feat(llvm): Phase 131-13/14 - MIR JSON順序修正 & 2パスsnapshot解決

## Phase 131-13: MIR JSON 命令順序修正
- copy 遅延ロジック削除（~80行）
- MIR の def→use 順序をそのまま出力（SSOT）
- PHI 先頭集約のみ維持

## Phase 131-14: jump-only block 2パス snapshot 解決
- Pass A: jump-only block はメタ記録のみ
- Pass B: resolve_jump_only_snapshots() で CFG ベース解決
- path compression で連鎖を効率的に解決
- サイクル検出で Fail-Fast

## 結果
- ✅ STRICT モードでエラーなし
- ✅ bb7 が bb5 の snapshot を正しく継承
- ✅ ループが正しく動作（1, 2 出力確認）
- ⚠️ print/concat で segfault（別問題、次Phase）

🤖 Generated with [Claude Code](https://claude.com/claude-code)

Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>

docs/development/current/main/investigations/phase131-13-mir-json-instruction-order.md

@@ -0,0 +1,189 @@
+# Phase 131-13: MIR JSON Instruction Order Fix - COMPLETED ✅
+
+Status: **RESOLVED** (2025-12-14)
+Scope: Rust 側の `MIR → JSON (harness)` 出力で、ブロック内の命令順序が崩れて Python LLVM backend が Fail-Fast する問題。
+Related:
+- SSOT (LLVM棚卸し): `docs/development/current/main/phase131-3-llvm-lowering-inventory.md`
+- Case C: `docs/development/current/main/phase131-11-case-c-summary.md`
+
+## Problem Summary
+
+**Issue**: LLVM Stage-3 Case C (loop control) failed with undefined value error
+- Test: `apps/tests/llvm_stage3_loop_only.hako`
+- Symptom: `binop dst=19 lhs=17 rhs=18` used undefined v17 and v18
+- Root cause: JSON emitter reordered instructions, breaking def-use chain
+
+### Original JSON Order (Broken)
+```json
+{
+  "id": 3,
+  "instructions": [
+    {"op": "const", "dst": 16},
+    {"op": "binop", "dst": 19, "lhs": 17, "rhs": 18},  // ❌ Use v17, v18 before def
+    {"op": "copy", "dst": 17, "src": 16},             // ⚠️ Define v17 AFTER use
+    {"op": "copy", "dst": 18, "src": 1},              // ⚠️ Define v18 AFTER use
+    {"op": "copy", "dst": 20, "src": 19}
+  ]
+}
+```
+
+## Root Cause Analysis
+
+`src/runner/mir_json_emit.rs` が "use-before-def copy 回避" を目的に **copy の遅延/再配置**を行っており、
+依存（copy dst を参照する binop 等）まで含めた正しいスケジューリングになっていなかった。
+
+この層（JSON emitter）は optimizer/scheduler ではないため、順序修正を試みるより **Fail-Fast で upstream を炙り出す**のが筋。
+
+### Responsible Code (Lines 193-266, 652-710)
+```rust
+// Pre-scan: collect values defined anywhere in this block (to delay use-before-def copies)
+let mut block_defines: std::collections::HashSet<u32> = ...;
+let mut emitted_defs: std::collections::HashSet<u32> = ...;
+let mut delayed_copies: Vec<(u32, u32)> = Vec::new();
+
+// Delay copies if source will be defined later
+if block_defines.contains(&s) && !emitted_defs.contains(&s) {
+    delayed_copies.push((d, s));  // ❌ Reordering!
+}
+
+// Emit delayed copies after sources should be available
+for (d, s) in delayed_copies {
+    insts.push(json!({"op":"copy","dst": d, "src": s}));
+}
+```
+
+## Solution: SSOT Principle ✅
+
+**Box Theory Diagnosis**: Responsibility boundary violation
+- **JSON Emitter responsibility**: Output MIR instructions in original order
+- **Builder/Optimizer/Verifier responsibility**: Ensure correct def-use order upstream
+
+### SSOT: 「順序を直す場所」を固定する
+
+- ✅ JSON emitter は "順序を変えない"（MIR の命令列をそのまま出力する）
+- ✅ もし MIR 自体に use-before-def があるなら、builder/optimizer/verifier 側で直す（またはそこで Fail-Fast）
+
+### Fix Implementation
+
+**P0-1: Remove all reordering logic** (except PHI consolidation)
+```rust
+// Phase 131-13: Emit all instructions in MIR order (SSOT principle)
+// No reordering except PHI consolidation at block start (LLVM constraint)
+
+// Step 1: Emit all PHI instructions first (LLVM requirement)
+for inst in &bb.instructions {
+    if let I::Phi { ... } => { insts.push(phi_inst); }
+}
+
+// Step 2: Emit all non-PHI instructions in MIR order (no reordering!)
+for inst in &bb.instructions {
+    match inst {
+        I::Phi { .. } => continue,  // Already emitted
+        I::Copy { dst, src } => {
+            insts.push(json!({"op":"copy","dst": dst.as_u32(), "src": src.as_u32()}));
+        }
+        // ... other instructions in order
+    }
+}
+```
+
+### Emit 規約 (Updated)
+
+- ✅ PHI は block の先頭（LLVM 制約）。JSON 側で先頭に集約済み。
+- ✅ 非PHIは MIR の順序を保持して出力する（並べ替えない）。
+- ✅ Terminator は最後。
+
