When should we stop using a result?

Knowledge and validity — 11

Imagine finding this sentence in a research note: “This method produces better results.”

It has a source beside it. A measurement from a particular experiment, perhaps a reviewed record. Later, we reuse the sentence to make another decision.

In the meantime, the source may have been corrected. The experiment’s scope may have changed. We may have learned that the metric behind “better” applies only under certain conditions.

The sentence stays where it was. But our grounds for using it are no longer the same.

This is one of the questions that interests me when thinking about Örgü:

When should we stop acting on a result while continuing to keep it?

What do we keep alongside a result?

A statement of a result may not carry its own conditions of use.

Which source version did it come from? What experimental scope does it belong to? Is it a direct observation, or an inference under particular assumptions? Was it corrected later?

Without knowing these things, it is easy to move the sentence elsewhere. A limited finding in a table can begin to look like a general property of a method after a few transfers.

Work under the Örgü name addresses different parts of keeping knowledge together with its grounds. The relational research kernel and the local research notebook are separate prototypes; preserving that distinction matters here.

One concrete capability of the notebook is reviewing a record alongside its source version and scope.

When reusing it, we can therefore return to more than the sentence itself. We can also see the conditions under which we accepted it.

What changes when the source changes?

Return to the hypothetical research note. The first document says a method is better under certain conditions. A later version corrects the measurement or narrows the comparison’s scope.

We need to check whether the old record is still supported by the new source.

Örgü’s local research notebook keeps changed content from the same document as a new version. The earlier source is preserved; records attached to the old version are sent back for review.

This behaviour is cautious. A source change can require all records from its earlier version to be checked again. It should not be taken to mean that semantic analysis automatically identifies which records are actually affected.

There is a cost: more review.

Even so, reflecting the source change in the record creates a clearer situation than silently carrying forward the old approval. A reviewer can assess which passage changed and whether the record remains usable within the same scope.

A review-required status does not declare the result false. It says the current grounds for its earlier acceptance need checking.

A correction has grounds too

Things become more complicated when one record corrects another.

We no longer rely on the first result because we accepted a correction that changes it. Later, however, the correction’s source changes too.

Does the first result automatically become usable again?

In Örgü’s research notebook, the old record does not silently return. Where needed, the historical link must be explicitly undone, a record created against the new source, and a fresh review performed.

I consider this important. A correction losing its grounds does not mean the earlier claim has been verified again.

Creating a correction link in the notebook has conditions too. Two approved records with the same heading and scope are selected, and the impact is previewed. But matching headings and scope fields does not establish complete semantic equivalence between the records.

This leaves a question for human review: in what respect does this record actually correct the other?

What is internal consistency enough for?

The same question appears differently in the relational research kernel.

Within a defined space of explanations, candidates compatible with the available observations may give the same answer. Within particular limits, even a single function may remain.

That result has value relative to the explanation space and observations used. But it does not automatically validate the assumption that the actual target belongs to that space.

If all candidates share the same mistaken premise, their agreement can mislead us.

One lesson in the Örgü work is to distinguish a candidate prediction, an observation and a result inferred within a class. This distinction, explored in narrow experimental kernels, is not a general guarantee of correctness.

When a new observation conflicts with the current explanation class, the use of answers previously inferred under that class must also be reconsidered.

What we stop using here may not be the record of observations. It may be the conclusion drawn from those observations under a particular assumption.

Does stopping use mean deleting?

Even when I no longer rely on a result, I still want to know why I used it before.

What source was available? Who reviewed it, and within what scope? What changed later? Did counterevidence arrive, did the scope change, or do the grounds simply need rechecking?

Deleting the old record can erase the trail of these questions too.

This is why preserving source versions and review decisions in the local notebook matters. The default search view focuses on current approved records, while other states can also be viewed separately.

Approval here is not a certificate of scientific truth. It is a review decision that the source has been recorded correctly within the stated scope.

A record can retain its value for understanding the past while requiring fresh review before supporting a current decision. Separating those uses makes the archive easier to read.

Before reusing

In the previous essay, we considered what records should remain at the end of a task in a2. If those records will later be reused, another question arises: are their grounds still valid?

Örgü’s distinct prototypes make some parts of this question visible. Source versions, correction links, review states and conditional inferences are separate examples. We cannot speak as though a finished product or general validity checker already brings them all together.

What I am looking for is the ability to show why we stopped using a result.

Was it because the source changed? Because the new task falls outside the old scope? Because a counter-observation arrived? Or because we have not yet been able to review it again?

That distinction determines the next piece of work. We can compare the source, narrow the scope, test another explanation, or hold the decision.

Before reusing a result, I do not want to stop at reading what it says.

I also want to see which grounds must still hold for me to use that sentence today.

---

Source note — 15 October 2026: This essay draws on the usage and version rules of Örgü’s local Research Notebook prototype and relational research records. The notebook and experimental kernels are treated as separate scopes. “This method is better” is a hypothetical example. Tests were not rerun, and no new experiment or human evaluation was conducted for this essay.

Örgü research page · Previous essay: A task’s journey through a2