运行清单（provenance manifest）——记录环境、参数、种子与输入输出的哈希

# 把一次网络分析需要复算的信息写进 JSON：Python 与关键包版本、平台、参数、随机种子、
# 输入文件与输出结果的 SHA-256。然后用同样的设置重跑一次，比较输出哈希是否一致。
# 解读：哈希一致说明在同一环境下可复算；换机器或升级包后若哈希改变，可据清单逐项排查。
# 浮点运算在不同平台或并行设置下可能有末位差异，此时应比较数值容差而非字节哈希。
import hashlib, json, platform, sys, tempfile
from pathlib import Path
import numpy as np
import networkx as nx
import scipy

def sha256(path):
    return hashlib.sha256(Path(path).read_bytes()).hexdigest()

def run_analysis(edge_file, out_file, seed, resolution):
    G = nx.read_edgelist(edge_file, nodetype=int)
    comms = nx.community.louvain_communities(G, seed=seed, resolution=resolution)
    Q = nx.community.modularity(G, comms)
    labels = {v: i for i, c in enumerate(sorted(comms, key=min)) for v in c}
    Path(out_file).write_text(json.dumps({"Q": round(Q, 10), "labels": [labels[v] for v in sorted(G)]}))
    return Q

workdir = Path(tempfile.mkdtemp())
edge_file = workdir / "network.edgelist"                      # 模拟的“原始数据”
nx.write_edgelist(nx.random_partition_graph([20, 20, 20], 0.3, 0.03, seed=5), edge_file, data=False)

params = {"algorithm": "louvain", "resolution": 1.0, "seed": 20261115}
hashes = []
for run in range(2):
    out = workdir / f"result_run{run}.json"
    Q = run_analysis(edge_file, out, params["seed"], params["resolution"])
    hashes.append(sha256(out))

manifest = {
    "python": sys.version.split()[0], "platform": platform.platform(),
    "packages": {"networkx": nx.__version__, "numpy": np.__version__, "scipy": scipy.__version__},
    "parameters": params,
    "inputs": {edge_file.name: sha256(edge_file)},
    "outputs": {"result_run0.json": hashes[0]},
    "modularity_Q": round(Q, 4),
}
(workdir / "manifest.json").write_text(json.dumps(manifest, indent=2, ensure_ascii=False))
print(json.dumps(manifest, indent=2, ensure_ascii=False))
print("两次运行输出哈希一致：", hashes[0] == hashes[1])