Lab
Can we detect decision cycles without psychologizing the user?
Short answer
A decision cycle can be described with what is on the record: the same question, handled on several dates, with the same options and nothing new logged in between. You can say so without ever attributing a feeling, a trait or a motive to the person. The Lab is studying this question and does not promise a feature.

Some questions keep coming back. Which invoicing tool to pick, whether to hire now, whether to stay with a hosting provider. You discuss it on a Monday, file it away, find it again three weeks later with the same options, and start from scratch. A memory that keeps decisions will see this back and forth. The real question is what it is allowed to say about it.
Our working answer is one rule: say what happened, never what the person feels or what kind of person they are. It is a writing rule as much as a design rule, and it can be tested. Here is how we phrase it in the Lab, along with its limits.
What a decision cycle is, and what it is not
We define a cycle by three things, all of them checkable:
- The same question is reopened (the subject can be identified, for example “choose the invoicing tool”).
- The same options come back, and no new one appears.
- Nothing new is recorded between two visits: no price change, no added constraint, no observed result.
It helps to say what does not count. Reopening a question because a quote went up is not a cycle. Revisiting a choice because a client changed their requirements is not one either. And coming back to a topic three times to sharpen the criteria, adding a constraint each time, is work in progress. Our piece on what happens to the old decision when a decision changes covers that legitimate case: the decision moves because the conditions moved.
So the difference comes down to one test: did anything change between two visits? That is also where caution starts, because a memory only knows what was written down.
Observable versus interpreted
A sentence can describe or it can interpret. The table below shows the line between them.
| Observable (you can check it) | Interpreted (someone is guessing) |
|---|---|
| This question was handled on March 4, March 19 and April 2. | You are struggling to make up your mind. |
| The two options have stayed the same. | You are avoiding a choice. |
| No new element is recorded between those dates. | Nothing has changed, you are going in circles. |
| The March 4 decision was never marked as made. | You are not standing behind your choice. |
The left column holds dates, counts and states. You can verify it by opening the history. The right column holds causes and traits: it gives the person an inner state that nobody has observed. A memory has no access to what people think or feel. It has access to what was written, when, and what replaced what.
Neutral wording rules
These are the rules we apply when we write this kind of output.
- Dates and numbers, not adjectives. “Three times between March 4 and April 2”, not “again” or “once more”.
- Say “recorded”. “No new element is recorded” describes the memory. “There is nothing new” describes the world, which the memory cannot see.
- No cause. We never say why the question keeps returning. The reader knows better than anyone, or does not know, and either way it is their business.
- No human subject. Sentences are about the question, the options, the dates. Never “you” followed by a verb about a state of mind.
- Open exits, including doing nothing. Reopen, close, or leave it as it is: three equally acceptable answers.
- No nagging. One statement, not a reminder. If it is dismissed, it does not come back in the same form.
An example: two ways to say the same thing
Imagine a freelancer, Camille, looking for accounting software. The example is made up. In her working memory there are three entries on the question, dated March 4, March 19 and April 2, each comparing the same two tools, with no trial result noted in between.
Here is the bad version: “You keep coming back to this choice. You may be afraid to commit. It is time to decide.”
It stacks up three faults. It assigns an emotion (“afraid”), it implies a diagnosis (“keep coming back” read as a character trait), and it applies pressure (“it is time”). It may also simply be wrong: Camille might just have been waiting for a trial period to end.
And here is the good version: “This question was handled on March 4, March 19 and April 2, with the same two options. No new element has been recorded since March 4. You can reopen it, close it, or leave it as it is.”
Both start from the same facts. The second leaves Camille free to interpret her own situation, including to answer: “I am waiting for a reply from their support team, it is noted elsewhere.”
What a memory can show, and what it must not do
A memory that keeps decisions, their reasons, the options set aside and the changes over time can show:
- the dates on which a topic was handled;
- the options on the table at each visit;
- what changed, or did not, from one visit to the next;
- the reason recorded at the last decision, so it can be reread (which is the point of retrieving past tradeoffs without delegating the decision).
It must not diagnose (“indecision”, “procrastination”), attach a label to the person, judge how good the process is, or send reminders that induce guilt. No counter that shows up like a reproach, no coaching tone. A cycle is a property of a topic’s history, not a flaw in the user.
A good starting point is also to keep the reasons with the decision: why preserve the reasoning behind a decision shows that a well documented decision gets reopened less by accident and more often for good reasons.
Consent and control
A statement like this should never arrive as a surprise. These are the principles we keep in mind while studying it:
- The user turns it on. By default, nothing is flagged.
- They can silence it, for one topic or for all, without having to explain.
- They can see what it rests on: the dates, the options, the items compared. A statement you cannot check is worth nothing.
- It stays in the private space. What starts private does not leave private on its own, and a note about how a person makes decisions has no reason to be shared.
These principles follow from what Nawame already says elsewhere: memory is a source, not an instruction, and sharing takes an explicit action.
Current limits
The Lab studies these questions. We do not promise a feature: what precedes is a method of wording and a set of principles, not the announcement of a product.
The first problem is false cycles. A memory only knows what was written. If a question is reopened because a phone call changed the data, and nobody logged it, it will look like a cycle without being one. The reverse risk exists too: a real repetition can go unnoticed if the topics are named differently each time.
Second limit: identifying the subject. Deciding that two entries are about “the same question” takes judgment, and that judgment can be wrong. Finally, neutral wording is not harmless wording: for some readers, even a date and a count can sound like a reproach. That is why the user’s control matters more than the choice of words.
If you want to try this by hand, with no tool at all: open your decision log, find a topic that keeps returning, and write two lines: the dates it was handled, and what changed between them. If the second line is empty, ask whether something changed without being written down. You can follow the Lab’s work on the Lab page, and to know when a piece of information has lapsed, how to tell that a memory has become obsolete offers concrete signals.


