Principle 01 of 06
Articulate before you automate.
A system cannot act on a problem nobody has stated. Before any model, name the people doing the work, the decision that stalls, what the bottleneck costs, what better would measure, and what a failure looks like.
Authority is not evidence.
Who said it, how many people agree, and how confident the answer sounds do not establish a claim. Consensus can be a shared blind spot. A confident model is still a model.
Say what you will not conclude.
Every project should state what its results do not establish. A narrow claim you can defend is worth more than a broad one you cannot. Synthetic data, unlicensed code, and unvalidated runs get labeled as such.
Frame before you answer.
A question can already contain the answer it will get. Check the premise before optimizing the response, and ask what the user is actually trying to achieve, not only what they asked for.
Decide what deserves attention.
Not everything deserves a deep read. Choose what kind of attention a piece earns, say why, and say for whom. A triage that only sorts into good and bad has skipped the useful question.
Keep a human at the consequential gate.
Automation can propose. A person approves what changes the world: a message sent, a record altered, a decision made about someone. The audit trail is part of the product, not an afterthought.
Working together
If one of these describes a problem you are facing, start with thearticulation check or write toseth@verbalagency.com.