测试与复现 / Tests and Reproduction¶
在本题目录编译，示例使用 Windows 可执行文件名。GCC 16.2.0，实际测量选项为 -std=c++17 -O2 -Wall -Wextra .Python 生成器仅依赖标准库。
Compile from this problem directory; the examples use Windows executable names. Measurements use GCC 16.2.0 with -std=c++17 -O2 -Wall -Wextra . The Python generator uses only the standard library.
g++ -std=c++17 -O2 -Wall -Wextra std.cpp -o std.exe
python tests/generate.py --plan tests/plan.json --out-dir data --std std.exe
python tests/generate.py --out-dir selected --case 01 --case 02 --std std.exe
--case 可重复指定；未指定则生成全部配方。未知编号会报错。省略 --std 只生成输入，提供它则运行指定程序生成 .out .同一配方和种子生成相同输入。正式大数据和平台二进制不随包提交；样例已包含。
Repeat --case to select several recipes, or omit it for all recipes. Unknown IDs are rejected. Omit --std for inputs only, or supply it to generate .out files by running the named executable. A fixed recipe and seed reproduce the same input. Large generated data and platform binaries are not distributed; samples are included.
配方 / Recipes¶
| ID | 分组 / Group | 分值 / Score | 目的 / Purpose |
|---|---|---|---|
| 01 | 1 | 10% | chain / literal |
| 02 | 1 | 10% | random / literal |
| 03 | 2 | 10% | chain / literal |
| 04 | 2 | 10% | star / literal |
| 05 | 2 | 10% | random / literal |
| 06 | 3 | 10% | chain / sparse |
| 07 | 3 | 10% | star / sparse |
| 08 | 3 | 10% | binary / zero |
| 09 | 3 | 10% | random / sparse |
| 10 | 3 | 10% | chain / accumulate |
独立验证 / Independent Validation¶
小规模逐点集合模型覆盖全部12操作。正式数据的小中规模使用任意更新；大规模使用全局基值加稀疏点差，并独立用距离判定路径和邻域成员，或保持全零测试任意大范围更新。包含长链、星形、二叉树、随机树、赋零、负值和接近10^18的总和。另有长标记下传回归反例。
Small literal vertex sets cover all 12 operations. Small and medium formal cases use arbitrary updates. Large cases use a global base plus sparse point differences, with independent distance-based membership, or arbitrary broad zero updates. Cases include chains, stars, binary and random trees, zero assignment, negative values, and sums near 10^18. A separate regression checks propagation of large accumulated tags.
独立小输入共 350 组，10份正式配方均重复生成并核对输入限制及分组；另检查6类非法输入。每份程序每个正式输入测时5次，并单独测峰值。
There are 350 independent small inputs. All 10 formal recipes were regenerated and checked against input and group constraints; six invalid-input classes were rejected. Each program has five timing runs per formal input and a separate peak-memory run.
工具和证据 / Tools and Evidence¶
仓库根目录的 check_r35.py 包含独立小模型和输入校验；benchmark_r35.py 重新生成正式数据、交叉检查并测量。先将编译产物放在 build/R35/<slug>-std.exe（有补充程序时也放 <slug>-alternative.exe）。基准支持 --only <slug> ，当前峰值测量依赖 Windows。校验工具的开发断言需正常 Python 模式运行；独立公开评分器不依赖断言。
From the repository root, check_r35.py supplies independent small models and input validation. benchmark_r35.py regenerates, verifies, and measures formal inputs. First place compiled programs at build/R35/<slug>-std.exe and, if applicable, <slug>-alternative.exe . The benchmark accepts --only <slug> ; peak-memory measurement currently requires Windows. Run development validators with normal Python assertions enabled; the standalone public checker does not rely on assertions.
逐次测量与哈希 / Timings and hashes · 环境与时限校准 / Environment and calibration