What happens when Second Brain remembers two things that contradict each other?
One of the earliest problems we ran into building v2 was this:
You decide something in January. You change your mind in March. You ask about it in July.
What should the system return?
A naive vector search returns whichever memory scores higher. Sometimes that is the old one. You get stale context presented as current fact, with nothing flagging it as potentially outdated.
Second Brain v2 handles this with conflict detection. When a new memory is stored that contradicts an existing one, the older memory is marked as superseded rather than deleted. Both remain in the graph, but recall scoring weights the newer one significantly higher.
You can still surface the old decision if you ask for it directly. But in normal recall, the system leads with what is currently true.
This matters most for:
• Decisions that evolve over weeks or months
• Preferences you have updated but not consciously logged
• Plans that changed after new information arrived
The old memory also stays linked to the new one, so you can always trace why you changed your mind, not just what you decided.
How are you currently handling stale context in your notes or memory tools?
Replies
Second Brain for AI
@benyamin_chenkirov Both. Recall leads with the current state by default. The older memory stays in the graph as superseded, linked to the newer one, so you can trace when your thinking changed and why. You get the answer without losing the history.
I've definitely run into this problem witrh my own notes. Finding the latest decision is easy but remembering why I changes my mind is the valuable part.
Second Brain for AI
The "why" is what the linking captures. Old memories stay linked to what replaced them, so the reasoning doesn't disappear.
This is a really important problem for AI memory. I don't want old information deleted I just want the system to understand what is still relevant.
Second Brain for AI
Superseded rather than deleted is the key. Old context stays linked and queryable, it just doesn't lead in recall anymore.
I've had documents where the newest update was buried under older versions. A system that understands this replaced that feels much closer to how humans actually work.
Second Brain for AI
The new leads in recall, but the old stays linked. You can trace what replaced what without the history burying the current answer.
This feels like a smart detail that will make AI assistant much more trustworthy. Knowing when something was replaced is a huge part of having useful memory.
Second Brain for AI
@hassan_ismail_rebe Trustworthiness is exactly the frame. A memory that knows when something changed and can show you that feels fundamentally different from one that just returns whatever scored highest.
you already lead with the right instinct: keeping the old memory linked rather than overwriting. the follow-up i'd think about is when to surface the conflict visually vs just letting the newer one win silently.
my take from getting burned by note apps: any time recall scoring shifts within a session, show a small "this was different last month" badge with old and new side by side. even if the user doesn't click, they know something moved. silent updates are where trust dies.