中级Intermediate在线实验室Online lab可打印Printable

猴子共和国：换个世界，冠军还会赢吗？

Monkey Republic: Will the Champion Win in Another World?

固定人数和收益，改变一个条件，重复几个种子。把个人得分、全体平均与分享率分开比较。

Fix population and payoffs, change one condition and repeat several seeds. Compare personal score, society mean and share rate separately.

一只猴子赢了一次，不一定在每个世界都赢。拿到有限的实验机会，你怎样用证据支持自己的策略？

One win need not transfer to another world. With a limited experiment budget, how will you support your strategy with evidence?

开始之前Before you start

预先固定实验条件与种子，比较至少两组重复运行，并用数据区分单局冠军和稳定表现。

Fix conditions and seeds in advance, compare repeated runs and use data to distinguish a single winner from consistent performance.

先做记忆课；会算平均数、读百分比。参数由老师操作，不要求每人有设备。

Complete the memory lesson; calculate means and read percentages. The teacher operates parameters; individual devices are unnecessary.

一步一步试试看Try it step by step

先猜（5分钟）：写下问题和评价指标。先固定 Win-Win 收益、九只猴子、100轮、五个种子；第一组动作翻转概率 0，第二组 0.15。Predict (5 min): write a question and metric. Fix Win-Win payoffs, nine citizens, 100 rounds and five seeds. Compare per-action flip probabilities zero and 0.15.
动手（10分钟）：先记预测，再运行两组。每个种子都记录，不挑喜欢的结果。平均香蕉是全体累计香蕉除以猴子数；分享率用全部轮次的动作统计。Run (10 min): predict, then run both conditions. Record every seed. Mean bananas is total accumulated bananas divided by citizens; share rate counts actions across all rounds.
观察（10分钟）：先看种子1，再看五种子的平均；分别比较学生得分、全体平均和分享率。一个指标高，不自动代表另一个也高。Observe (10 min): compare seed one with the five-seed mean. Separate student score, society mean and sharing; improving one metric need not improve another.
挑战（10分钟）：只改一个学生参数，或在新世界测试。若比较世界预设，要记住收益和失误率可能同时变化；要归因于其中一项，必须固定另一项。Challenge (10 min): change one student dial or test another world. Presets may change both rewards and mistakes; hold one fixed to attribute differences to the other.

一起算一个例子Work through one example

固定上述条件：种子1的每猴累计香蕉为148与184；五种子平均约185.91与148.04。这里只描述这一组代码实验，不说明差异必然发生或统计显著。

Under these conditions, seed one gives 148 versus 184 cumulative bananas per citizen; the five-seed means are about 185.91 versus 148.04. This describes this code experiment, not an inevitable or statistically significant effect.

先想一想，再看答案Think first, then check

种子1的方向和五种子平均相反，能删掉种子1吗？同种子能保证改参数后每一轮的配对都完全相同吗？

May we discard seed one because its direction differs from the five-seed mean? Does the same seed guarantee identical pairings after changing a parameter?

查看参考答案Show the expected answer

不能删；种子在运行前选定。同种子能复现同一个配置，但分支可能改变随机数消耗，跨配置不保证每轮配对相同。学习单：A=30,10,20，均值20；B=24,24,24，均值24；A有最高单次，B均值更高。真数据五种子均值185.9111与148.0444，种子1反向。15%指每个动作独立翻转；一对至少一方翻转概率是1−0.85²=27.75%。只看分享率无法唯一算出香蕉，因为还缺每对组合和收益表。

Do not discard it; select seeds before running. A seed repeats one configuration, but changed branches can consume random draws differently, so pairings need not match across configurations. Worksheet: A=30,10,20 has mean 20; B=24,24,24 has mean 24. A has the highest single result, B the higher mean. Actual five-seed means are 185.9111 and 148.0444; seed one reverses direction. Fifteen percent applies to each action; at least one flip in a pair has probability 1−0.85²=27.75%. Share rate alone does not determine bananas without pairing outcomes and payoffs.

这个实验说明了什么What this experiment shows

使用固定版本的真实 Society 计算引擎，但这仍是人工假设下的模型。六滑块不是神经网络，也不是人的性格测量。Shared Well 没有实际水库存或再生。结论不能直接推广到真实社会或评价孩子品格。

This uses a pinned Society engine but remains a model with designed assumptions. Six dials are neither a neural network nor a personality test. Shared Well has no water inventory or regeneration. Results do not directly establish real-society claims or grade children’s character.