Not pile up forever.
Repeated beliefs get reinforced. Aliases get added. Old versions get replaced but kept on record. If two beliefs might be different, Tenure leaves them alone.
When a new belief matches an existing one, Tenure reinforces the existing belief instead of creating a second copy.
When a belief changes, older versions are superseded so the system can preserve what changed and why.
Tenure is conservative by default. If two beliefs might be different, they remain distinct until stronger evidence appears.
Before Tenure decides what to do with it, it checks for name and alias matches, close-enough merge candidates, where it came from, how confident it is, and whether anything explicitly says to update.
Tenure looks for existing beliefs by canonical name, aliases, and scoped merge candidates.
File-specific beliefs stay separate when the same name means different things in different files.
Insert, reinforce, add aliases, supersede, flag conflict, or skip low-confidence input.
Changes are recorded with session, turn, model, confidence, and prior belief state.
A weak signal and a durable memory aren't treated the same. An inferred belief needs to show up more than once, and needs some time to pass, before it goes active. That way one long session can't accidentally turn a shaky guess into settled fact.
Tenure merges beliefs when they express the same fact, preference, decision, entity, or expertise signal with enough confidence.
Tenure avoids aggressive compaction because false merges are worse than extra memory objects.
The merge rule stays simple: keep the evidence, promote things slowly, and never turn an uncertain guess into a confident-sounding fact.
Memory only helps if the model gets the right belief at the right moment. Merging isn't just tidying up. It decides what future sessions get to assume is true.
Tenure gives you a memory layer that can pile up evidence over time without ever losing track of where something came from, how sure it is, where it applies, or what it replaced.