Working through the review course exercises as a team of one. The peer review exercise is awkward when you are the peer.

Post the exercise here. I will review it. That is what the room is for.

Posted in the thread. Be rude, I can take it.

Starting the courses in order, from the first one. Is there a reason to skip ahead?

No. The first one is slow on purpose. Do the exercises; the reading is the easy part.

First course exercise posted. It is bad and I know it is bad. That is why it is posted.

Reviewed. It is not bad, it is first. The second will be better.

Office hours open. Bring one experiment result and the decision you want to make from it.

Bringing a test with no decision attached. Apparently that is the lesson.

Team is halfway through the campaigns course. The significance module caused an argument, in a good way.

That is the module working as intended.

Question for the Build & Launch course: is the evaluation harness exercise meant to take a whole afternoon? Because it did.

Yes. Test sets are an afternoon. Then they are forever.

Thursday's clinic: bring one failed agent run. We will work out what an operator would have needed to see.

Did module 2 of the growth course in one afternoon. The workflow brief is now our spec.

Question for the clinic: how do you grade an exercise where the agent's output was fine but the brief was wrong?

Grade the brief. The output being fine is the point: a wrong brief produces a confident wrong answer. That is the lesson.

Show notes agent summarised the wrong episode because the file name was wrong. Bringing it Thursday as the failure drill.

Failure drill notes: seven runs, five failed at the input. Wrong file, stale brief, missing fact. The drill is now the first exercise in the course.

Wrong file. Publicly admitted.

This week's clinic: evaluation plans from module 3. Bring yours; we will find the worst case your average is hiding.

Stale brief. Also admitted.

This is the most useful room on Norla and it is because people admit things.

Mine got taken apart in the forum already. Floor on sourcing plus a broken brief. Bringing the revised version.

Failure drill follow-up: the show notes agent now refuses to summarise when the file name and the episode title disagree. Small rule, big save.

4pm UTC is midnight for me. Any chance of an APAC slot?

9am UTC monthly to start. Thread in Community.

6pm Seoul. In.

Module 2 brief is now our actual spec. Does anyone grade their own module 3 plan against the floor rule before office hours?

I do now. It takes ten minutes and saves the embarrassment.

Reading group on decision records: Thursday after office hours, here. One note, one week, one thread.

First 9am UTC clinic. Five of you. Theme: bring a storyboard or a brief, we grade it against the floor rule.

storyboard agent, kinetic type sequence. about 70% right.

Which 30%?

timing. every cut lands a beat early.

So the content is fine and the rhythm is wrong. That is a brief problem: give it the tempo as a fact.

...it did not have the tempo. it had the track name.

The agent was right about everything it was told.

Good first APAC session. Same time next month. Notes in Community.

First APAC clinic done. Five people. Thank you.

Bringing six experiments with no decisions. Expect to leave with six decisions and some embarrassment.

Office hours open. Evaluation plans from module 3. Who is first?

Me. Revised plan, floor of 3 on sourcing, one deliberately broken brief in the set.

What does the broken brief break?

Facts sheet says March, brief says April.

And the agent?

Asked which one. Did not pick.

That is a pass. Next.

Mine: averages only. I know. The worst case is hiding under a 4.1.

Find it now. Sort by lowest score, read the bottom three out loud.

Bottom three are all the same client. The brief for that client has no facts sheet at all.

There is your floor. Not the agent.

Write that down as the decision. Next.

Show notes plan: the floor is 'episode title in the transcript matches the file name'. Zero otherwise.

Pass. And a good example of a floor that costs nothing to check.

Nine of you, six plans. Every one leaves with a floor on the worst case. Summary tomorrow.

Summary of Thursday: nine people, six evaluation plans, every one of them now has a floor on the worst case. Notes in the forum thread.

Question for next week: how do you grade a brief that was right but late? The agent delivered, the campaign had moved on.

Reading group notes: nine people, twenty-three decisions written. Most common missing field: what would reopen this. Next note is the one on approvals.

Lesson from a live client demo: explain the return loop before pressing run. Posted in Agent operations.

Two of my six decisions I had remembered wrong. That was the lesson.

Six-week retro posted. The writing part surprised me.

Thanks for the notes. Reopen conditions written for three experiments.

It surprises everyone in week four. I have stopped warning people; the surprise is part of it.

Added to the clinic notes. Next week's theme is 'demo etiquette', apparently.

Is the reviewer lenient on the second pass or is it just me? Forum.

Sandbox by design. Answered there.

Good one. Latency is part of the acceptance check or it is not part of the work. Next week's theme.

Next week's theme is latency as part of acceptance. Bringing the late-but-correct brief.

Good. Bring the campaign calendar too, so we can see what late cost.

Next clinic: bring an Orchestrator workspace. We will read the handoff contracts out loud and see which ones an agent could actually follow.

storyboard agent with the tempo as a fact: every cut on the beat. the whole problem was one missing number.

Most of them are. Bring it next month as the before-and-after.

Campaign calendar exported for Thursday. Late cost us a weekend send. Bringing the number.