docs: mark Phase 2b rook 7th rank + connected passers as implemented

Benchmark results: score 80% (+15% vs 2a), mate rate 60%, rep rate 20%, 0 losses.
RunId: phase2b-rook7th-connpassers, commit 52c7aba.

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

Author: brainit-consulting

docs/ENGINE_ROADMAP.md

@@ -118,6 +118,20 @@ Goal: Reduce early threefolds without changing time caps.
     - Decision: accepted.
   - Remaining Phase 2 items (not yet implemented): rook 7th-rank bonus, connected passers, tapered evaluation.
 
+- Phase 2b - Rook on 7th rank + connected passed pawns (implemented, commit 52c7aba)
+  - Added: rook on 7th rank bonus (+25cp, +10cp extra if opponent king on back rank), connected passed pawn bonus (+15cp base + 5cp/rank advanced).
+  - Both terms apply to all difficulty levels.
+  - Added rookSquares tracking to EvalContext.
+  - 6 new unit tests (134 total passing).
+  - Self-play results (runId `phase2b-rook7th-connpassers`, Hard 1000ms vs Max 3000ms, 10 games, seed 9001, swap, fenSuite):
+    - W/D/L (Hard perspective): 6-4-0, score 80%.
+    - Mate rate: 60% (up from 50% in Phase 2a).
+    - Repetition rate: 20% (down from 30% in Phase 2a).
+    - Avg plies: 95.8 (shorter = faster conversion).
+    - Zero losses (improved from 1 in Phase 2a).
+    - Decision: accepted.
+  - Remaining Phase 2 items: tapered evaluation (Phase 2c).
+
 ## Phase 3 - Search and ordering improvements (moderate risk)
 
 - Change list (files)

plans/PHASE2B_ROOK7TH_CONNECTED_PASSERS.md

@@ -0,0 +1,45 @@
+# Phase 2b: Rook on 7th Rank + Connected Passed Pawns - Execution Record
+
+## Status: ACCEPTED
+
+## Change Summary
+- **Commit**: 52c7aba
+- **Files**: `src/ai/evaluate.ts`, `src/ai/__tests__/ai.test.ts`
+- **Roadmap Phase**: 2b
+
+## Implementation
+- Rook on 7th rank bonus: +25cp per rook on 7th rank
+- Rook on 7th king trapped bonus: +10cp extra when opponent king on back rank
+- Connected passed pawn bonus: +15cp base + 5cp per rank advanced beyond rank 3
+- Added rookSquares tracking to EvalContext (populated during piece iteration)
+- Both features apply to ALL difficulty levels
+
+## Test Results
+- 134 tests passed (128 existing + 6 new)
+- Build clean (no new TypeScript errors)
+
+## Benchmark Results (runId: phase2b-rook7th-connpassers)
+- Config: Hard 1000ms vs Max 3000ms, 10 games, seed 9001, swap, fenSuite
+- W/D/L (Hard perspective): 6-4-0
+- Score: 80%
+- Mate rate: 60% (Phase 2a baseline: 50%)
+- Repetition rate: 20% (Phase 2a baseline: 30%)
+- Avg plies: 95.8 (Phase 2a baseline: 103.3)
+- Losses: 0 (Phase 2a baseline: 1)
+- Early repetitions: 0
+
+## Comparison vs Phase 2a Baseline
+
+| Metric | Phase 2a | Phase 2b | Delta |
+|--------|----------|----------|-------|
+| Score % | 65% | 80% | +15% |
+| Mate rate | 50% | 60% | +10% |
+| Repetition rate | 30% | 20% | -10% |
+| Avg plies | 103.3 | 95.8 | -7.5 |
+| Losses | 1 | 0 | -1 |
+
+## Decision
+ACCEPTED - Score +15%, mate rate +10%, repetition rate -10%, zero losses. Shorter games suggest faster conversion. All criteria exceeded.
+
+## Remaining Phase 2 Items
+- Tapered evaluation (opening/middlegame/endgame weights) - Phase 2c