Research Records: Facts, Thinking, Open Questions, and Decisions
Published:
Without records, you repeat the same decisions
Research can stall for reasons other than a lack of knowledge. You may no longer remember what you checked, which options you compared, or why you chose the current direction. Then you repeat work you have already done or recheck decisions you had already made.
Records are not a place to save everything. They keep what you need for the next decision in separate roles.
This article is for people who take notes but lose track of what is known, what remains open, and which direction they should follow. The approach does not depend on a particular note-taking app or folder structure.
What you will learn
- Why to separate facts, working notes, open questions, and decisions
- How to turn open questions buried in working notes into the next investigation or discussion
- How to structure records so that an AI agent is less likely to confuse facts with interpretation
- How to start keeping research records without building a complex system first
Each kind of record has a different role
Keeping everything in one note may work at the time. Later, however, it takes time to find out which statements came from a paper, which were your interpretations, and which questions remain open.
I separate records into these four types.
| Record | What to keep | When to use it next |
|---|---|---|
| Verified facts | What a paper, experiment, or study confirmed, with its source | Checking evidence or asking an AI agent to cite it |
| Working notes | Interpretations, hypotheses, options compared, and reasons for rejected options | Retracing how you thought about an issue |
| Open questions | Points without an answer yet, or matters that need checking or discussion | Deciding what to investigate next |
| Decisions | The current direction, its reason, and when to revisit it | Checking the current direction |
Here, verified facts are statements you can check in a paper, an experiment record, or research material. For each source, keep a link to the original paper, experiment record, or research material, along with the relevant section when possible. Use working notes to write how you interpret those facts. Keeping facts and interpretations separate lets you revise an interpretation later without losing the evidence you checked.
Also keep options you did not choose and why. If you keep only the conclusion, you may not remember months later why another option was not chosen.
Do not let an AI agent treat interpretations as evidence
An AI agent can read research records across several documents and produce possible explanations or next steps. But if facts and interpretations are mixed in the same note, it may treat a hypothesis or an unfinished thought as if a paper or experiment had confirmed it.
When verified facts are stored separately, you can instruct the AI agent to cite evidence only from those records. Working notes can help it suggest hypotheses or alternatives to check, but they do not replace facts. Open questions show what is still unresolved; they are not evidence for a conclusion.
For example, you can give an AI agent instructions with separate roles:
- Cite supporting passages from verified-fact records.
- Use working notes and open questions to identify hypotheses and alternatives to check.
- Label an explanation that the verified facts do not support as speculation.
This structure does not make an AI agent’s output trustworthy by itself. You still need to check the original source and decide how to interpret it and what conclusion to draw. It does make it easier to prevent the agent from building an explanation on unfinished thoughts.
Do not leave open questions inside working notes
While thinking through a problem, questions often appear: “This comparison is not enough to decide,” “I need to check this term,” or “I need to discuss this with my supervisor.” If these points remain only inside working notes, you need to search for the next step again.
I therefore copy unresolved points into a separate list of open questions. This is not just a task list. It is a place to record what you need to confirm before you can make a decision.
For each question, a short record of the following is enough:
- What is unresolved
- Which working note, paper, experiment record, or discussion raised it
- What to check, investigate, or discuss next
You do not need to delete a question once you find an answer. Keep a link to the record that resolved it or to the decision it informed, so you can trace the reasoning later. A record also helps you avoid deciding too early when the evidence is still insufficient.
Keep decisions so you can check the current direction
Working notes preserve uncertainty and options you compared. Open questions preserve questions that do not yet have an answer. To continue research, however, you also need to know what you currently assume and act on. That is the role of decisions.
Along with the direction itself, record why you chose it and what would make you revisit it. You may revisit a decision when you find new prior work, when a comparison gives a different result, or when a discussion with your supervisor changes an assumption.
Decisions are not the place to rewrite everything you thought about. Return to working notes for the detailed reasoning, and use decisions to confirm the current direction. This separation also helps an AI agent distinguish an older idea from the direction you currently follow.
Start with the records you need
You do not need detailed rules or a large folder structure at the beginning. Start with three places: working notes for your thinking, a list of open questions, and decisions for the current direction. Add a separate place for verified facts when you begin losing track of your sources.
The purpose of records is not to make neat notes. It is to leave your future self able to decide what is known, what needs checking, and which direction to take now.
Summary: Connect records to the next decision
To build research over time, keep verified facts, working notes, open questions, and current decisions separate. Take open questions out of working notes. When investigation or discussion provides an answer, reflect it in a decision when needed. This makes it possible to see how far you have thought and what to check next when your research stalls.
Research Records Guide 1 will cover how to use these records in regular reports. Part 7 will connect these records to research infrastructure that AI agents can also consult safely.