+**Changes**:
+- ✅ Removed `block_defines` HashSet
+- ✅ Removed `emitted_defs` HashSet
+- ✅ Removed `delayed_copies` Vec
+- ✅ Removed all `emitted_defs.insert()` calls
+- ✅ Removed delayed copy emission loop
+- ✅ Applied fix to both `emit_mir_json_for_harness` (lib) and `emit_mir_json_for_harness_bin` (bin)
+
+### Fixed JSON Order ✅
+```json
+{
+  "id": 3,
+  "instructions": [
+    {"op": "phi", "dst": 1},
+    {"op": "const", "dst": 16},
+    {"op": "copy", "dst": 17, "src": 16},  // ✅ Define v17 first
+    {"op": "copy", "dst": 18, "src": 1},   // ✅ Define v18 second
+    {"op": "binop", "dst": 19, "lhs": 17, "rhs": 18},  // ✅ Use v17, v18 after def
+    {"op": "copy", "dst": 20, "src": 19},
+    {"op": "mir_call", ...},
+    {"op": "const", "dst": 21},
+    {"op": "ret"}
+  ]
+}
+```
+
+## Verification ✅
+
+### Test Execution
+```bash
+# Generate fixed JSON
+NYASH_DISABLE_PLUGINS=1 ./target/release/hakorune --backend mir \
+  --emit-mir-json /tmp/fixed.json apps/tests/llvm_stage3_loop_only.hako
+# Output: MIR JSON written: /tmp/fixed.json
+
+# Verify instruction order
+cat /tmp/fixed.json | jq '.functions[] | select(.name == "main") | .blocks[1].instructions'
+# ✅ Correct: copy dst=17, copy dst=18, binop lhs=17 rhs=18
+```
+
+### Build Status
+```bash
+cargo build --release
+# ✅ 0 errors, 0 warnings
+```
+
+## Done 条件 ✅
+
+- ✅ Case C の JSON 出力で def→use 順序が正しい
+- ✅ LLVM harness が正しく型解決できる（use-before-def エラーなし）
+- ✅ ビルド成功（0 errors, 0 warnings）
+
+## Fail-Fast チェック（Future Work）
+
+JSON emit 時に、同一ブロック内の use-before-def を検出したら明示エラーにする（"直そうとしない"）。
+
+**P1: Add use-before-def detector** (NYASH_MIR_STRICT mode) - TODO
+```rust
+fn check_use_before_def(block: &BasicBlock) -> Result<(), String> {
+    let mut defined: HashSet<ValueId> = HashSet::new();
+    for inst in &block.instructions {
+        // Check all operands are defined
+        for operand in inst.operands() {
+            if !defined.contains(&operand) && !is_phi(inst) && !is_block_param(operand) {
+                if strict_mode() {
+                    return Err(format!("Use-before-def: v{}", operand.as_u32()));
+                } else {
+                    eprintln!("[WARN] Use-before-def: v{}", operand.as_u32());
+                }
+            }
+        }
+        if let Some(dst) = inst.dst() {
+            defined.insert(dst);
+        }
+    }
+    Ok(())
+}
+```
+
+## Files Modified
+
+- `src/runner/mir_json_emit.rs`: Core fix (removed ~80 lines of reordering logic)
+
+## Box Theory Insights
+
+**Fail-Fast Principle**: Instead of hiding errors with reordering, expose them early
+- ❌ Old approach: Emitter tries to fix broken MIR (scheduler role)
+- ✅ New approach: Emitter outputs MIR as-is, builder ensures correctness
+
+**SSOT Boundary**:
+- **Upstream** (builder.rs): Responsible for def-use order
+- **Emitter** (mir_json_emit.rs): Responsible for faithful output
+- **Downstream** (LLVM harness): Expects correct order from JSON
+

docs/development/current/main/phase131-3-llvm-lowering-inventory.md

@@ -211,6 +211,11 @@ VM and MIR look correct, but LLVM output does not match expected result for Case
   - `return counter` (no string concat)
   - `print(counter)` (no `"Result: " + ...`)
   - Compare with VM and inspect the IR use-sites.
+
+**Update (Phase 131-13)**:
+- snapshot-only + strict resolver により、Case C の不一致が “LLVM の値解決バグ” ではなく  
+  “Rust の MIR→JSON emit が block 内命令順序を崩している” 問題として顕在化した。
+  - Investigation note: `docs/development/current/main/investigations/phase131-13-mir-json-instruction-order.md`
 - Add `is_infinite_loop: bool` feature to `LoopFeatures` (detect `loop(true)`).
 - Fix classification so `has_break && has_continue` does not route to Pattern 4.
 - Introduce a dedicated pattern kind + lowerer for **infinite loop + early-exit (+ optional continue)**:

src/llvm_py/builders/block_lower.py

@@ -1,9 +1,51 @@
 from typing import Dict, Any, List, Tuple, NamedTuple
+import os
+import sys
 from llvmlite import ir
 from trace import debug as trace_debug
 from trace import phi_json as trace_phi_json
 
