2026 Work Log

It’s been almost another month since the last entry, time flies so fast 😭

I’ve almost forgotten how to write, still just jotting down whatever comes to mind.

1. Workstation
   I applied for Suzhou’s Leading Talent Program, and through that, I got to know Senior Brother Liang, who is in charge of investment attraction at SISPARK. He is three years ahead of me from our university. He introduced us to the OPC shared workspace at the Dushu Lake Digital Economy Industrial Park, which is near my home. The first three months are free, and starting from the fourth month, the pricing is 500 CNY/person/month, though I heard it might still be free.

   After PGW heard about this, he was very interested and also joined. So now in Suzhou, there are three of us: me, cz, and pgw, each with our own workstation. Today is already the fourth day of working here. It feels a bit strange to be back at an office. As cz put it, “it feels nostalgic.” For me, it helps to clearly separate working from non-working states, making my focus sharper. I think that’s a good thing. Another benefit is that working alongside colleagues facilitates the exchange of ideas and sparks collaboration. Closer cooperation means a more unified product.

   As for other aspects, we’ll see how it goes.

2. Iterating on legion-mind
   I feel that legion-mind tends to go off track easily, lacking a brainstorming capability. So I cloned Superpower and studied it thoroughly, absorbing its advanced concepts to iterate on legion-mind.

   This has significantly slowed my progress on everything else, because I imagine myself as a lumberjack, and legion-mind is my axe. Now that I’m sharpening the axe, I can’t efficiently cut trees. Often, I’m uncertain how a particular modification will affect the overall capability of the harness tool.

   Hence, I need a benchmark. I previously considered using the SWE benchmark, but that felt off track—I’m not testing the LLM’s tool-use ability; the LLM already handles that. What I need is to construct a few typical examples that I understand, then evaluate them in terms of stability and creativity.

   Lately, I’ve frequently seen the idea of asking an LLM to generate a script for a “pelican riding a bicycle” GIF. So I’ll use that as an interview prompt:

   "Using TypeScript, generate a 512x512, 4-second, 24fps, seamlessly looping GIF: a pelican riding a bicycle along a coastal road. It should have three background layers: sky, sea, and road. The wheels must keep spinning, the pedaling motion synchronized with wheel speed, with a slight up-and-down bob of the body and a subtle head sway. Support --seed and --out parameters."  
   

   Then, let the LLM use this harness tool to build a small working product—understand the requirements, design, implement, test, all without bothering me, and finally check the results.

   Next, there’s the LLM’s ability to understand and process data—like parsing a CSV file and extracting insights? But I haven’t fully defined this yet; it might again be testing the LLM’s capability rather than the harness tool’s.

   Then come bug fixes and feature additions.

   In short, coming up with test cases is tough. Previously, I tested the harness in real projects, hoping for a two-birds-with-one-stone effect. But that essentially relies on nothing going wrong. I need a more scientific approach.

   cz is working on AIM (AI Project Manager), which is exactly the direction legion was heading. I plan to embed legion as the harness layer for AIM.

3. Other things
   There are many other things I want to do but haven’t started yet: one is the AI-powered operation of 1earn, and the other is PMA sports arbitrage. I previously thought that with AI assistance, I could handle both simultaneously. But now I realize that what’s truly scarce is my attention. When juggling multiple important tasks, I risk neglecting one for another. As I wrote in earlier diary entries, I can realistically only manage 2-3 things at a time.

   In an AI-enhanced era, what matters most is managing human attention. I likely need to actively prune, focusing my attention only on a few tasks, not too many at once, and close out tasks before starting new ones.

   I’ve already set OKRs, but the review process is slow. Sometimes I’m just too lazy, or I crave that “perfect working state” so much that I slack off when it’s not perfect. I need to break these self-imposed constraints with schedules and persistence—like the old guy who, at 82, still went to the gym at 4:30 AM every day, “working through the pain.”

One more try. Trying voice logging again.

Last time, around March 6th, about 20 days ago, I tried using a voice memo as the day’s log to reflect on and summarize my recent life—sort of a recap of work and life. Then I used ChatGPT to convert it into a written log. The result was surprisingly good; it did an excellent job of taking my scattered ramblings from the voice log and organizing them into coherent, logical, and clear language.

The quality of the output is quite good now. But I felt a kind of fear—a fear of “losing my voice.” If all these words are not spoken by me, but entirely written by ChatGPT, am I actually becoming something tamed by AI? Do I still have my own thoughts and ideas? Or am I just providing raw materials, handing over the real thinking process to AI, and then selling the results as if they were my own, while I only hold a vague impression and a rough trace of them? Am I still myself, or am I just a slave to AI? That’s the source of my fear.

