问HN：系统可以在没有内存的情况下正常运行吗？
我一直在思考，记忆是否实际上是正确性的运行时需求，还是仅仅是一种训练信号，其效果在不再需要回忆时依然存在。
在实践中，许多分布式系统因将状态视为需要协商的对象而非需要强制执行的对象而积累了复杂性。当系统试图调和过去的状态以便在当前采取行动时，往往会产生重试、调和逻辑和协调开销，以保持一致性。
我一直在尝试一种以强制执行为优先的模式，其中每个交互都被视为首次请求，无效状态被拒绝而不是修复，状态由请求本身明确携带。我整理了一个最小的参考实现，以使这个想法更具具体性。
我很想知道人们认为这个模型在实践中在哪些方面会失效，以及哪些类型的系统真正需要叙述状态而非局部约束强制执行。
https://github.com/SvengsFuture/stateless-substrate
查看原文
I’ve been thinking about whether memory is actually a runtime requirement for correctness, or just a training signal whose effects persist even when recall is no longer needed.<p>In practice, many distributed systems accumulate complexity because they treat state as something to negotiate with rather than something to enforce against. When systems attempt to reconcile past states in order to act in the present, they often generate retries, reconciliation logic, and coordination overhead just to stay consistent.<p>I’ve been experimenting with an enforcement-first pattern where every interaction is treated as a first-time request, invalid states are refused rather than repaired, and state is carried explicitly by the request itself. I put together a minimal reference implementation to make the idea concrete.<p>I’m curious where people think this model breaks down in practice, and which classes of systems truly require narrative state rather than local constraint enforcement.<p>https://github.com/SvengsFuture/stateless-substrate