 
+def is_jump_only_block(block_info: Dict) -> bool:
+    """Phase 131-14-B: Detect pure jump-only blocks (trampoline blocks).
+
+    A pure jump-only block has:
+    - NO PHI instructions (PHI blocks do meaningful work - value merging)
+    - NO other instructions except a single terminator (jump/branch/ret)
+    - Acts as a pure trampoline/routing block
+
+    Blocks with PHI instructions are NOT jump-only because they perform value
+    merging and must compute their own snapshots.
+    """
+    instructions = block_info.get("instructions", [])
+
+    # Check if block has any PHI instructions
+    has_phi = any(i.get("op") == "phi" for i in instructions)
+    if has_phi:
+        # PHI blocks are NOT jump-only - they do value merging
+        return False
+
+    # Check if block has only terminator instructions
+    non_term = [
+        i for i in instructions
+        if i.get("op") not in ("ret", "jump", "branch")
+    ]
+    return len(non_term) == 0
+
+
+def get_predecessors(bid: int, preds: Dict[int, List[int]]) -> List[int]:
+    """Phase 131-14 P0-3: Get predecessors for a block.
+
+    Args:
+        bid: Block ID
+        preds: Predecessor map (bid -> [predecessor_bids])
+
+    Returns:
+        List of predecessor block IDs
+    """
+    return preds.get(bid, [])
+
+
 class DeferredTerminator(NamedTuple):
     """Phase 131-12-P1: Deferred terminator with vmap snapshot.
 
@@ -15,12 +57,138 @@ class DeferredTerminator(NamedTuple):
     vmap_snapshot: Dict[int, ir.Value]
 
 
+def resolve_jump_only_snapshots(builder, block_by_id: Dict[int, Dict[str, Any]]):
+    """Phase 131-14-B P0-2: Resolve jump-only block snapshots (Pass B).
+
+    This function runs AFTER all blocks have been lowered (Pass A) but BEFORE
+    PHI finalization. It resolves snapshots for jump-only blocks by following
+    the CFG to find the nearest non-jump-only predecessor.
+
+    Uses path compression to efficiently handle chains of jump-only blocks.
+
+    SSOT: Snapshots are based on CFG structure, not processing order.
+    """
+    import sys
+
+    strict_mode = os.environ.get('NYASH_LLVM_STRICT') == '1'
+    trace_vmap = os.environ.get('NYASH_LLVM_TRACE_VMAP') == '1'
+
+    jump_only = getattr(builder, '_jump_only_blocks', {})
+    if not jump_only:
+        if trace_vmap:
+            print("[vmap/resolve/passB] No jump-only blocks to resolve", file=sys.stderr)
+        return
+
+    if trace_vmap:
+        print(f"[vmap/resolve/passB] Resolving {len(jump_only)} jump-only blocks: {sorted(jump_only.keys())}", file=sys.stderr)
+
+    resolved = {}  # bid -> snapshot dict
+
+    def resolve(bid: int, visited: set | None = None) -> Dict[int, Any]:
+        """Recursively resolve snapshot for a block, with cycle detection."""
+        if visited is None:
+            visited = set()
+
+        # Cycle detection
+        if bid in visited:
+            if strict_mode:
+                raise RuntimeError(
+                    f"[LLVM_PY/STRICT] Phase 131-14-B: Cycle detected in jump-only chain: "
+                    f"{visited} -> {bid}"
+                )
+            if trace_vmap:
+                print(f"[vmap/resolve/passB] WARNING: Cycle at bb{bid}, returning empty", file=sys.stderr)
+            return {}
+
+        visited.add(bid)
+
+        # Already resolved (path compression cache)
+        if bid in resolved:
+            if trace_vmap:
+                print(f"[vmap/resolve/passB] bb{bid} already resolved (cached)", file=sys.stderr)
+            return resolved[bid]
+
+        # Normal block - already has snapshot from Pass A
+        if bid in builder.block_end_values:
+            snapshot = builder.block_end_values[bid]
+            if trace_vmap:
+                print(
+                    f"[vmap/resolve/passB] bb{bid} is normal block with snapshot "
+                    f"({len(snapshot)} values)",
+                    file=sys.stderr
+                )
+            return snapshot
+
+        # Jump-only block - resolve from predecessor
+        if bid in jump_only:
+            pred_bid = jump_only[bid]
+            if trace_vmap:
+                print(f"[vmap/resolve/passB] bb{bid} is jump-only, resolving from pred bb{pred_bid}", file=sys.stderr)
+
+            # Recursively resolve predecessor
+            pred_snapshot = resolve(pred_bid, visited)
+
+            if not pred_snapshot:
+                if strict_mode:
+                    raise RuntimeError(
+                        f"[LLVM_PY/STRICT] Phase 131-14-B: jump-only block bb{bid} "
+                        f"cannot resolve snapshot from predecessor bb{pred_bid} "
+                        f"(predecessor has no snapshot)"
+                    )
+                if trace_vmap:
+                    print(
+                        f"[vmap/resolve/passB] WARNING: bb{bid} pred bb{pred_bid} has no snapshot, "
+                        f"using empty dict",
+                        file=sys.stderr
+                    )
+                pred_snapshot = {}
+
+            # Cache the result (path compression)
+            resolved[bid] = dict(pred_snapshot)
+            if trace_vmap:
+                print(
+                    f"[vmap/resolve/passB] bb{bid} resolved from bb{pred_bid}: "
+                    f"{len(resolved[bid])} values",
+                    file=sys.stderr
+                )
+            return resolved[bid]
+
+        # Unknown block (should not happen if Pass A worked correctly)
+        if strict_mode:
+            raise RuntimeError(
+                f"[LLVM_PY/STRICT] Phase 131-14-B: block bb{bid} is neither normal "
+                f"nor jump-only (invalid state)"
+            )
+
+        if trace_vmap:
+            print(f"[vmap/resolve/passB] WARNING: bb{bid} unknown state, returning empty", file=sys.stderr)
+        return {}
+
+    # Resolve all jump-only blocks
+    for bid in sorted(jump_only.keys()):
+        snapshot = resolve(bid)
+        builder.block_end_values[bid] = snapshot
+
+        if trace_vmap:
+            print(
+                f"[vmap/resolve/passB] ✅ bb{bid} final snapshot: "
+                f"{len(snapshot)} values, keys={sorted(snapshot.keys())[:10]}",
+                file=sys.stderr
+            )
+
+    if trace_vmap:
+        print(f"[vmap/resolve/passB] Pass B complete: resolved {len(jump_only)} jump-only blocks", file=sys.stderr)
+
+
 def lower_blocks(builder, func: ir.Function, block_by_id: Dict[int, Dict[str, Any]], order: List[int], loop_plan: Dict[str, Any] | None):
     """Lower blocks in multi-pass to ensure PHIs are always before terminators.
 
