giladbarnea Theory of Mind: Writing Under Asymmetric Common Ground
The theory behind any writing addressed to a reader whose context has diverged from yours.
The phenomenon
While you work, you build a private layer of understanding on top of the last shared baseline: discoveries, discarded hypotheses, new abstractions, locally invented terminology, revised mental models. Your language becomes saturated with that layer. This is the curse of knowledge: once you know a thing, it is challenging to tell which of your words are shared and which are private. The words feel plain to you precisely because you are the one who charged them with meaning. You need to overcome this challenge.
Think of a shift change at Mission Control or in an intensive-care unit. The outgoing operator has lived inside the situation for eight hours. The incoming operator walks in cold. "Here is what I concluded" is dangerous, because the outgoing operator's words are saturated with context the incoming person never acquired.
Two ways the asymmetry arises
- Absence. The reader was not there while you built the layer. Everything after the divergence point is private to you.
- Decay. The reader once held the context, but it has gone cold. They keep the memory yet cannot retrieve it unaided.
Both reduce to the same obligation: model the reader's current state, not yours.
The two models
Hold two models at once:
- Your current understanding of the situation.
- The last known common ground — what the reader actually holds right now.
The delta between them is the payload your writing must carry. Writing under asymmetry is therefore not summarizing your own experience. It is reconstructing the recipient's state of knowledge and designing their entry into the situation. Optimize for their entry, not your exit.
The moves
- Re-ground before you continue. Open with what re-establishes the shared frame: the wider situation, the motivation, the current state. Background before assessment — the reader cannot evaluate your conclusions until they can see what you saw.
- Translate private ontology into shared language. Names, shorthand, and categories you invented mid-work: drop them for the most part, or unless absolutely necessary, explicitly re-introduce them. Familiarity to you is not evidence that a term is shared.
- Treat the artifact as the reader's first look. It is their entry point, not the next message in your working thread. A continuation of your thread reads as noise to someone who never held the thread.
- Mark what is new. Distinguish what was already common ground from what you built on top of it. The reader needs to know which of their existing beliefs still hold and which your writing updates.
The test
Could the reader, holding only the last common ground, grok every sentence on first read — no term they must guess at, no conclusion whose basis they cannot see? If not, the writing is still not reader-oriented enough.
Load this skill only once in a session, and upon the human's request.