Memory & Context
How to tell that a memory has become obsolete
Short answer
A memory is obsolete when it has been replaced, contradicted, has expired, no longer fits the situation or has become moot. The signals are a date, a change of context, a new decision or a contradiction. Obsolete does not mean false: mark it and replace it instead of deleting it.

Your notes say: “The site is down for maintenance until Friday.” The note is three weeks old. It was true, it no longer is, and nothing says so. If you skim it, or if an AI retrieves it and reports it back, you head off on information that has no standing anymore.
This kind of mistake is hard to spot because nothing was wrong at the start. Nobody wrote anything false. Time did the damage. Here is how to recognize that a memory has gone obsolete, and what to do about it.
Obsolete does not mean false
Start by separating two things that often get mixed up. A false memory was never accurate: a date copied wrong, a figure misunderstood. An obsolete memory was accurate, and then the world moved on.
The difference matters for what comes next. You correct an error, and sometimes you delete it. You do not delete an obsolete memory, because it records what was true at a given moment, and that story can be useful: why the rate changed, when the tool was swapped, what you tried before. What you want to avoid is letting it pass for the current state.
This nuance is at the heart of our piece on what happens when a decision changes: the replaced decision stays in the history, but it should stop posing as today’s rule. Here we widen the view to every kind of memory, not only decisions.
Five ways to go obsolete
Take an independent consultant who keeps notes on clients. The example is invented, but each case is common.
- Replaced. “Day rate: 450 euros.” In January it went up to 500. The old line is still there, superseded by a newer one.
- Contradicted. One note says the client pays in 30 days, another says 45. Both date from the same period and neither is marked as the right one.
- Expired. “Site down for maintenance until Friday.” Friday has come and gone. The note carried its own end date, but nobody reread it.
- Out of context. “We invoice with this tool.” True for the old client. The new client uses something else, and the note floats with no attachment.
- Moot. “Wait for the town hall’s reply before starting.” The project was canceled. There is nothing left to wait for, yet the note still asks for action.
These five cases do not call for the same response. A replaced memory gets linked to what replaced it, an expired one is closed, a contradicted one is settled, an out-of-context one is tied back to its scope, a moot one is filed away.
Signals that should raise a flag
You cannot reread everything you have written, so you watch for signals. Four come up often.
- The date. The older a piece of information, the more it deserves a look, especially if it concerns something that moves (a price, a status, a deadline).
- A change of context. New client, new team, new tool, new budget year: everything noted before is worth rereading in light of the change.
- A new decision. Whenever a choice is replaced, look for what depended on it. The reasons kept alongside the decision help: see why preserve the reasoning behind a decision.
- A contradiction. Two notes that disagree are a strong signal, even when you do not yet know which one is wrong.
A fifth signal is subtler: silence. A piece of information you have not looked at for a long time, on a subject that changes, is not necessarily stale. But it has not been confirmed either.
Write the shelf life with the fact
The best moment to deal with obsolescence is when you write. When you note a fact you know to be temporary, add its shelf life:
- “Under maintenance until Friday the 14th”
- “Rate valid for the first half of the year, revisit in July”
- “Contact: Sophie, while Marc is away”
Three extra words, and the note carries its own expiry signal. Someone who reads it later, human or AI, does not have to guess. The habit is simple, and rarely kept, because when you write you think about what is true, not about what will still be true.
Do not overdo it. Not every piece of information has an end date: a client’s name, the address of a head office or a working principle can last for years. Stamping an expiry you do not actually know amounts to inventing false precision.
Replace instead of deleting
Faced with a stale memory, the temptation is to delete it. That is tidy, and it is often a mistake. You lose the history, you can no longer answer “why did we change?”, and you cannot check what held before if a doubt comes up.
The better practice is to mark the old one as replaced, say by what and on what date, and leave it readable. You read the current state first, and the history only when it helps.
Nawame works this way: a reference item has a life cycle, it can be corrected or replaced, and the history of replacements is part of what you can export. The old one is not lost, it is just no longer presented as the reference. The same principle works in a plain spreadsheet: a “replaced by” column and a “date” column are enough.
What an AI should do with a doubtful memory
When an AI retrieves a piece of information that looks old, or seems to contradict another, the right reaction is neither to ignore it nor to present it as settled. It is to flag it.
Compare two answers to the question “what is this client’s rate?”:
- “450 euros a day.”
- “A note from last year says 450 euros a day, but a more recent note mentions 500 euros. I cannot tell which one is current. Can you confirm?”
The second costs one more sentence and prevents an invoicing mistake. It assumes the AI has access to dates and to the links between memories, otherwise it has nothing to flag. That is one reason to keep provenance and chronology alongside the content.
The same principle holds for other kinds of information: an outdated hypothesis or an old instruction should not start speaking again without anyone noticing. We tell these categories apart in the difference between a memory, a fact, a decision, a hypothesis and an instruction. And found does not mean true, any more than it means current: see why found does not mean true.
Limits: you cannot always tell what has gone stale
The most awkward part is that obsolescence is not always visible. A note with no date looks like a fresh one. An old fact can remain perfectly valid (the founding date of a company does not expire), and a recent fact can already be out of date. A date is a clue, not a verdict.
There is also the opposite risk: archiving too fast. A piece of information filed as obsolete when it still holds drops out of the results, and nobody notices what they no longer see. That is one argument for a memory that does not keep everything, as explained in why a good memory knows when not to write, but that keeps replacements visible.
Finally, an obsolescence error can spread unseen: an old fact gets copied into a summary, then cited as current. That is the subject of the errors a long-term memory can carry forward. No method promises you will catch everything. The only serious defense is to keep dates and replacements visible.
Three questions for a light review
Once a month, or at every major change, take ten minutes and run through your working notes with three questions.
- Is this still true today? If you do not know, mark it “to check” with today’s date.
- What replaced it, contradicted it or made it moot? Link the old to the new, or file it away.
- Has a deadline passed? Close it or extend it explicitly.
No need to reread everything: start with what concerns prices, statuses and deadlines, where time does the most damage fastest. If you want help keeping replacements and their dates, the FAQ describes what Nawame keeps and how it handles changes. Otherwise, a spreadsheet with a “status” column already does a lot.