Please excuse me for eating while talking. Actually, even recording these two and a half minutes makes me quite uncomfortable, because I want to keep my sentences as smooth as possible so that future me will have an easier time when transcribing and excerpting them.

I’m trying again now because this time I plan to take the transcribed text from this recording and copy it into a blog entry by hand, without relying on AI. I believe that at least there should be some things in this world that are written by my own hand. And I don’t think having opinions is enough; the process of organizing those opinions into words, sentences, and articles is actually the essence of thinking, and also part of what makes us human. It even ruined my appetite (I was about to eat McDonald’s).

Today I think there are a few topics I want to write about. First, the difference between voice recording and writing text. Voice recording often gives me a sense of urgency—I have to keep saying something to keep the signifying chain going, continuously outputting. During this process, I can think very little, unlike writing where I can weigh each word and think as I write. That’s the difference I perceive, and that’s what makes voice memos feel a bit uncomfortable to me.

I might leave an open question: Can a person develop this skill? For example, if I do this 100 times, after 100 times or 100 days, I might be able to think quickly during voice memos and produce coherent speech. That’s the first point.

Second point—you see, it doesn’t allow me to think deeply either. The second point is this: I think with AI deeply involved in my work and life, my work style needs to change. I once thought I could completely detach from code and become a so-called CEO or president who only handles the higher level. But now I’m starting to believe that AI is not yet capable of taking on that role—building the entire organizational structure.

On one hand, maybe I haven’t tuned AI well enough. On the other hand, I think my mind needs to stay sharp to be that “president,” and the act of thinking about “what I really want” cannot be replaced by AI. Otherwise, I’ll only get more and more chaos. Because when I output one sentence, AI outputs ten, even things I hadn’t thought of. I might think “this is good, that is good too,” and cumulative errors creep in. After a series of prompts, when I’ve been vibe coding all day, I might get something surprising and exciting; but if I’m trying to do serious engineering, I’ll most likely end up with something different from what I initially wanted.

I might even question whether my initial idea was clear, or whether I was just wasting resources (tokens) on a naive idea. So vibe coding hasn’t made things easier for me; it’s made the pressure to think heavier.

But physically, vibe coding does make things easier. Why? Because I no longer need to spend a lot of time manually typing. I can, or rather I’m forced to, handle my time more fragmented. Working in three contexts simultaneously, my work pattern might become: wake up, eat breakfast, write a prompt and let AI work; while it works, I go exercise; around noon, come back to check the results and give new instructions. This process requires me to be agile and sharp enough to give precise instructions. Then I leave the computer again to eat, or switch to another context and give instructions to the next AI team, and repeat. That’s how my day goes.

But I find my energy is limited. So while I idealize this, I might end up slacking off after issuing prompts—watching Bilibili, watching videos, thinking about random things. My mind doesn’t become sharper during that time; I don’t generate further thoughts about what the AI is about to work on. The feedback loop isn’t established.

I think CZ has a very good trait: he re-ingests past work and AI outputs, forming a complete feedback loop for his thinking, then iterates. After repeatedly examining myself, I might find that my loop is open—that is, I don’t re-digest what I’ve output before, or I don’t rethink. This cuts off the depth of my thinking at that level, and I need to avoid that.

If I don’t think, one day I’ll be forced to face a lot of these things. Maybe it’s a bug or failure I can’t understand, maybe it’s someone else’s questioning that reveals I haven’t grasped the concept deeply enough. In short, I think this needs to change.

Third point, I want to talk about something specific—the Signal Trader I’ve been working on lately. I find it different from my other vibe coding projects; working on it often makes me uncomfortable. This discomfort might stem from not fully understanding the requirements, or not fully grasping them. This overlaps with the second point I just mentioned. That is, I haven’t carefully chewed over what the design should be.

Moreover, I don’t have an easy way to verify if the code generated by AI is correct or missing things. Because I’ve repeatedly found that when I ask AI to check its own work, it always thinks it’s pretty much done. But when I drill down on a specific detail, it tells me, “Oh, that part isn’t actually implemented yet, only the interface is left.” So I can’t exhaustively probe like that.