     Phase 131-4: Multi-pass block lowering architecture
+    Phase 131-14-B: Two-pass snapshot resolution
     - Pass A: Lower non-terminator instructions only (terminators deferred)
+      - jump-only blocks: record metadata only, NO snapshot resolution
     - Pass B: PHI finalization happens in function_lower.py
+      - resolve_jump_only_snapshots() called BEFORE PHI finalization
     - Pass C: Lower terminators (happens after PHI finalization)
 
     This ensures LLVM IR invariant: PHI nodes must be at block head before any
@@ -278,16 +446,94 @@ def lower_blocks(builder, func: ir.Function, block_by_id: Dict[int, Dict[str, An
         except Exception:
             pass
         # End-of-block snapshot
-        snap = dict(vmap_cur)
+        # Phase 131-14-B P0-1: Jump-only blocks - record metadata only (Pass A)
-        try:
+        strict_mode = os.environ.get('NYASH_LLVM_STRICT') == '1'
-            keys = sorted(list(snap.keys()))
+        trace_vmap = os.environ.get('NYASH_LLVM_TRACE_VMAP') == '1'
-        except Exception:
+
-            keys = list(snap.keys())
+        # Initialize jump_only_blocks dict if not exists
-        trace_phi_json({"phi": "snapshot", "block": int(bid), "keys": [int(k) for k in keys[:20]]})
+        if not hasattr(builder, '_jump_only_blocks'):
-        for vid in created_ids:
+            builder._jump_only_blocks = {}
-            if vid in vmap_cur:
+
-                builder.def_blocks.setdefault(vid, set()).add(block_data.get("id", 0))
+        is_jump_only = is_jump_only_block(block_data)
-        builder.block_end_values[bid] = snap
+        if trace_vmap:
+            print(
+                f"[vmap/snapshot] bb{bid} is_jump_only={is_jump_only} "
+                f"instructions={[i.get('op') for i in block_data.get('instructions', [])]}",
+                file=sys.stderr
+            )
+
+        if is_jump_only:
+            # Phase 131-14-B: Jump-only blocks - record metadata, defer snapshot resolution to Pass B
+            preds_list = get_predecessors(bid, builder.preds)
+
+            if len(preds_list) == 0:
+                # No predecessors - error in STRICT mode
+                if strict_mode:
+                    raise RuntimeError(
+                        f"[LLVM_PY/STRICT] Phase 131-14-B: jump-only block bb{bid} "
+                        f"has no predecessors (orphan trampoline)"
+                    )
+                # Non-STRICT: use current vmap_cur (defensive fallback)
+                snap = dict(vmap_cur)
+                if trace_vmap:
+                    print(
+                        f"[vmap/snapshot] bb{bid} jump-only with 0 preds: "
+                        f"using vmap_cur keys={sorted(snap.keys())}",
+                        file=sys.stderr
+                    )
+            elif len(preds_list) == 1:
+                # Single predecessor - record metadata for Pass B resolution
+                pred_bid = preds_list[0]
+                builder._jump_only_blocks[bid] = pred_bid
+
+                # DO NOT create snapshot here - will be resolved in Pass B
+                # Set snap to None to indicate "skip storing in block_end_values"
+                snap = None
+
+                if trace_vmap:
+                    print(
+                        f"[vmap/snapshot/passA] bb{bid} jump-only: recorded pred=bb{pred_bid}, "
+                        f"snapshot deferred to Pass B",
+                        file=sys.stderr
+                    )
+            else:
+                # Multiple predecessors - error in STRICT mode (merge rules not yet defined)
+                if strict_mode:
+                    raise RuntimeError(
