A wording correction is made once and enforced; a content correction is new thinking every time

'No em-dashes' is a rule an agent can keep; 'sound more human' is not. Each jarring word becomes a line in the behaviour policy, restored after every summary and checked by the validator, so the corrections move from language to content, which is the measure that the voice is working.

Decided by. Nobody yet: a candidate for Dinis to read, comment on and decide (keep, rewrite, merge, cut, article first, later, not doing it).

Note. Sits next to every-mistake-added-a-rule: that idea is the trap (rules that grow), this one is the kind of rule that works (specific, enforced, few).

Where it comes from

How the voice works The rules; What the corrections are about now (newsroom)
Re-anchoring the policy printed back after every summary (newsroom)

What it answers in the 2018 book

none named

Where it goes

Chapter How to work with agents, part Working with agents (drafting); topic How to work with agents: briefs, voice, one shared state.

Other ideas proposed for the same chapter: An agent is a session with a focus: one session per topic, named, with a role file, The brief is the unit of work, and writing it as an article forces the research, Omnichannel programming: brief by voice, review on the phone, one shared state, Every mistake added a rule: the complexity trap of the agent operator.

Node idea:a-correction-becomes-a-rule in graph.json; see it in the graph. Source: admin/content/graph/ideas.json.