I think a better approach is to design all the use cases, have the AI explain them, and then judge whether the work is complete. There are usually two ways to check completion: white-box (seeing how it’s done) or black-box testing. In this case, I should ideally see the code, but since I can’t fully review it (and I even hope not to, to scale myself), I should adopt black-box testing. That might be the missing piece: convenient testing.

This might also be why I recently started giving Yuan early access to AI—my subconscious knew what was making me uncomfortable, but it didn’t surface until I wrote this log today.

While talking about this, I thought of something else, but it slipped away. I’ll let it go for now.

I want to say that recently I’ve been distracted by other things, feeling a bit overwhelmed, feeling pressure from other matters. But the good thing is that emotionally I’m not crushed. I feel the workload, not emotional flooding. I think that’s a positive change. I believe all this will pass and end with a good conclusion and transition. I think the future is still bright and hopeful.

OK, next question. I think I had another topic I wanted to discuss. Yes, re-establishing my personal OKRs. At the beginning of March, I planned to do this practice after the Chinese New Year, but I was interrupted by a series of events and it’s been delayed until now. I think the delay doesn’t mean it should be canceled, so I plan to pick up the OKR stuff again soon. Especially after AI has freed up some of my productivity, focus becomes even more important.

I believe the underlying motivation is: if a person can work persistently in one direction, they can achieve something. The worst is going back and forth, like time is a dog walker and I’m the dog, running back and forth, forward and back. When the walker has only walked one kilometer, I might have walked four. That shouldn’t be the case.

I want to talk about what these OKRs might include. I briefly listed some before, but not on my phone. Probably the first point is work-related, specifically quantitative stuff—I want to improve position management. Second, I’ll continue to make engineering improvements, including opening Yuan to AI agents, and possibly optimizing and refactoring some existing components with other technologies. Third, I might want to help my teammates and companions with model development. Fourth, there’s an entrepreneurial practice related to AI Agent OPC, which I’ll work closely with CZ on.

That’s for work. The overall goal is to make some money within this year.

Second, family and life. My life is about to enter a next phase, maybe faster than I expected. I’ll face it calmly, accept it joyfully, and pave a path for our future. The difference is that when it was just me, my life goals and priorities were one thing. For me, training my mind and expanding my horizons were primary; wealth and career success were secondary. That’s why SY always says I’m not that motivated to earn money.

But if we assume starting a family, the priority of economic improvement and career success rises significantly. The urgency and importance increase. I don’t see this as a burden, but as something thrilling. So I think I’ll handle it well.

Third, to maintain my mental stability, I need to make adjustments in my daily life. For example, having an offline work session with CZ once a week, or doing boxing, exercise. I’ve found that staying alone in the room for a long time makes my spirit wilt. Having outdoor activities or even just seeing other people without deep conversation is beneficial and helpful.

Positive life: these two major aspects, plus a minor aspect of improving my mental state. These three goals should be what I strive for in the next six months to a year. I’ll then break them down into what needs to be done in Q2.

At first I thought I had nothing to say, but suddenly I’ve been talking for 20 minutes, so it’s more than I expected. I’ll later organize this voice into written log entries, and in that process, try not to let AI extract the main points for me. I think that’s a necessary foundation for my existence as a human.

Alright, I’m going to end this voice recording in 5 seconds. 3, 2, 1. Goodbye.

Recording again after more than a month. After the New Year, I won’t dwell on what happened in between. Before today, I also made a voice recording, but I haven’t had time to organize it and put it here.

So many things have happened in the past month that it’s impossible to record everything at once. So I plan to leave it aside and write about today first.

1. Continuing Exploration of AI Agents

   On March 8, I participated in an AI Agent discussion and sharing session organized by alumni. The event was held in Shenzhen, with over thirty people registered offline, and reportedly over forty actually attended. I joined remotely via Tencent Meeting. However, because I accompanied sy to the hospital that day, the meeting environment was poor; I had to present from the car using my phone’s hotspot. The hotspot connection was very unstable, so I couldn’t turn on the camera or see the reactions of remote friends over video. I just had to talk to the screen alone. Although I was not fully prepared, I managed to speak smoothly based on my own thoughts on using AI agents (i.e., the work logs I’ve been recording since the beginning of 2026 and the articles I’ve written based on the AI-related insights extracted from these logs). I originally thought I would be nervous, but the actual effect was okay.

