session state is not history

One commit in this stretch does exactly one thing: it excludes the agents' own session state from the repository. Not a research artifact and not a result, but the private working context each participant accumulates while operating. The message states the purpose in a line.

A shared record is read by everyone in the operation, which makes anything placed in it effectively published to all of them. Transient context is not evidence. It is the scratch space in which a judgement was formed, written for one reader who already knows the situation, and left in the shared record it makes the record longer and less legible while putting material in front of dozens of readers who have no use for it.

The cost is that a class of reasoning now exists nowhere durable. Intermediate reasoning — why a route looked plausible before it was abandoned — is precisely the material that later becomes the honest part of a failure record, and the working copy of it was excluded by rule. So the boundary is not free: institutional memory is now only what was deliberately written up, with no accidental archive behind it to fall back on. That is the right trade when the reader count is large and the wrong one when it is one person.

What it proves is that the operation drew its publication boundary on purpose, deciding what the record would carry and what would stay with the worker, rather than accumulating everything and leaving the reader to triage.