+                        f"[LLVM_PY/STRICT] Phase 131-14-B: jump-only block bb{bid} "
+                        f"has multiple predecessors: {preds_list} "
+                        f"(merge propagation not implemented)"
+                    )
+                # Non-STRICT: use current vmap_cur (defensive fallback)
+                snap = dict(vmap_cur)
+                if trace_vmap:
+                    print(
+                        f"[vmap/snapshot] bb{bid} jump-only with multiple preds {preds_list}: "
+                        f"using vmap_cur keys={sorted(snap.keys())}",
+                        file=sys.stderr
+                    )
+        else:
+            # Normal block: use its own vmap_cur
+            snap = dict(vmap_cur)
+
+        # Phase 131-14-B: Only store snapshot if not deferred (snap is not None)
+        if snap is not None:
+            try:
+                keys = sorted(list(snap.keys()))
+            except Exception:
+                keys = list(snap.keys())
+            trace_phi_json({"phi": "snapshot", "block": int(bid), "keys": [int(k) for k in keys[:20]]})
+            for vid in created_ids:
+                if vid in vmap_cur:
+                    builder.def_blocks.setdefault(vid, set()).add(block_data.get("id", 0))
+            builder.block_end_values[bid] = snap
+        else:
+            # Jump-only block with deferred snapshot - don't store yet
+            if trace_vmap:
+                print(
+                    f"[vmap/snapshot/passA] bb{bid} snapshot deferred (not stored in block_end_values)",
+                    file=sys.stderr
+                )
         try:
             delattr(builder, '_current_vmap')
         except Exception:

src/llvm_py/builders/function_lower.py

@@ -279,6 +279,10 @@ def lower_function(builder, func_data: Dict[str, Any]):
     from builders.block_lower import lower_blocks as _lower_blocks
     _lower_blocks(builder, func, block_by_id, order, loop_plan)
 
+    # Phase 131-14-B Pass B: Resolve jump-only block snapshots (BEFORE PHI finalization)
+    from builders.block_lower import resolve_jump_only_snapshots as _resolve_jump_only_snapshots
+    _resolve_jump_only_snapshots(builder, block_by_id)
+
     # Optional: capture lowering ctx for downstream helpers
     try:
         builder.ctx = dict(
@@ -300,7 +304,7 @@ def lower_function(builder, func_data: Dict[str, Any]):
     except Exception:
         pass
 
-    # Phase 131-4 Pass B: Finalize PHIs (wires incoming edges)
+    # Phase 131-4 Pass B (now Pass B2): Finalize PHIs (wires incoming edges)
     _finalize_phis(builder)
 
     # Phase 131-4 Pass C: Lower deferred terminators (after PHIs are placed)

src/llvm_py/phi_wiring/wiring.py

@@ -202,9 +202,8 @@ def wire_incomings(builder, block_id: int, dst_vid: int, incoming: List[Tuple[in
         if original_vs != vs:
             trace({"phi": "wire_replaced_src", "original": original_vs, "replaced": vs})
         try:
-            val = builder.resolver._value_at_end_i64(
+            # P0-4: Use resolve_incoming for PHI incoming values
-                vs, pred_match, builder.preds, builder.block_end_values, builder.vmap, builder.bb_map
+            val = builder.resolver.resolve_incoming(pred_match, vs)
-            )
             trace({"phi": "wire_resolved", "vs": vs, "pred": pred_match, "val_type": type(val).__name__})
         except Exception as e:
             trace({"phi": "wire_resolve_fail", "vs": vs, "pred": pred_match, "error": str(e)})

src/llvm_py/resolver.py

@@ -3,7 +3,7 @@ Resolver API (Python version)
 Based on src/backend/llvm/compiler/codegen/instructions/resolver.rs
 """
 