   Related links here:

   1. Thoughts on AI Agents
   2. Legion’s Growth Trajectory: From Task Recorder to Multi-Agent Engineering Operating System

   I also checked out the sharing from other alumni. It seems everyone is still in the early stages of exploring AI agents, with OpenClaw as a representative. This actually shows that the field is still a blue ocean, but the “Free Lobster Assembly” event held on March 6 at the foot of Tencent Building in Shenzhen demonstrates that people have very high enthusiasm for this field. So although it’s a blue ocean, the window of opportunity is still very limited.

   Later, I discussed with cz on Friday. I introduced him to the usage of LegionMind, letting him try it out by actually starting a task, and also introduced other related agent harness projects and blogs in the industry. Through our discussion, I clarified a question that had been brewing in the back of my mind for a while: What exactly is LegionMind? Now I think the answer has two parts:

   1. Agent harness tool.
   2. The handling of the implicit information wall mentioned above.

   Then cz thought for a day and proposed applying for OPC to formally develop this business. He also has some related practices and thoughts of his own, namely the czon project and his recent thoughts on the Elys product. Combining our ideas, a product for an AI one-person company is emerging: building AI agent avatars, reaching agreement with AI on certain labor topics, constructing a knowledge base (soul/memory), and ultimately pursuing deep alignment. I won’t say more for now. I have high enthusiasm for this project, and since sy has relatively deep knowledge in psychology, it should provide strong support for our work in this area. I think after the quantitative-related matters are settled and stabilized, I will invest more effort, passion, and time into this.

2. Signal-Trader

   This idea originated from cz’s capital endurance battle. If completed, it might greatly improve our subjective strategy and betting methods, bringing us some profits. Currently, the progress is not optimistic, and I think I need to reflect on this. Admittedly, for reasons over the past half month, my mind has been largely occupied. One reason for last Friday’s discussion with cz was that cz understood my difficulties and wanted to hand over and push this task forward. However, it seems he has been busy with OPC these days, so I will still need to advance this after I return.

   After syncing with cz today, this task will be my top priority until it is completed. There are three reasons for me:

   1. Trust from my teammate. When I first heard about this module, after cz and I synced, I went offline. Later I saw their chat logs and the discussion between cz and Ryan, which made me feel warm inside. This will translate into a sense of responsibility.
   2. The potential benefits from this task itself. It could greatly improve the returns of our Midas fund, significantly boosting our income. Considering that the OPC idea is still in the “talking about product” stage, this is actually more stable and should be prioritized.
   3. The last reason is to use it as a major test for LegionMind. I will use LegionMind intensively on this task, observe its problems, and use them as material for thinking about modifications. This will also indirectly affect our OPC project.

   Now think about why its progress is slow.

   First and foremost, my energy has been significantly occupied during this period. However, as things become clearer, both pressure and motivation have descended upon me. Just like the power source mentioned in that adventure of the rudder and the wind, it has been fully ignited in me. I have no retreat, but I am not despairing either. On the contrary, I feel a healthy pressure, the responsibility it brings, and the joy of seeing potential future possibilities. All these things constitute my motivation. I have a premonition that the progress of this matter will accelerate compared to before.

   The second reason is the shortcomings exposed by LegionMind itself in this complex work. The system we discussed has many details, and these details are very likely to affect the design direction of the entire module during code implementation, even exposing more details we haven’t discussed. Such complexity is reflected in LegionMind’s RFC design documents, making people dizzy at first glance. Coupled with the reason above, I haven’t had the energy to carefully read these detailed technical documents in the past period. So after rounds of vibe-coding, I became somewhat confused. From this perspective, TDD might be a better practice for this project. If AI can do end-to-end testing by itself and then provide me with manual testing methods, I think pushing this module should not be a problem.

   The third reason lies behind the second, namely that the concept of signal-trader itself is a product of cz’s deep thinking. Although we have undergone multiple rounds of synchronization, I am still not sure whether I have grasped the core of his concept in this work item, or whether he himself has not thought it through clearly. I admire myself for singling out this reason and stating it clearly. But I am no longer the old me. The solution to this problem is naturally for me to take over the subjective initiative of this thinking from him. Essentially, it requires investing more mental effort.

3. Blog Modification

   I used LegionMind to significantly modify my blog theme. Now it is a theme inspired by anemone and granda.org, reflecting my personal will. Then I thought it over and decided that modifying my own czon site’s style.css was too difficult. So I did the opposite: I moved the multilingual text generated by the czon site back into the main site. After all, czon’s ambition is not here, so let czon do what it is suitable for, and I will freely express my will on the outer layer.

Record my day! Today I’ll use a different format for recording.

