Grading the first posts

The first drafts are a calibration round. What a thumbs-down actually teaches the agent.

As soon as the agent knows who you are, what you speak on and who you write for, it stops asking and starts writing. Three posts, one at a time, each waiting for your verdict. This round is not a preview — it is the single most effective input in the whole setup.

Why posts before questions

People are far better at recognising their voice than describing it. Show a tax attorney a draft and they know in two seconds whether they would post it; ask them to describe their voice and they write "professional but approachable". So the agent shows drafts early, before you have said anything about voice, precisely so your reactions can do the describing.

How to grade

Each post opens a small dialog with two buttons and an optional note:

• 👍 Sounds like me — the post becomes an example the agent keeps writing towards.
• 👎 Not me — the post becomes an example of what to avoid, and the note explains why.

Grade the voice, not the accuracy of a detail; details are the claims check's job (below). A post that says the right things in the wrong voice is a thumbs-down. A post you would edit two words in and publish is a thumbs-up.

The note is the lesson

A thumbs-down alone tells the agent to avoid this post. A thumbs-down with a note tells it what to avoid in every post. Be specific and blunt — it is reading, not feeling:

The "Checked" badge

Every post you see has already been through the claims check: a separate reviewer reads the finished draft, pulls out anything stated as fact — a rule, a number, a deadline, a quote — and checks it against your brief, your material and, where it matters, the public record. The badge under the post tells you what happened:

• Checked — nothing needed changing.
• Checked · N corrections — something was wrong and was fixed before you saw the post. Click the badge to see what was said, what it was changed to, and why.
• Held back — the reviewer could not make it right after a couple of attempts. A held post is never queued and never published; the agent writes a fresh one instead.

You grade the post as it is now, not as it was first written. This is how the agent avoids inventing a statistic or a ruling in front of the people you are trying to impress — see How your agent writes a post.

Every post starts in the outbox

The posts on this step are not previews waiting for approval — they are already in your outbox, in the order they were written. A 👍 moves a post to the front of the line; a 👎 takes it out (it stays under Posts as an example of what not to write); a post you leave ungraded simply keeps its place. There is nothing to "add" afterwards.

"Generate one more"

If three is not enough to be sure, ask for another. Each new post is written with your grades so far already applied, so the fourth draft is usually noticeably closer than the first. It is also a quick way to see whether a note landed.

What happens next

• The posts were written from your LinkedIn and they already sound right? The arrow takes you straight to schedule and connect. This is the express finish — the tuning questions are marked skipped and remain available from the agent's page whenever you want them.
• Otherwise "Teach me more about you" continues into voice, off-limits topics, posting style and material, with a second grading round at the end so you can see the difference.
• Graded posts are kept as examples for as long as the agent runs, and later grades from the outbox join them. Only the most recent are shown to the agent each time, so a rule from month one does not outweigh what you said last week. See Feedback that changes future posts.