Why I Keep Research as an Archive, Not a Feed

In research, the thing that disappears most easily is not the conclusion. It is the path that led to it.

A few months later, a conclusion may remain — rates may fall, a company looked attractive, a system design seemed better — while the evidence, assumptions, and conditions that shaped it have disappeared. The sentence in a notes app, the useful conversation, the code behind an analysis, and the sources left open in a browser have all ended up somewhere else.

I did not build The Archive to publish more often. I built it so that a thought I had once could still be examined, revised, and reused later.

Keep the reasoning, not just the conclusion

Suppose I record only the sentence, “Interest rates are likely to fall.” Months later, I can check whether the forecast was correct, but little else. A real decision requires more context:

• Which inflation and employment data did I use?
• What was already reflected in market prices?
• Which assumptions made a rate cut appear more likely?
• How did that view lead to a decision about portfolio duration?
• What new evidence would make me change my mind?

Those questions make a wrong outcome useful. They help distinguish poor reasoning from a reasonable decision followed by bad luck. A good outcome, on the other hand, does not automatically validate weak reasoning.

Being wrong is not the part that bothers me most. What bothers me more is reaching a point where I can no longer explain why I made the decision in the first place.

That is why these articles try to keep the following distinct:

• Question
• Facts and data
• Assumptions
• Interpretation and reasoning
• Conclusion
• Conditions for revision

The purpose is less to display finished knowledge than to preserve the path by which a conclusion was formed.

Why put different fields in one place?

Investing, AI, optimization, mathematics, and coding look like separate fields. In practice, they repeatedly return to a similar problem:

How do we interpret incomplete information and make decisions under real-world constraints?

In investing, we infer an uncertain future from incomplete information and turn that view into a position. In AI, we decide how probabilistic model outputs should be controlled and used inside real workflows. Optimization asks how to choose among conflicting objectives under constraints such as cost, time, capacity, or risk. Mathematics provides the language of probability, estimation, and optimization. Code makes those ideas executable and testable.

The point is not that all disciplines are the same. It is that they often share a decision structure. Keeping them in one archive makes it easier to preserve the connections that disappear when each field is treated as an isolated folder.

Public documents put useful pressure on thinking

Private notes tolerate phrases such as “I know what I mean.” Public writing does not. A claim must be stated clearly, facts must be separated from interpretation, and numbers and sources must be checked again.

That pressure is uncomfortable but useful. Publishing something is not a declaration that I am right. It is a way of making the basis of my thinking inspectable by someone else. Other readers may find a counterexample, point out missing data, challenge an assumption, or expose an error. That scrutiny makes the research more legible.

Research should not be frozen after publication

I want these pieces to behave more like versioned research notes than finished artifacts frozen at publication. New data may change a conclusion. An explanation may need correction. Significant changes should retain their date and context, and a changed conclusion is itself useful information.

A durable research note is not one that never changes. It is one that can survive revision.

Written for people, but not only for people

The first reader of this archive is a person. A title should reveal the question, the argument should make sense from the text alone, and every cited source should be open to inspection.

More information is likely to be encountered indirectly: through search systems and agents that retrieve, summarize, and connect material on a reader’s behalf. A good public research document should be clear enough that both humans and machines can preserve its context. Stable URLs, explicit publication and update dates, clear section hierarchy, source links, descriptive figures, relevant code and calculations, visible claims and limitations, and a clear Korean-English pairing all help.

This does not mean the articles are written to please agents. The order is the other way around:

A document that is clear and verifiable for people is usually easier for machines to read accurately.

Documents that are clear and verifiable for people tend to be easier for machines to interpret as well. Being “agent-friendly” is therefore not the main purpose of this archive. It is a distribution and rediscovery advantage that follows from good information architecture.

Research you can rerun, not just read

Some pieces may include executable calculations, interactive charts, adjustable assumptions, or small models that make a claim easier to test. If an article about multi-objective optimization lets a reader change the relative importance of cost and carbon and immediately see how the preferred solution changes, the article becomes more than an explanation. It becomes a small research instrument.

I want some pieces of research to be things you can rerun, not just read. For me, an archive is not only a place to store documents. It is a way to preserve ideas so they can be tested again.

An archive, not a feed

A feed optimizes for what is visible now. An archive optimizes for what is worth finding again later.

A feed favors recency and constant publishing; the next item pushes the previous one downward. An archive favors retrievability, durability, connection, revision, and reuse. It chooses a different optimization target without needing to dismiss the value of a feed.

For that reason, this site values the following standards more than publishing frequency:

• begin each article with one clear question whenever possible;
• distinguish facts, interpretation, and personal judgment;
• state the relevant date and the assumptions that matter;
• link sources, calculations, and code when they matter;
• make limitations and uncertainty visible;
• explain why a view changed;
• connect related pieces instead of leaving them isolated;
• keep the Korean and English versions aligned in substance;
• write for people first, while preserving enough structure for search and agents to recover the context; and
• where useful, let readers test or manipulate the underlying model themselves.

I know I will not follow these principles perfectly. Some arguments will age badly. Some explanations will need to be rewritten. Some views will turn out to be wrong.

But I would rather leave behind research that can be revisited, challenged, revised, and reused than conclusions that disappear into a feed. I would rather build an archive that can be reused by my future self and by others than a feed that only reflects what I happen to be thinking today.