Where does research end, and a new question begin?

How do we know that we have reached the end of a study?

When the code runs? When the experiment is complete? When the expected difference appears? Or when we can explain the result?

Each is an important threshold. They do not, however, mean the same thing.

An implementation may work correctly while the idea under examination fails to find the expected support. An experiment may finish without providing enough evidence for a decision. We may obtain a strong finding under particular conditions while still not knowing what happens elsewhere.

This distinction is one of the concerns I explore in Autorite: being able to state which responsibility a study has fulfilled, while separately preserving what it leaves open.

Knowing what we are trying to finish

Consider the question: “Is information about the past useful?”

As stated, the question has no clear endpoint. We can change the length of the history, choose other tasks or try different models. Every result opens another possible comparison.

To make the question testable, we need to bound it. In which task, with which historical information and against which comparison are we looking for a benefit?

Autorite's RP002A study therefore distinguishes different conditions: when the current observation is sufficient, when history carries additional information, and when history varies without adding information relevant to the future task.

These distinctions replace a broad statement such as “history is useful” with more specific questions.

They also create a place where the work can close. Once the predefined comparisons are complete, decision rules have been applied and limitations recorded, a particular research package can be closed.

The wider question can remain alive.

Changing the question after a result

During research, we may notice a pattern we had not considered at the start. A condition may behave differently from what we expected. A seemingly successful method may lose its advantage in another comparison.

These are productive moments. But a pattern noticed afterwards should not be confused with a claim that was designated for testing at the outset.

When we develop a new idea that could explain a difference, it is still a candidate explanation. The earlier experiment may not have been designed to test it.

Likewise, failing to meet a benefit criterion does not automatically establish that two methods are equivalent. Equivalence requires its own criterion and test.

Sometimes this is precisely where the new question begins: where the available record cannot answer it, but makes clearer what we need to ask.

What should remain at closure?

Simply marking research “complete” leaves too little information for a later return.

Which claim found support? Which did not? Which comparison remained inconclusive? Were there deviations from the method? Under which assumptions does the result hold?

In Autorite's research programme, the ability to close cycles is considered alongside recording these questions. Results should be accompanied by a synthesis, open questions and a fixed snapshot of the knowledge state at that stage.

The next study can then inherit more than the most interesting finding. It can also inherit counterexamples, unsuccessful attempts and matters left outside the scope.

That is the value of closure: not leaving the next study's starting point to an uncertain memory.

Not every new question requires us to continue

Many possible directions may emerge at the end of a study. We cannot pursue all of them at once.

Some are necessary to understand the current result. Some require a different method, data or resources. Others are simply interesting for now.

Making this distinction also allows research to pause. Recording an open question is not a commitment to answer it immediately.

What matters is being able to state why it remains open and what will be needed when it is revisited.

A study that has fulfilled a particular responsibility can close. A new question can wait for its own scope, method and success criterion.

This is how I think about continuity in research: advancing by preserving what we closed and why, rather than keeping everything open.

Sometimes a study's final sentence is an answer. Sometimes it is a question we can now ask more precisely.

Source note: This essay draws on Autorite's approach to research contracts, cycle closure and programme continuity. It reports neither a new experimental result nor the current status of a cycle.

Autorite · From a question to testable research · How do we publish an unsuccessful experiment? · Tracing a thought