docs: the root README table shows both mcpp columns, and the guard checks both

和 bench/README 同样的问题:根 README 只放默认配置,把 `bmi_schedule=on` 降级成
脚注里的一句「cold 35.4s」。读者看到的是 `cold 1.2x`、`edit-body 1.1x`,而实际
可达的是 **3x** 和 **3x**。

| 场景 | schedule=on | 默认 | cmake |
|---|---|---|---|
| cold | **35.43s · 3x** | 79.54s · 1.2x | 92.33s |
| touch-hub | **0.22s · 377x** | 0.40s · 207.9x | 83.39s |
| edit-body | **30.17s · 3x** | 76.24s · 1.1x | 85.64s |
| edit-comment | **0.18s · 458x** | 0.38s · 217.2x | 82.96s |

数据本来就在已发布的 JSON 里(五个场景全 ok),只是没被用。

⚠️ **差点重新发布一个已知的假数字。** 按 `mcpp-linux-gcc-5way.json` 直接重建表时,
xmake 的 cold 是 **0.60s · 153x** —— 那正是路径翻倍导致「测了一棵已构建好的树」的
幽灵。README 现用的 90.30s 来自 `mcpp-linux-gcc-xmake-refixed.json`,所以重建时
xmake 列必须继续取修正文件。**一份结果目录里同时躺着修正前后的两份数据,是个陷阱**;
两个文件名的区别是唯一的提示。

e2e 233 §5 当场拦下了表格变形(「did not parse — has its shape changed?」),这正是
它存在的意义。解析器已教会新形状,并且**新增的 schedule 列也纳入对数** —— 那是读者
目光最先落到的一列,不查它是最贵的一种漏。两个方向都验过:改掉该列的数字会红,
还原后绿。

Author: speak-agent

README.md

@@ -310,19 +310,20 @@ Building **mcpp itself** — 137 module interface units, 57k lines, every one of
 them `import std;` — with three engines given the **same compiler binary**.
 Each cell is the median wall-clock and how many times faster it is than cmake.
 
-| scenario | what changed | **mcpp** | cmake | xmake |
-|---|---|---|---|---|
-| `cold` | nothing built yet | **79.54s** · 1.2x | 92.33s · 1.0x | 90.30s · 1.0x |
-| `noop` | nothing at all | **0.16s** · 1.8x | 0.28s · 1.0x | 0.38s · 0.7x |
-| `touch-hub` | mtime on a widely-imported interface, content unchanged | **0.40s** · 208x | 83.39s · 1.0x | 82.08s · 1.0x |
-| `edit-body` | a real edit inside a function body | **76.24s** · 1.1x | 85.64s · 1.0x | 84.61s · 1.0x |
-| `edit-comment` | a comment added to a widely-imported interface | **0.38s** · 218x | 82.96s · 1.0x | 82.73s · 1.0x |
-
-<sub>mcpp in its DEFAULT configuration. Linux x86_64 · i9-13900K · gcc 16.1.0 ·
-n=1 · pinned workload `a749e9f`. The opt-in `[build] bmi_schedule = "on"` takes
-`cold` to 35.4s, but it has an unresolved correctness bug on incremental
-rebuilds and is therefore not quoted here — see
-[`bench/README.md`](bench/README.md).</sub>
+| scenario | what changed | **mcpp** `bmi_schedule=on` | mcpp default | cmake | xmake |
+|---|---|---|---|---|---|
+| `cold` | nothing built yet | **35.43s** · 3x | 79.54s · 1.2x | 92.33s · 1.0x | 90.30s · 1.0x |
+| `noop` | nothing at all | **0.16s** · 2x | 0.16s · 1.8x | 0.28s · 1.0x | 0.38s · 0.7x |
+| `touch-hub` | mtime on a widely-imported interface, content unchanged | **0.22s** · 377x | 0.40s · 207.9x | 83.39s · 1.0x | 82.08s · 1.0x |
+| `edit-body` | a real edit inside a function body | **30.17s** · 3x | 76.24s · 1.1x | 85.64s · 1.0x | 84.61s · 1.0x |
+| `edit-comment` | a comment added to a widely-imported interface | **0.18s** · 458x | 0.38s · 217.2x | 82.96s · 1.0x | 82.73s · 1.0x |
+
+<sub>Linux x86_64 · i9-13900K · gcc 16.1.0 · n=1 · pinned workload `a749e9f`.
+**Both mcpp columns are shown because either alone misleads**: the default is
+what you get today, `bmi_schedule = "on"` is one opt-in manifest key. It is
+opt-in because it still has an unresolved correctness bug on incremental
+rebuilds — reproducible on the generated fixture, not on any of the three real
+trees measured — see [`bench/README.md`](bench/README.md) §8b.</sub>
 
 * **`touch-hub` and `edit-comment` are where the day goes.** cmake and xmake
   decide by timestamp and rebuild everything downstream; mcpp compares the BMI

tests/e2e/233_bench_matrix.sh

@@ -292,16 +292,26 @@ if not default:
     raise SystemExit(0)
 
 readme = open(os.path.join(root, "README.md"), encoding="utf-8").read()
-rows = re.findall(r"^\| `([\w-]+)` \| [^|]+ \| \*\*([\d.]+)s\*\* · [\d.]+x \| ([\d.]+)s · 1\.0x",
+# Six columns now: scenario | what changed | schedule=on | default | cmake | xmake.
+# The bolded cell is the SCHEDULE arm and the `1.0x` one is cmake; the default
+# mcpp column sits between them and is checked too — publishing one mcpp column
+# alone understated the engine badly enough to be a defect in its own right.
+rows = re.findall(r"^\| `([\w-]+)` \| [^|]+ \| \*\*([\d.]+)s\*\* · [\d.]+x \| ([\d.]+)s · [\d.]+x \| ([\d.]+)s · 1\.0x",
                   readme, re.M)
 if not rows:
     print("FAIL: the root README benchmark table did not parse — has its shape changed?")
     raise SystemExit(1)
 
 bad = []
-for sc, mcpp, cmake in rows:
-    for engine, claimed in (("mcpp", mcpp), ("cmake", cmake)):
-        have = truth.get(default if engine == "mcpp" else "cmake", {}).get(sc)
+for sc, sched, mcpp, cmake in rows:
+    # The schedule arm is checked too. It is the column a reader's eye goes to,
+    # so an unchecked number there is the most expensive kind to get wrong.
+    for engine, claimed in (("mcpp", mcpp), ("cmake", cmake),
+                            (default + "+schedule=on", sched)):
+        key = ("cmake" if engine == "cmake"
+               else default if engine == "mcpp"
+               else engine)
+        have = truth.get(key, {}).get(sc)
         if have is None or abs(float(claimed) - have) >= 0.01:
             bad.append(f"README {sc}/{engine}={claimed}s but the run says {have}")
 if bad: