1 the mask form loses (§9.3).
 
+**At 65,000 users** (one run; the same fixture shape, so more distinct pairs per focus) the pattern holds and sharpens:
+
+- 64 groups (956 conjunctions): F is 3.3× faster than B+ (391 vs 1,291 ns).
+- 256 groups (8,160 conjunctions, 255 KiB of masks): F's mask form is 2.1× slower than B+ (23.9 vs 11.1 µs); the slot-bit form is 1.2× faster (9.1 µs).
+- Composite membership: F is 1.8–2.8× faster at every focus size.
+- Sweep at 1,495 groups: B+ 218 ns, F 70 ns, F+users 57 ns. The F+users table here is 2 MiB, at the edge of L2, and still costs nothing extra per evaluation.
+
 ### 9.5 Constraints 3 and 4: preparation accounting and amortization
 
 **Preparation at 64 focus groups, 8,000 users** (W5a, composite fixture). Each phase alone; allocations are counted by a global counting allocator.