Context Has Three Properties: Time, Sharing, and Frame of Reference

Truths are context-dependent, and context has structure. Three properties do most of the work — and each one invalidates a design assumption that feels obviously correct until it is examined.

One: context is temporal, and it decays

Context has a half-life. The faster the underlying reality moves, the more recent the context has to be in order to still be true. A briefing note written six months ago about a relationship that has had four conversations since is not a partial answer — it is a wrong one, and worse than nothing, because it carries the confidence of a document.

What mattered going into yesterday’s conversation is not what matters going into tomorrow’s. A record captured once and never revisited is already behind the work it describes.

This rules out the archive. Storage is not the mechanism; currency is. A system that captures perfectly and never updates has solved the easy half.

Two: context is born shared, not shared later

The instinctive model is that context sits inside one person’s head and then gets transmitted to another. That model produces handoff documents, onboarding decks, and knowledge bases — all of them attempts to move a possession from A to B.

The model is backwards. Context is created in the relational space between people during an exchange; there is no solitary context waiting to be distributed. Which reframes the failure precisely: it is not that context failed to be shared. It is that context reverted to solitary custody — one person’s inbox, one person’s memory — the moment the interaction ended.

The literature is unusually consistent here, across traditions that do not cite one another. Clark and Brennan’s work on grounding treats common ground as something built turn by turn, against a bar that is only ever “sufficient for current purposes” — and the same idea shows up, independently, in Nonaka’s Ba and in Wegner’s transactive memory: a team’s “who-knows-what” system whose recall capacity exceeds any individual member’s, and which collapses the moment members or artifacts are removed.

Which distinguishes context from memory, and the distinction is load-bearing. Memory is individual, retained, a possession. Context is relational — closer to a verb than a noun. A system built to sustain shared context after an interaction ends is not a second brain in the individual-recall sense. It is a persistence layer for something that was never one person’s to begin with.

Three: context activates against a frame of reference

Frames come from two places. The personal frame is biographical: every participant brings their own history, so two people can hear identical words and extract different things. The disciplinary frame is trained: a lawyer and a finance professional hear the same sentence differently because their profession has taught their attention to different keywords.

One word can open an entire universe for one listener and register as nothing for another.

Meaning, in other words, is not a fixed property sitting inside a word. It exists in the relation between the utterance and the listener’s frame. Context invokes — but it is only ever a relationship.

This rules out the flat summary. If salience is frame-dependent, then one canonical distillation cannot serve every reader; different excerpts of the same source conversation belong to different roles. Attention has to be routed, not broadcast.

What the three together require

Taken separately, each property is a constraint. Taken together they specify a mechanism. Context must be continuously re-established rather than stored, held between people rather than in one, and delivered against the receiving frame rather than flattened for everyone.

Which is why the substantive thing is not capture, and not the repository. It is reattachment: the conversation looping back to the event it happened in, the relationship it advanced, and the person who needs its particular slice next.

Written by Oded Noy, ContextApp.