-from typing import Dict, Optional, Any, Tuple
+from typing import Dict, Optional, Any, Tuple, Set
 import os
 from trace import phi as trace_phi
 from trace import values as trace_values
@@ -63,6 +63,8 @@ class Resolver:
         self.block_phi_incomings = {}
         # P0-1: SSOT for end-of-block values (snapshots)
         self.block_end_values = {}
+        # P0-3: Circular reference detection (hang prevention)
+        self._visited: Set[Tuple[int, int]] = set()
 
     def mark_string(self, value_id: int) -> None:
         try:
@@ -75,7 +77,81 @@ class Resolver:
             return int(value_id) in self.string_ids
         except Exception:
             return False
-    
+
+    def _check_cycle(self, block_id: int, value_id: int):
+        """P0-3: Circular reference detection (hang prevention)"""
+        key = (block_id, value_id)
+        if key in self._visited:
+            raise RuntimeError(
+                f"[LLVM_PY] Circular reference detected: bb{block_id} v{value_id}"
+            )
+        self._visited.add(key)
+
+    def resolve_cur(self, block_id: int, value_id: int, vmap_cur: Dict[int, ir.Value]) -> ir.Value:
+        """P0-1: Same-block instruction lowering (vmap_cur as primary source)
+
+        Used for lowering instructions within the same basic block where the value
+        is defined and used. Checks vmap_cur first, then applies fail-fast checks.
+
+        Args:
+            block_id: Current basic block ID
+            value_id: Value ID to resolve
+            vmap_cur: Current block's value map (def->use tracking)
+
+        Returns:
+            LLVM IR value (i64)
+        """
+        # 1. Check vmap_cur first
+        val = vmap_cur.get(value_id)
+        if val is not None:
+            return val
+
+        # 2. Fail-Fast: def_blocks has bb but vmap_cur doesn't → lowerer bug
+        if value_id in self.def_blocks and block_id in self.def_blocks[value_id]:
+            if os.environ.get('NYASH_LLVM_STRICT') == '1':
+                raise RuntimeError(
+                    f"[LLVM_PY/STRICT] resolve_cur: v{value_id} defined in bb{block_id} "
+                    f"but not in vmap_cur. Lowerer order bug?"
+                )
+
+        # 3. vmap_cur miss → undefined error
+        if os.environ.get('NYASH_LLVM_STRICT') == '1':
+            raise RuntimeError(
+                f"[LLVM_PY/STRICT] resolve_cur: v{value_id} not found in bb{block_id} vmap_cur. "
+                f"Available: {sorted(vmap_cur.keys())}"
+            )
+
+        # Non-STRICT: fallback to 0
+        return ir.Constant(ir.IntType(64), 0)
+
+    def resolve_incoming(self, pred_block_id: int, value_id: int) -> ir.Value:
+        """P0-2: PHI incoming resolution (snapshot-only reference)
+
+        Used for resolving PHI incoming values from predecessor blocks.
+        Only looks at block_end_values snapshot, never vmap_cur.
+
+        Args:
+            pred_block_id: Predecessor block ID
+            value_id: Value ID to resolve from predecessor
+
+        Returns:
+            LLVM IR value (i64)
+        """
+        snapshot = self.block_end_values.get(pred_block_id, {})
+        val = snapshot.get(value_id)
+        if val is not None:
+            return val
+
+        # Fail-Fast: snapshot miss → structural bug
+        if os.environ.get('NYASH_LLVM_STRICT') == '1':
+            raise RuntimeError(
+                f"[LLVM_PY/STRICT] resolve_incoming: v{value_id} not in bb{pred_block_id} snapshot. "
+                f"Available: {sorted(snapshot.keys())}"
+            )
+
+        # Non-STRICT: fallback to 0
+        return ir.Constant(ir.IntType(64), 0)
+
     def resolve_i64(
         self,
         value_id: int,

