diff --git a/BMM/BMM_Theory/bmm_theory/io_csv.py b/BMM/BMM_Theory/bmm_theory/io_csv.py
new file mode 100644
index 0000000..68c36e9
--- /dev/null
+++ b/BMM/BMM_Theory/bmm_theory/io_csv.py
@@ -0,0 +1,101 @@
+"""csv 输入输出层.
+
+case 输入 csv 列 (表头, 大小写不敏感, 缺省列取默认值):
+    case_id, batch_a, batch_b, m, n, k,
+    dtype_a, dtype_b, dtype_c, trans_a, trans_b, has_bias, deterministic_level
+
+方案输入 csv (评估模式): ImplPlan 全部字段, 见 ImplPlan.csv_fields().
+
+输出 csv:
+  - 方案推荐模式: case 列 + plan_* 列 + 时延评估列 + feasible/bound_type/advice
+  - 方案评估模式: 同上, plan 来自用户输入, feasible/violations 反映约束校验
+"""
+
+from __future__ import annotations
+
+import csv
+from pathlib import Path
+
+from .models import BmmCase, ImplPlan, EvalResult
+
+
+_TRUE = {"1", "true", "yes", "y", "是"}
+
+
+def _to_bool(v, default=False):
+    if v is None or str(v).strip() == "":
+        return default
+    return str(v).strip().lower() in _TRUE
+
+
+def _to_int(v, default=0):
+    if v is None or str(v).strip() == "":
+        return default
+    return int(float(v))
+
+
+def _to_str(v, default=""):
+    if v is None:
+        return default
+    v = str(v).strip()
+    return v if v else default
+
+
+def load_cases(path: str | Path) -> list:
+    """读取 case csv -> list[BmmCase]."""
+    cases = []
+    with open(path, newline="", encoding="utf-8-sig") as f:
+        for i, row in enumerate(csv.DictReader(f)):
+            r = {k.strip().lower(): v for k, v in row.items() if k}
+            try:
+                cases.append(BmmCase(
+                    case_id=_to_str(r.get("case_id"), f"case_{i}"),
+                    batch_a=_to_int(r.get("batch_a"), 1),
+                    batch_b=_to_int(r.get("batch_b"), 1),
+                    m=_to_int(r.get("m"), 1),
+                    n=_to_int(r.get("n"), 1),
+                    k=_to_int(r.get("k"), 1),
+                    dtype_a=_to_str(r.get("dtype_a"), "bf16"),
+                    dtype_b=_to_str(r.get("dtype_b"), "bf16"),
+                    dtype_c=_to_str(r.get("dtype_c"), "bf16"),
+                    trans_a=_to_bool(r.get("trans_a")),
+                    trans_b=_to_bool(r.get("trans_b")),
+                    has_bias=_to_bool(r.get("has_bias")),
+                    deterministic_level=_to_int(r.get("deterministic_level"), 0),
+                ))
+            except ValueError as e:
+                raise ValueError(f"{path} 第{i+2}行解析失败: {e}") from e
+    return cases
+
+
+def load_plans(path: str | Path) -> list:
+    """读取方案 csv -> list[ImplPlan] (按 case_id 与 case 关联)."""
+    plans = []
+    with open(path, newline="", encoding="utf-8-sig") as f:
+        for row in csv.DictReader(f):
+            r = {k.strip(): v for k, v in row.items() if k}
+            plans.append(ImplPlan.from_row(r))
+    return plans
+
+
+def save_results(path: str | Path, results: list) -> None:
+    """list[EvalResult] -> csv."""
+    if not results:
+        return
+    rows = [r.to_row() for r in results]
+    fields = list(rows[0].keys())
+    with open(path, "w", newline="", encoding="utf-8-sig") as f:
+        w = csv.DictWriter(f, fieldnames=fields)
+        w.writeheader()
+        w.writerows(rows)
+
+
+def save_plans(path: str | Path, plans: list) -> None:
+    """list[ImplPlan] -> csv (标准结构体列)."""
+    if not plans:
+        return
+    with open(path, "w", newline="", encoding="utf-8-sig") as f:
+        w = csv.DictWriter(f, fieldnames=ImplPlan.csv_fields())
+        w.writeheader()
+        for p in plans:
+            w.writerow(p.to_row())