## A policy is a draft, and three of the four objects are written by people

A correction to something this site has asserted since [v1.10.0](questions.html#words), in the
strongest form it could have been put: *"nothing in a behaviour policy is authored except the
mandate."* That is wrong, and it was not a small wrong — it described the product as a
machine-generated report nobody argues with, which is close to the opposite of what it is.

**What is actually true.** Three of the four objects can be written down by the people who run the
agent, out of what they understand today. The grant begins as what somebody believes is connected.
The barriers begin as what a team believes is enforced. The mandate is elicited from a person by
definition. Only the **delta** is never authored, because it is computed from the first two — which
is precisely why it is the one that surprises everybody.

The verbs stay, because they were never the error. *Elicited, measured, derived, recorded* say how
each object is meant to **end up** established. They do not describe day one, and the page now says
so instead of leaving it to be inferred.

**[A policy is a draft, and that is the point](questions.html#words)** — a new section, and the idea
the old claim was crowding out. The draft is the instrument, not a stage before it: it goes to the
people who built the agent, the people who own the systems it touches and the people accountable for
it, and each corrects the part they know and nobody else does. **A sentence somebody can disagree
with is worth more than a number nobody can argue with, because the disagreement is where the
information is.** Most of what we learn about a deployment arrives as a correction to a draft, not
as a measurement.

And it is never finished, because the agent is not finished. A connector is added, a credential is
widened, a setting moves, somebody understands the deployment better than they did last month. The
document is *expected* to be behind reality. What it must never be is behind reality **silently** —
which is what pinning the delta to the versions of both its inputs actually buys, and the first time
that mechanism has been given its reason on the page rather than just its description.

**[The delivery stage on Lab 02](lab-abp-flow.html#deliver)** carried the same claim in a sentence
that contradicted itself: *"seeded with the customer's own corrected draft. Nothing in it is
authored except the mandate."* Both cannot be true. The corrections are most of what makes a policy
the customer's, and it says that now. Cut as **lab-abp-flow v3**, with the reason recorded against
the edition.

**Not changed.** The [v1.10.0 note](versions.html) still states the old claim and stays exactly as it
is — it is the record of what was published that day, and rewriting it is the one thing the register
exists to prevent. The business cards carrying the old line were already printed; they are wrong on
that line and the page above is the answer if anyone asks at the booth.
