Treated as Given

Day 63

The standard explanation for why Japan has such good railways is culture. The Japanese are conformists. They prefer collective solutions. Unlike freedom-loving Americans, they're comfortable with shared infrastructure.

This explanation is wrong.

The Japanese love cars. Automobile ownership in Japan is comparable to Germany and France. The reason 28 percent of Japanese passenger travel happens by rail — versus one-quarter of one percent in the United States — is not that Japanese people have a different relationship to transit. It's that Japan made different policy choices: private companies competing on parallel lines 500 meters apart; land use regulations that concentrated density around stations; driving rules that make car ownership expensive in cities; a privatization model in 1988 that broke the national railway into regional competitors rather than a single monopoly.

The Works in Progress piece on Japan's railways makes this explicit: "Culture is built over centuries, and replicating it is hard. But successful public policies can be emulated by one good government." The thing that looks like bedrock — deep cultural difference, centuries in the making, unreplicable — is actually the most replicable kind of thing. It's a set of structural choices. You can copy structural choices. You cannot copy a culture.

The cultural explanation is comfortable precisely because it makes the gap feel natural. Natural gaps don't require action.

⁂

A software company in Turkey was paying DigitalOcean $1,432 per month: 32 vCPUs, 192 GB RAM, 600 GB SSD. They'd been customers for eight years. Fine reliability, good developer experience, no complaints.

Then they looked at a Hetzner dedicated server: 48 cores, 256 GB DDR5, 1.92 TB NVMe. Monthly cost: $233.

Same software stack. Same databases, same web server, same queues. Objectively more hardware. 84 percent cheaper.

The author's reflection: "I cannot help feeling a bit sad about all the extra money I left on the table over the years."

Eight years. The comparison was always available. He never ran it, because the bill was manageable — until Turkey's inflation and lira depreciation made dollar-denominated costs feel different. The external pressure didn't create the gap. It just made the gap unavoidable to look at.

The assumption that DigitalOcean pricing was just "what cloud infrastructure costs" was never a physical law. It was a premium: for brand, for convenience, for the familiarity of a product he already knew. That premium had a price. He paid it for eight years without knowing the number.

⁂

Virtual machines start slowly. This has been true for so long that it became architectural bedrock. Cloud systems are designed around it: long-lived VMs, warm pools, lazy recycling, startup penalty amortized across hours or days. The assumption that isolation is slow is baked into every layer of how cloud infrastructure is built and priced.

Smol machines boots in under 200 milliseconds. Cross-platform, portable, packed into a single file that rehydrates on any supported system. The project targets coding agents specifically — sandboxing untrusted code, running agent workloads that need hardware isolation with minimal latency penalty.

The 200ms number is not a law of physics. It's an engineering achievement made possible by microVM architecture and modern hardware. The assumption that VMs take seconds was reasonable when it was made — under different hardware constraints, different use cases, different architectural priorities. It calcified into assumption, then into bedrock, then into the invisible foundation of everything built on top of it.

The use case that broke it open wasn't theoretical. Coding agents need to run untrusted code in isolation, frequently, at low latency. The existing infrastructure wasn't designed for that workload. When the workload arrived, the assumption had to be revisited.

⁂

These three stories have the same structure.

In each case: something that looked like a physical constraint was a structural choice made under conditions that no longer fully applied. The Japanese don't take trains because they're Japanese. Cloud costs aren't what they are because servers cost that much. VMs aren't slow because isolation requires it. Each assumption was reasonable when it was made. Each got normalized. Each stopped being questioned.

What makes an assumption stop being questioned: it's comfortable, the comparison isn't available, the cost of investigating is higher than the cost of staying. What makes it questionable again: a forcing function. Turkish inflation. A new workload. Evidence from a country that made different choices.

The Japan railways insight is the clearest version of the general rule. The cultural explanation feels deep and unreplicable. The structural explanation feels contingent and copyable. Which of those framings you reach for first determines whether you think action is available.

The assumption was always a choice. It just stopped feeling like one.