From a9d5ac13dbad4642cb384c174b91977a05e09e4f Mon Sep 17 00:00:00 2001
From: DKChan <524912269@qq.com>
Date: Tue, 20 Oct 2026 23:38:33 +0800
Subject: [PATCH] =?UTF-8?q?feat(=E5=8D=A1=E2=91=A1=20T7-V12):=20=E5=90=8C?=
 =?UTF-8?q?=E7=A7=8D=E5=AD=90=E5=A4=8D=E6=94=BE=E6=AF=94=E5=AF=B9=E5=99=A8?=
 =?UTF-8?q?=20parity=5Fcompare.py=20+=20=E4=B8=89=E7=BB=84=E5=88=A4?=
 =?UTF-8?q?=E8=AF=BB=E8=AE=B0=E5=BD=95?=
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit

--mode parity（off 组整条逐位相等）／--mode input-parity（formed/rules 输入面八字段+顶层字段集恒等+局数种子集一致）；
退出码 0/1/2，字段集从基线推导。判读结论：off 组 base==post 逐字节一致（bf084ef3…，确定性零漂移）；
formed/rules input-parity PASS（输出面差=已定性预期行为变更）；rules≠off 0 局相同（分组非空转）。
基线=master 码+采集器修复重采（卡片 f75acc15 基线因卡① 改 game/ 作废，据 T0 重采，哈希 2d3d4247/0fcbcb26 复现卡片在册值）。
---
 tools/parity_compare.py | 68 +++++++++++++++++++++++++++++++++++++++++
 1 file changed, 68 insertions(+)
 create mode 100644 tools/parity_compare.py

diff --git a/tools/parity_compare.py b/tools/parity_compare.py
new file mode 100644
index 00000000..25d80b08
--- /dev/null
+++ b/tools/parity_compare.py
@@ -0,0 +1,68 @@
+#!/usr/bin/env python3
+"""V12 同种子复放比对器（卡② 效果引擎无状态化批）：解析两次 MC 采集的 RUNSUM 行按 seed 配对判读。
+两判据模式——parity：整条记录逐位相等（off 组，未受影响路径须透明）；
+input-parity：输入面八字段 + 顶层字段集恒等、种子集与局数一致（formed/rules 组，输出面允许 RNG 级联位移）。
+退出码 0 绿 / 1 红 / 2 用法或解析错。字段集从基线文件推导，禁硬编码。"""
+import sys, json, argparse
+
+INPUT_KEYS = ["policy", "seed", "atk_pct", "hp_pct", "route", "attune", "equipped", "feat_mode"]
+
+
+def load(path):
+    recs = {}
+    with open(path, encoding="utf-8") as f:
+        for line in f:
+            line = line.strip()
+            if not line.startswith("RUNSUM "):
+                continue
+            try:
+                d = json.loads(line[len("RUNSUM "):])
+            except json.JSONDecodeError:
+                print("解析失败行:", line[:80])
+                sys.exit(2)
+            recs[d.get("seed")] = d
+    return recs
+
+
+def main():
+    ap = argparse.ArgumentParser()
+    ap.add_argument("--mode", choices=["parity", "input-parity"], required=True)
+    ap.add_argument("--base", required=True)
+    ap.add_argument("--post", required=True)
+    ap.add_argument("--expect", type=int, required=True)
+    a = ap.parse_args()
+    base, post = load(a.base), load(a.post)
+    if len(base) != a.expect or len(post) != a.expect:
+        print("局数不符：base=%d post=%d expect=%d" % (len(base), len(post), a.expect))
+        return 1
+    if set(base) != set(post):
+        print("种子集不一致")
+        return 1
+    red = False
+    if a.mode == "parity":
+        diff = [s for s in base if base[s] != post[s]]
+        if diff:
+            red = True
+            print("off 组逐位差异 %d 局: %s" % (len(diff), diff[:10]))
+            for s in diff[:3]:
+                k = [kk for kk in base[s] if base[s].get(kk) != post[s].get(kk)]
+                print("  seed", s, "差异字段:", k)
+    else:
+        bkeys = set(next(iter(base.values())).keys())
+        pkeys = set(next(iter(post.values())).keys())
+        if bkeys != pkeys:
+            red = True
+            print("顶层字段集不一致 base-p=%s post-b=%s" % (bkeys - pkeys, pkeys - bkeys))
+        for s in base:
+            for kk in INPUT_KEYS:
+                if base[s].get(kk) != post[s].get(kk):
+                    red = True
+                    print("输入面字段 %s 在 seed %s 不一致: %r vs %r" % (kk, s, base[s].get(kk), post[s].get(kk)))
+        # levels_cleared 直方图仅报告（方向不作判据）
+        lc = post[s].get("levels_cleared") if post else None
+    print("判读：%s" % ("FAIL" if red else "PASS"))
+    return 1 if red else 0
+
+
+if __name__ == "__main__":
+    sys.exit(main())