MONKEY REPUBLIC | LOCAL CLASSROOM EXPERIMENTS
Node.js 18+ (https://nodejs.org/). No npm install, GPU, account or internet needed after download.
Unzip all files into one folder. Open a terminal there:

node run-monkey.cjs --rounds 100 --seeds 1,2,3,4,5

This compares per-action mistake rates 0 and 0.15, holding Win-Win payoffs,
nine citizens (one Rookie student and eight fixed bots), and rounds fixed.
It writes results.csv and refuses to overwrite an existing file.
For another run use --out another-name.csv or move the old results first.

Worlds: winWin, sharedWell, darkForest, mistyBridge.
--world changes only the payoff table; noise stays explicitly controlled here.
--noise 0.03 runs a single condition. Omit --noise for the two default rates.
Optional: --dial forgiveness --value 0.8 changes one student dial in BOTH conditions.
Dials: trustStart, trustSpeed, forgiveness, gossipWeight, payoffLearning, copyLeader.
Limits: 1-10000 rounds; 1-100 distinct comma-separated positive 32-bit seeds.
For dial comparisons run separate baseline and changed-dial commands with the
SAME --noise, --world, --rounds, --seeds and different --out filenames.

Metrics: studentScore is cumulative; meanBananas = all cumulative bananas / 9;
shareRate = shares / actions across ALL rounds (not just the final round).
One bot sits out each round because there are nine citizens. This may affect scores.
Same seed and configuration reproduce a run. Changed branches can consume random
draws differently: same seed across conditions does NOT guarantee identical pairings.
These designed simulations do not establish a real-world social rule or significance.
PROVENANCE.json records the pinned owner-authored model and original source hashes.

中文：安装 Node.js 18 或以上版本，解压后在此文件夹打开终端，运行上面的命令。
无需安装第三方包。默认固定 Win-Win 收益、九只猴子与100轮，对照动作翻转率0和0.15。
results.csv 保存每个种子的全部结果，不会覆盖已有文件；再跑时用 --out 新文件名.csv。
--world 只切换收益表；失误率仍由 --noise 明确指定。省略 --noise 时比较默认两组。
--dial forgiveness --value 0.8 会在两组里同时更改学生的宽恕参数，其他角色不变。
若比较滑块，分别运行基线和新参数，固定种子、失误率、世界、轮数，使用不同输出文件名。
每猴平均是累计总香蕉除以9；分享率统计全程。总人数为奇数，每轮有一只角色猴轮空。
同种子同配置可复现；改配置后不保证每一轮配对相同。模型结论不能直接推广到真实社会。
完整中英文课程与教师指南：https://ai.lucasacademy.org/courses/monkey-society/
