A system represents unobserved entities as continuing to exist with
definite properties. It carries an entity's state through the
unobservable interval, predicts where it should be, and re-binds new
observations to the same entity on re-emergence. The commitment is
sharper than memory: ongoing existence between observations, not mere
retention of the last-seen value.
How would you explain it like I'm…
Still There When Hidden
When a ball rolls behind the couch, it's still there — it didn't disappear just because you can't see it. Your mind keeps a little picture of it: where it is, where it's going, that it's the SAME ball. So when it rolls back out, you're not surprised. Things keep existing even when you're not looking.
The Ball Behind the Couch
Object permanence is treating something you can't see right now as still being there, with real properties, even while it's hidden. If a ball rolls behind a couch, your mind keeps track that it still exists, roughly where it is, and which ball it is, so when it comes out the other side you know it's the same ball. This is more than just remembering: remembering is about past sightings, while object permanence is about the thing continuing to exist in the gap between sightings. So the hidden time isn't treated as a blank where the ball's status is unknown; it's treated as the ball quietly carrying on.
Existence Between Glimpses
Object permanence is the pattern of representing unobserved things as continuing to exist with definite properties. The defining commitment is that a system's internal model keeps an entity's existence and identity alive across the times it can't be seen: a ball rolled behind a couch still exists in the model, its position and velocity are carried through the hidden stretch, and it's re-matched to the same entity when it reappears. This is sharper than memory, which only persists past observations; object permanence is the commitment to ongoing existence between observations. So the unobserved interval isn't a gap where the entity's status is undefined, it's a continuation of the entity's history, a stretch of time over which it still has a state even though nothing is confirming it. That single distinction is the whole content, and it relies on a precise set of parts: a persistent entity, alternating visible and hidden intervals, a carried-through state, a prediction step, and a re-matching step on reappearance.
Object permanence is the structural pattern of representing unobserved entities as continuing to exist with definite properties. The defining commitment is that a system's internal model maintains entity-existence and entity-identity across periods when the entity is not directly observable: a ball rolled behind a couch still exists in the model, and its position, velocity, and identity are propagated through the unobservable interval and re-bound to the same entity when it re-emerges. The commitment is sharper than memory — memory persists *past* observations, whereas object permanence is the commitment to *ongoing existence between* observations. The system treats the unobserved interval not as a gap in which entity-status is undefined but as a continuation of the entity's history: a stretch of time over which the entity has state even though no observation is arriving to confirm it. That distinction is the whole content of the prime, and it carries a precise architecture: an entity treated as persistent; an alternation of observable and unobservable intervals; a persistent-state representation carried through the unobservable stretch; a prediction mechanism that propagates that state in the absence of observation; a re-binding mechanism that matches new observations to the same entity on re-emergence; and an underlying design choice — not a given — to model this particular entity as persistent rather than ephemeral. The pattern is substrate-independent because none of these roles names a medium. Its formalization is well known: the Kalman filter, with a predict step that propagates latent state between measurements and an update step that re-binds incoming observations, is engineered object permanence, and state-space modeling generally is the latent-state grammar this prime names.
Makes an implicit representational commitment explicit — "do we treat
this entity as persisting?" — which exposes a whole class of permanence
failures (ID switches, account collisions, re-meeting a known entity as
a stranger) as instances of one failure: loss of identity binding
across a gap.
Sorts cross-substrate moves into three: what cues re-bind, what
guarantees track continuity, and what detects identity failure — one
reusable toolkit instead of unrelated bug-fixes.
Supplies a latent-state grammar: unobservable things have ongoing
state the system models, and observations update the model without
constituting the entity's existence — the spine shared by state-space
modeling, Bayesian filtering, theory of mind, and developmental
cognition.
A Kalman filter tracks an aircraft between radar sweeps: it propagates
the state estimate (inflating its uncertainty), then gates the next
return to the same track — and swaps identities, the signature
failure, when two aircraft cross and the gating mis-assigns.
Object Permanence is not Associative Memory because it commits to ongoing existence between observations, whereas memory persists information past observations — retaining a last-seen value rather than modeling a continuing entity.
Object Permanence is not Determinism because it is a claim about representation (an unobserved entity persists), whereas determinism is a claim about dynamics (state evolves by fixed law) — orthogonal to whether evolution is stochastic.
Object Permanence is not Belief Formation because it is the prior structural commitment that an entity continues to exist, whereas belief formation is the process of arriving at what is held true, operating over that scaffolding.