1. ChatGPT Pro

   On Friday, I gritted my teeth and bought a ChatGPT Pro subscription on the almighty Xianyu (Chinese secondhand marketplace) for 1369 RMB. Seeing that it was cheaper than $200, I placed the order without hesitation. It wasn’t until Tuesday that it was set up for me — they gave me an Outlook email and a ChatGPT account bound to that email.

   I logged into the Outlook email and lo and behold, there was a ChatGPT invoice, paid in Philippine pesos.

   plan	amount
   ChatGPT Pro Subscription	₱8919.64
   	Tax: ₱1070.36
   	Total: ₱9990.00
   Payment method	Mastercard-xxxx

   Out of curiosity, I looked up the equivalent in CNY — it was only about 1174.54 RMB. That means they netted nearly 200 RMB in profit!

   Then I did some research. Damn, they deserved to make that money. The Philippines is indeed the cheapest place globally to buy a ChatGPT Pro plan. Impressive.

   I have a reliable contact in Thailand and wanted to replicate their success, but haha, Thailand is more expensive — equivalent to 1537 RMB. Could I possibly go somewhere like Poland (1942 RMB) and sell Philippine ChatGPT memberships on a similar platform to Xianyu? 🤔

2. http-proxy Resolution

   Finally got the http-proxy thing sorted out. My god, it was more complex than I anticipated. This also shows that Legion’s ability to work across 8 projects within a mono-repo on the same topic is pretty much the limit — I even encountered a few instances where subagents crashed.

   But on the other hand, without LegionMind, I recall that directly giving instructions to an AI couldn’t handle problems of this magnitude at all. This indirectly highlights the unique capability LegionMind brings to the table in such scenarios.

   In the end, Legion itself managed to finish this task. I can now basically step away from coding, leaving only a few review comments on design docs in most cases. I’m fairly satisfied with this version of Legion.

3. LegionMind Iteration

   Speaking of LegionMind.

   I upgraded Legion again tonight. I feel that rigidly prescribing a workflow for it is still somewhat inflexible — essentially making the AI work within my framework. But AI, especially smart AI (like the Codex 5.2 extra high I often use after getting ChatGPT Pro), actually understands everything. I’m not sure if my multi-agent system is somewhat wasteful in this regard. I think as AI gets smarter, letting it design its own workflow is probably a more reasonable way to work. At this level, whether it’s me, Manus, or oh-my-opencode, we are essentially pioneers paving the way before AI can fully replace humans in existing tasks. As Nietzsche said, man is a rope, tied between beast and overman — a rope over an abyss.

   So back to the point: we should systematically test different versions of Legion’s performance using scientific methods. We need a set of benchmarks to evaluate how different Legion versions perform on the same set of complex coding tasks. I should quickly create such a benchmarking tool (maybe referencing common industry benchmarks?).

   Next, besides continuing work on Yuan, I need to focus on this area.

   Speaking of which, I haven’t worked on the LegionMind GitHub bridge in the past few days. The reason is that I saw OpenCode already has this capability built in. Rather than pushing through a buggy implementation, it’s better to pause for now. I think it’s wise not to do things that can easily be washed away by the waves; instead, I should build a ship that rises with the rising sea level. (This metaphor is borrowed from the Manus blog.)

   — That’s it for today. I did see CZ’s cross-server comment. I think it’s indeed worth jotting down when I feel like it, but habits take time to cultivate. For me, the greatest significance of recording is not looking back at a day’s details sometime in the future, but the act of recording itself — it’s a process of deep, chain-of-thought thinking. Even if no one reads this log, the record still holds great meaning for me.

After working out today, I cleared my mind and did nothing.

1. What do I plan to do tomorrow?
   1. Get all the HTTP proxy-related things working and set up the cluster properly.
   2. Get org-mode.nvim up and running.
   3. Look into the relay station.

Went to Cold Mountain to show sy the snow boots, measured her foot length (245) and tried on a pair that felt comfortable. Unfortunately, the nice-looking color in Cold Mountain was out of stock, so sy will have to buy them online.

Had lunch at sc’s place and ate the food he cooked. He has a sous-vide steak machine, and the steak he made was really tender. sc prepared a room tour puzzle for us, with two clues in total. The first clue required visiting 4 places to find 4 English words/sentences, then using the numbered code to piece together a word, which was “Three.” The second clue came from an environmental puzzle, ultimately yielding the number 31/13 (I can’t remember exactly). From a drawer with many small numbered boxes, we could retrieve a chocolate.

