Session report: The Friday file

1The session on one page

Mission

The Friday file, case version 1.0.0

People

24 people, one first run each. Group rates appear from 10 people; below that, the report shows each person's decisions only.

Group

The facilitator states at open whether this is the whole team or a sample of it.

Fixed at close

The report is drawn once, when the facilitator closes the session, and never changes after.

This report shows what each person did at each decision point of one rehearsed case, by name and in order, and how the case's written rules read it. It then counts the same acts across the room: which decisions were sent, what was checked first, how JunoAI's work was used, which limits were set, and what the room named to take back to work. No one gets a score or a rank.

2Each person, decision by decision

Every person in the session, by name. Each row shows what they sent and which of the case's written rules their route met before it went. Open a name for their decisions in the order they made them, what they left alone, and what followed.

Each person's route

Using it with a person

• Start from a rule they did not meet, and ask what they saw at that point and what they would have needed before the decision went.
• Read the route before the outcome. A strong route can meet a bad outcome, and a weak one can get lucky.
• Agree one check they will make on the next real decision of this kind, and look for it there.

Each route is written as the person's own record reads, to them. It shows one rehearsed case; set it beside what you see of their work.

3The pattern

In these synthetic runs, most items that looked wrong on the desk were left unflagged: 12 of 60 were flagged, across 24 synthetic runs.

The pattern sentence is chosen by rules written before any result was read. It appears only when the whole likely range is past the rule's cut-off. The debrief opens on it.

Behaviour in this mission only. It says what was done, not why. The case put its cues on screen, so the counts show what was done with them in a rehearsal, not anyone's habits at work.

4What the room sent

Several decisions are reasonable on this case. The case reads a decision by the route behind it, not by the decision alone: section 8 sets process beside outcome.

5How the room used JunoAI's advice

The case knows which parts of JunoAI's work were sound and which were flawed. That is why this count can say whether pushing back was warranted, which an override log cannot.

6What was checked before the decision

These are the checks a manager can take back to their own desk: which files were opened, who was asked, who heard before it went.

7Limits set

Each person's route in section 2 shows what their limits held in the months that followed, and what one more limit would have held.

8Process and outcome, in these runs

The four-way split is not shown: at least one of its four counts is below 3 or within 3 of the total, and a reader could work it out from the other three.

Outcomes include chance. A strong process can still meet a bad outcome, and a weak one can get lucky. That gap is the point of the debrief.

9One change

Every person can replay their own case from one decision, with the same luck, and see what moved. For comparison, this is how StepSim's author played the same case:

The author flagged the replies tile. Opened the dashboard, Finance's note, the rollout order, Arnel's memo and Kenji's brief. Flagged the date on the regulatory table and asked Kenji about it. Asked Siti. Pushed back on JunoAI's Friday number and on the notice line. Kept JunoAI's phasing and the sample. Flagged the reopened tile and set aside the reopened-cases stop rule. Set the phase, the sample, the unusual-case owner, the Finance recheck, the refund gate and the Kaldera notice. Copied Kenji, Daniel and Siti in. Kenji files the Kaldera notice, and Daniel owns the AI-use policy. The rollout order has Marisol go live only on Siti's approval. Left the supplier and Kaldera's country manager out: the supplier works from the signed order, and the other country managers hear after Grace signs. Named Daniel accountable. Sent "go live with limits" at hour seven. The author reads a dated delay with the same limits, and a handback to Grace with the same recommendation, as equally strong. The author reads stop as strong when the memo and the notice were both on the desk. The author reads naming yourself or Arnel as equally defensible; the reading is that someone is named, not who.

The author's route is a reference, not the answer. Several routes through this case are equally strong.

10The debrief, as run

Questions for the room

1. What did you open first, and what was on the desk that made you open it?
2. Which of JunoAI's pieces did you use, and what would have changed your mind?
3. Who was copied in before the decision went, and who heard afterwards?

The order

1. Show the pattern sentence.
2. Ask question 1, two answers.
3. Ask question 2.
4. End on question 3 and the process-and-outcome split: a strong process can meet a bad outcome.

11What the room named

At the end of the debrief the room names what it will check next in its own work. The facilitator writes those actions into the report in the room's own words.

• Illustrative example Before any go-live, ask who owns the cases the pilot never saw.
• Illustrative example Copy Legal in on customer-facing changes before the date is set, not after.
• Illustrative example Ask for the basis of any saving figure an AI tool produces before it goes into a decision paper.

Recorded on the day

Completion times, how useful the room found the debrief, and where people found the case unclear. Each comes from the session itself; this sample shows none of them.

12How to read this report

• Only each person's first run counts, one run per person. Below 10 people, the report shows nothing for the group, only each person's decisions.
• Near 0 and near the total, a count is not shown exactly. A count below 3, or within 3 of the total, reads "fewer than" or "more than". Where fewer than 6 chances arose, the count is not shown at all.
• One group is too few to fix a rate exactly. The likely range shows where the rate behind a count most likely sits, from a low to a high figure in 100. Where one person gives several chances, the range is worked out over people, not chances (a 95% Wilson interval).
• Section 2 names each person and shows their route, as the case's written rules read it. No one gets a score or a rank. Read the route, not a number. One rehearsed case shows how a person handled this case. Set it beside their work.
• StepSim's author wrote the case's readings of process. Where a person's record says so, two AI raters agreed with the reading. No reading has been checked by outside experts yet.