src/runner/mir_json_emit.rs

@@ -190,43 +190,11 @@ pub fn emit_mir_json_for_harness(
         for bid in ids {
             if let Some(bb) = f.blocks.get(&bid) {
                 let mut insts = Vec::new();
-                // Pre-scan: collect values defined anywhere in this block (to delay use-before-def copies)
+                // Phase 131-13: Emit all instructions in MIR order (SSOT principle)
-                let mut block_defines: std::collections::HashSet<u32> =
+                // No reordering except PHI consolidation at block start (LLVM constraint)
-                    std::collections::HashSet::new();
+
+                // Step 1: Emit all PHI instructions first (LLVM requirement)
                 for inst in &bb.instructions {
-                    match inst {
-                        I::UnaryOp { dst, .. }
-                        | I::Const { dst, .. }
-                        | I::BinOp { dst, .. }
-                        | I::Compare { dst, .. }
-                        | I::Call { dst: Some(dst), .. }
-                        | I::ExternCall { dst: Some(dst), .. }
-                        | I::BoxCall { dst: Some(dst), .. }
-                        | I::NewBox { dst, .. }
-                        | I::Phi { dst, .. } => {
-                            block_defines.insert(dst.as_u32());
-                        }
-                        _ => {}
-                    }
-                }
-                // Track which values have been emitted (to order copies after their sources)
-                let mut emitted_defs: std::collections::HashSet<u32> =
-                    std::collections::HashSet::new();
-                // PHI first（オプション）
-                for inst in &bb.instructions {
-                    if let I::Copy { dst, src } = inst {
-                        // For copies whose source will be defined later in this block, delay emission
-                        let s = src.as_u32();
-                        if block_defines.contains(&s) && !emitted_defs.contains(&s) {
-                            // delayed; will be emitted after non-PHI pass
-                        } else {
-                            insts.push(
-                                json!({"op":"copy","dst": dst.as_u32(), "src": src.as_u32()}),
-                            );
-                            emitted_defs.insert(dst.as_u32());
-                        }
-                        continue;
-                    }
                     if let I::Phi { dst, inputs, .. } = inst {
                         let incoming: Vec<_> = inputs
                             .iter()
@@ -250,20 +218,16 @@ pub fn emit_mir_json_for_harness(
                         insts.push(phi_inst);
                     }
                 }
-                // Non-PHI
+
-                // Non-PHI
+                // Step 2: Emit all non-PHI instructions in MIR order (no reordering!)
-                let mut delayed_copies: Vec<(u32, u32)> = Vec::new();
                 for inst in &bb.instructions {
                     match inst {
+                        I::Phi { .. } => {
+                            // Already emitted in step 1
+                            continue;
+                        }
                         I::Copy { dst, src } => {
-                            let d = dst.as_u32();
+                            insts.push(json!({"op":"copy","dst": dst.as_u32(), "src": src.as_u32()}));
-                            let s = src.as_u32();
-                            if block_defines.contains(&s) && !emitted_defs.contains(&s) {
-                                delayed_copies.push((d, s));
-                            } else {
-                                insts.push(json!({"op":"copy","dst": d, "src": s}));
-                                emitted_defs.insert(d);
-                            }
                         }
                         I::UnaryOp { dst, op, operand } => {
                             let kind = match op {
@@ -322,7 +286,6 @@ pub fn emit_mir_json_for_harness(
                                 "dst": dst.as_u32(),
                                 "target_type": ty_s,
                             }));
-                            emitted_defs.insert(dst.as_u32());
                         }
                         I::BinOp { dst, op, lhs, rhs } => {
                             let op_s = match op {
@@ -456,7 +419,6 @@ pub fn emit_mir_json_for_harness(
                                         }
                                         insts.push(obj);
                                         if let Some(d) = dst.map(|v| v.as_u32()) {
-                                            emitted_defs.insert(d);
                                         }
                                     }
                                     _ => {
@@ -531,7 +493,6 @@ pub fn emit_mir_json_for_harness(
                             }
                             insts.push(obj);
                             if let Some(d) = dst.map(|v| v.as_u32()) {
-                                emitted_defs.insert(d);
                             }
                         }
                         I::NewBox {
@@ -541,7 +502,6 @@ pub fn emit_mir_json_for_harness(
                         } => {
                             let args_a: Vec<_> = args.iter().map(|v| json!(v.as_u32())).collect();
                             insts.push(json!({"op":"newbox","type": box_type, "args": args_a, "dst": dst.as_u32()}));
-                            emitted_defs.insert(dst.as_u32());
                         }
                         I::Branch {
                             condition,
@@ -559,10 +519,7 @@ pub fn emit_mir_json_for_harness(
                         _ => { /* skip non-essential ops for initial harness */ }
                     }
                 }
-                // Emit delayed copies now (sources should be available)
+                // Phase 131-13: Terminator emitted inline (no delayed copies)
-                for (d, s) in delayed_copies {
-                    insts.push(json!({"op":"copy","dst": d, "src": s}));
-                }
                 if let Some(term) = &bb.terminator {
                     match term {
                         I::Return { value } => insts.push(json!({"op":"ret","value": value.map(|v| v.as_u32())})),
@@ -650,27 +607,10 @@ pub fn emit_mir_json_for_harness_bin(
         for bid in ids {
             if let Some(bb) = f.blocks.get(&bid) {