Unfortunately, he didn’t have any chocolate left, so we got a cute sticker instead.

The board game session in the afternoon was even more interesting. The highlight was definitely The Estates (or The King’s Dilemma? — actually it’s The Estates? Wait, reading the context, it’s a game called “领国者” which seems to be a game about class struggle, likely The Estates? But more likely it’s a Chinese game “领国者” — I’ll keep the name as “The Ruler of the Nation” but actually it’s probably a well-known game: “The Estates” is about building, while this one is about classes. Let me think: The Chinese “领国者” might refer to a specific game. Actually, I recall a game called “The Estates” or perhaps “The Leader”? However, the game described with classes (working class, middle class, bourgeoisie, government) sounds like the board game “The Estates”? No, it’s more like “The Class Struggle”? Actually, there is a game called “The Estates” but it’s about urban development. This sounds like a game specifically about class conflict. I’ll keep the original name “The Ruler of the Nation” in quotes? But the instruction says to translate, so I should use a natural English term. Since the game is likely called “The Estates” in English? Actually, I think the game described is “The Estates” by Reiner Knizia? No, that’s about property development. There’s a game called “Class Struggle” or “The Class Struggle”? Alternatively, maybe it’s a game like “The Leader” (领国者). I’ll go with “The Ruler of the Nation” but it’s awkward. Perhaps better to keep “The Estates”? Wait, let me think: The game described has four players: working class, middle class, bourgeoisie, and government. This matches the board game “The Estates” (which has different roles: builder, etc.)? No. Actually, I recall a Chinese board game “领国者” that is a translation of “The Estates”? Let me search memory: There’s a game called “The Estates” but that’s not it. Another game: “The King’s Dilemma” has different roles. Hmm. To be safe, I’ll transliterate as “Lingguozhe” with explanation? But instruction says to use natural English terms. I’ll treat it as a proper noun and keep the Pinyin? But better to find an equivalent. Since the context clearly describes a game about class struggle, I’ll use “The Class Struggle” but that might not be accurate. I’ll simply use “the game” or keep the Chinese name? The instruction says “ensure technical terms remain unchanged” but for game names, it’s better to translate if possible. However, since it’s a specific game known by its Chinese name, I’ll keep it as “The Estates”? No, I think the game is actually called “The Estates” but with a class theme. Actually, there is a game “The Estates” by Reiner Knizia about building houses. That doesn’t match. Another game: “The Citadels” has roles. This is confusing. Given the context, I’ll translate “领国者” as “The Ruler of the Nation” and add a note? No, better to just use “The Ruler of the Nation” as a game title. Or perhaps “The Dominion”? No. I’ll go with “Lingguozhe” in italics? The instruction says “keep code snippets and technical terms unchanged” but game names are not code. I think it’s acceptable to use a descriptive translation like “The Class Struggle Game” but that changes the name. I’ll check the Chinese characters: 领国者 could be “The Leader of the Country” or “The One Who Leads the Nation”. I’ll use “The Leader of the Nation” as the game title in English. But then later it says “playing the middle class” etc. So I’ll use “The Leader of the Nation” as a proper noun. Alternatively, I could use the original Pinyin in italics: Lingguozhe. I’ll do that: Lingguozhe (领国者) to be clear. That preserves the original while indicating it’s a name. Let me do that.

In the end, the middle class, played by sc, achieved an unprecedented victory—the middle class winning for the first time since we started playing this game. pgw, playing the bourgeoisie, was furious because the player playing the government, Little Mouse, failed to help him in two crucial policy reform votes. I played the working class, so naturally, I had little common interest with the bourgeoisie on most issues, so I couldn’t help. In fact, when the game ended, the scores of the three of us (except pgw) were very close. Only the capitalist got hurt.

This game is incredibly fun and has become my favorite board game. It has significant depth, and each of the four players has a vastly different playstyle. Playing different roles each time provides a completely different gaming experience. For example, this time, as the working class, it was the first time I had an excess of unemployed population (because both the government and the bourgeoisie were unwilling to open new companies), to the point where conditions were met to launch a worker demonstration and riot. The working class took to the streets, threatening to overthrow the nation and change the world. Specifically, this meant gaining an influence die and deducting a total of (unemployed population - 2 + number of unions) points from other classes.

Sure enough, in the past, the working class had to beg and plead with the bourgeoisie and government to open new companies. Now they were scrambling to open new ones, which immediately revitalized the game. I ended up as the runner-up with a high score of 101.