数据生成与检验 / Test recipes¶
仅需 C++17 与 Python 3 标准库。在本题目录执行： Only C++17 and the Python 3 standard library are needed. Run in this problem directory:
g++ -std=c++17 -O2 -Wall -Wextra std.cpp -o std.exe
python tests/generate.py --out-dir ../../../build/data/digging-the-subway --std ./std.exe
python tests/validate.py ../../../build/data/digging-the-subway/01.in
plan.json 列出固定种子、用途和10份各10%的正式文件。--plan 指定其他计划；可重复 --case 01 --case 02 选择编号；省略 --std 只生成输入。文本使用LF换行。生成器核验参考程序结果；构造题允许任意合法最优输出。复现环境为 Python 3.12.14。
plan.json lists seeds, purposes, and ten formal files worth 10% each. Use --plan for another plan, repeat --case to select files, and omit --std for input generation only. Text uses LF line endings. The generator checks reference results; constructive answers may use any optimal valid output. The reference environment uses Python 3.12.14.
覆盖 / Coverage¶
直接路径并集枚举；原贪心反例、最优集合不嵌套、千点链/星形/平衡/随机树；星形两类叶子枚举与单源边际收益独立核验。 / Literal path unions; greedy and non-nesting regressions; 1000-vertex chains, stars, balanced and random trees; independent star leaf counts and single-source marginal gains.
仓库检验 / Repository checks¶
独立小模型工具核验五题，包含输入拒绝检查；测量工具重复生成数据、验证答案、记录五次耗时，并另测整进程峰值内存。先将程序编译至仓库 build/R38/<slug>-std.exe ；传送门的补充程序编译为 nether-portal-alternative-deadline.exe . 在仓库根目录执行：
The independent-model tool checks all five problems and rejects invalid inputs. The benchmark tool regenerates inputs, checks answers, records five timings, and separately measures whole-process peak memory. Compile programs into the repository's build/R38/<slug>-std.exe , plus nether-portal-alternative-deadline.exe for the portal alternative. Run from the repository root:
python tools/check_r38.py
python tools/benchmark_r38.py --only digging-the-subway
逐次耗时、输入输出和源码哈希 / Timings and hashes。时限取标准程序最慢中位数的两倍、向上取整至0.5秒；更换环境需重新校准。
The limit is twice the slowest reference median, rounded up to 0.5 seconds; recalibrate in another environment.