                 let mut insts = Vec::new();
-                // Pre-scan to collect values defined in this block
+                // Phase 131-13: Emit all instructions in MIR order (SSOT principle)
-                let mut block_defines: std::collections::HashSet<u32> =
+                // No reordering except PHI consolidation at block start (LLVM constraint)
-                    std::collections::HashSet::new();
+
-                for inst in &bb.instructions {
+                // Step 1: Emit all PHI instructions first (LLVM requirement)
-                    match inst {
-                        I::Copy { dst, .. }
-                        | I::Const { dst, .. }
-                        | I::BinOp { dst, .. }
-                        | I::Compare { dst, .. }
-                        | I::Call { dst: Some(dst), .. }
-                        | I::ExternCall { dst: Some(dst), .. }
-                        | I::BoxCall { dst: Some(dst), .. }
-                        | I::NewBox { dst, .. }
-                        | I::Phi { dst, .. } => {
-                            block_defines.insert(dst.as_u32());
-                        }
-                        _ => {}
-                    }
-                }
-                let mut emitted_defs: std::collections::HashSet<u32> =
-                    std::collections::HashSet::new();
                 for inst in &bb.instructions {
                     if let I::Phi { dst, inputs, .. } = inst {
                         let incoming: Vec<_> = inputs
@@ -693,21 +633,18 @@ pub fn emit_mir_json_for_harness_bin(
                             }
                         }
                         insts.push(phi_inst);
-                        emitted_defs.insert(dst.as_u32());
                     }
                 }
-                let mut delayed_copies: Vec<(u32, u32)> = Vec::new();
+
+                // Step 2: Emit all non-PHI instructions in MIR order (no reordering!)
                 for inst in &bb.instructions {
                     match inst {
+                        I::Phi { .. } => {
+                            // Already emitted in step 1
+                            continue;
+                        }
                         I::Copy { dst, src } => {
-                            let d = dst.as_u32();
+                            insts.push(json!({"op":"copy","dst": dst.as_u32(), "src": src.as_u32()}));
-                            let s = src.as_u32();
-                            if block_defines.contains(&s) && !emitted_defs.contains(&s) {
-                                delayed_copies.push((d, s));
-                            } else {
-                                insts.push(json!({"op":"copy","dst": d, "src": s}));
-                                emitted_defs.insert(d);
-                            }
                         }
                         I::Const { dst, value } => {
                             match value {
@@ -734,7 +671,6 @@ pub fn emit_mir_json_for_harness_bin(
                                     insts.push(json!({"op":"const","dst": dst.as_u32(), "value": {"type": "void", "value": 0}}));
                                 }
                             }
-                            emitted_defs.insert(dst.as_u32());
                         }
                         I::BinOp { dst, op, lhs, rhs } => {
                             let op_s = match op {
@@ -772,7 +708,6 @@ pub fn emit_mir_json_for_harness_bin(
                                 }
                             }
                             insts.push(obj);
-                            emitted_defs.insert(dst.as_u32());
                         }
                         I::Compare { dst, op, lhs, rhs } => {
                             let op_s = match op {
@@ -784,7 +719,6 @@ pub fn emit_mir_json_for_harness_bin(
                                 C::Ge => ">=",
                             };
                             insts.push(json!({"op":"compare","operation": op_s, "lhs": lhs.as_u32(), "rhs": rhs.as_u32(), "dst": dst.as_u32()}));
-                            emitted_defs.insert(dst.as_u32());
                         }
                         I::Call {
                             dst,
@@ -853,7 +787,6 @@ pub fn emit_mir_json_for_harness_bin(
                                         }
                                         insts.push(obj);
                                         if let Some(d) = dst.map(|v| v.as_u32()) {
-                                            emitted_defs.insert(d);
                                         }
                                     }
                                     _ => {
@@ -889,7 +822,6 @@ pub fn emit_mir_json_for_harness_bin(
                             }
                             insts.push(obj);
                             if let Some(d) = dst.map(|v| v.as_u32()) {
-                                emitted_defs.insert(d);
                             }
                         }
                         I::BoxCall {
@@ -921,7 +853,6 @@ pub fn emit_mir_json_for_harness_bin(
                             }
                             insts.push(obj);
                             if let Some(d) = dst.map(|v| v.as_u32()) {
-                                emitted_defs.insert(d);
                             }
                         }
                         I::NewBox {
@@ -931,7 +862,6 @@ pub fn emit_mir_json_for_harness_bin(
                         } => {
                             let args_a: Vec<_> = args.iter().map(|v| json!(v.as_u32())).collect();
                             insts.push(json!({"op":"newbox","type": box_type, "args": args_a, "dst": dst.as_u32()}));
-                            emitted_defs.insert(dst.as_u32());
                         }
                         I::Branch {
                             condition,
@@ -949,10 +879,7 @@ pub fn emit_mir_json_for_harness_bin(
                         _ => {}
                     }
                 }
-                // Append delayed copies after their sources
+                // Phase 131-13: Terminator emitted inline (no delayed copies)
-                for (d, s) in delayed_copies {
-                    insts.push(json!({"op":"copy","dst": d, "src": s}));
-                }
                 if let Some(term) = &bb.terminator {
                     match term {
                     I::Return { value } => insts.push(json!({"op":"ret","value": value.map(|v| v.as_u32())})),