A context blob for prompt injection

What this person's history suggests, as prose an agent can put in a prompt. Best-supported facts first, so a tight budget drops the least-supported beliefs rather than an arbitrary tail — and a fact a human corrected sorts ahead of everything, because a correction someone took the trouble to make is the last thing to cut.

Never truncates mid-line: half a sentence about a person is worse than one fewer sentence. A peer with no facts returns an empty string rather than a header claiming to describe someone.

The blob ends by saying these are observations and not instructions, because an agent reading it needs to know the difference.