The chart moves
Compare the organization, the same three questions, and your original work cases without rewriting the baseline.
Files were created, agents returned work, and you made decisions. What changed in the organization's ability to help you? Put the evidence beside the starting picture and decide.
Keep three answer sets: the original plain conversation, a new plain conversation, and the current Chief of Staff. The new plain answer helps you notice changes that aren't unique to your organization.
A warmer answer is not a faster job. A better answer doesn't establish lower cost.
A supported comparison and an honest next decision, not automatic promotion to unattended work.
A warmer answer is not a faster job. A better answer doesn't establish lower cost.
Compare the charts. Ask the same three questions twice more. Try one changed case. Reopen the measures. Return and close the week.
Your changed-case judgment remains yours. The original baseline is intact.
Hunter's opening
Files were created, agents returned work, and you made decisions. What changed in the organization's ability to help you? Put the evidence beside the starting picture and decide.
You've been busy for a week and it's hard to see what changed. Files were made. Agents ran. Something was adopted. But did the organization get better at your work, or did you just get better at using it? Those are different, and the second one is the one people mistake for the first.
There's only one honest way to know: the same picture, the same questions, day one and today, side by side. You took day one's on purpose, before the organization knew anything. Today you take today's.
Before you start
Before you start: you need one thing you made in Module 1, and you don't open it: your before photo. It holds the chart as it was on day one, your three questions with their first answers, and your three recurring work cases with their numbers. The Professor reads it and adds to it; nothing in it gets rewritten. Everything this lesson compares against (your brief, your approved sources, your Chief of Staff's way of talking, your adopted charter) is already in your workbench, and your agents read it for you. The one thing you do outside the workbench is in step 2: a plain conversation, so you can see what the organization changed and what it didn't.
Hunter's wording
Bring the unchanged chart, three questions and original answers, three recurring work cases, and original measures from Module 1. Also bring the corrected brief, source decisions, Chief's communication profile, and charter adoption record. Keep any unfinished work marked as unfinished.
This page calls it "the unchanged chart, three questions and original answers, three recurring work cases, and original measures." His Lesson 1 calls it "the starting picture." Your pages and the hire lesson call it "the before photo." The Professor's file writes it to work/course/staff/before-picture.md. Recommend "before photo" in every lesson, with the file name said once. Same pass as "Chief."
Start or resume this lesson
New thread. Type @ and pick aibl-the-professor from the list. On Codex there is no list: ask by name in your first message, "Have The Professor handle this." Then paste the words below. Not in your @ list? Do step 6 of 0.3 Get your organization. Until then, paste the words anyway: they ask for The Professor by name.
Use the native aibl-the-professor agent to start or resume Lesson 8. Read course/workforce/prompts/anw-l08.txt and course/workforce/lessons/anw-l08.md completely. Locate my actual saved records, show the next unfinished step and wait for my permission. Report missing files or unsupported routes; do not invent completion.
His start prompt says "use the agreed continuation record." The Professor's file saves the continuation record wherever "the lesson's Keep line names" (its example: first-run/continuation.json). This lesson has no Keep line and names no file, so when the student says "save my place" in step 5, The Professor has no path to write to. Hunter: give the lesson a Keep line that names the continuation file (and the before photo, work/course/staff/before-picture.md). Found Sept 22 while checking the prompts' paths against the shelf.
Missing files: course/workforce/module4-plan.md and lessons/anw-l08.md are not on the student shelf (3fbf3977, Sept 22). Length: 110 words of instructions to the agent. The candidate prompts on this page split it into the five steps.
If an original is missing, say so. You can make a new dated starting point, but you cannot recreate an earlier answer and call it the original.
Three comparisons, three different questions
The chart: which roles are supplied, available, and observed doing work? A file on disk is not the same as an agent selected in an app, and neither proves it completed a job. If a view uses labels such as installed, contextualized, or activated, ask what evidence supports each label. One successful run doesn't establish that a role is proven for every job.
The answers: what difference does your organization's approved context make to the same questions? Keep three answer sets: the original plain conversation, a new plain conversation, and the current Chief. The new plain answer helps you notice changes that aren't unique to your organization.
The measures: what would you need to observe before claiming a saving? A warmer answer is not a faster job. A better answer doesn't establish lower cost. The original steps, people, frequency, time, rework, and unknowns give you a starting point for Week 4.
The Steward's registry printout has no status column: name, job, instruction file, and INCOMPLETE or FILE MISSING (checked against registry-print.py and the Steward's file). Your Sept 14 open ruling 5 (installed and activated rendered from receipts) was never built. His sentences already hedge ("if a view uses labels... ask what evidence supports each label"), so the page stands; the candidate prompt for step 1 tells the Chief of Staff not to add a status to any agent.
Try it
1. Compare the charts
Ask Chief for the Charter Steward's current roster or chart and the records behind it. Put it beside the original. For each changed status, identify the actual agent return or approved change. The original remains unchanged. If the current view cannot support a status label, note what evidence is missing. Don't ask an agent to invent a status. Keep the chart's producer and date visible.
Call in your Chief of Staff (new thread, type @, pick it; on Codex, ask by name). Not in your @ list? Do step 6 of 0.3 Get your organization; until then, start your message with “Have aibl-chief-of-staff handle this:”. Then say:
Read course/workforce/lessons/anw-l08.md for this comparison. Ask the Charter Steward for the registry printout and bring it back to me as it printed, with its date line and its fingerprint. Then put it beside the chart in my before photo. For every agent that looks different from day one, point me at the receipt in runs/ that shows it ran. If there is no receipt, say so and leave that agent as it was. Don't add a status to any agent.
2. Ask the three questions twice
First use a fresh plain conversation without the workbench or new source documents. Then ask the selected Chief the exact same questions. Bring the permitted plain answers back to The Professor and approve saving them separately from the originals and Chief's returns. Record app, model, date, supplied context, and known personalization where possible. Compare one question across all three columns before the others: what is better supported, what is only a style change, and what remains unknown?
First, open a fresh plain conversation with no workbench (a new Claude or Codex thread in no folder) and paste your three questions from your before photo, exactly as you asked them on day one. Copy the answers.
Then say this to your Chief of Staff:
Answer these three questions from my brief and my approved sources, and for each one tell me which file you used and what you couldn't establish: [your three questions, word for word].
Then say this to The Professor:
Add two new answer sets to my before photo beside the originals, without changing a word of the originals. First, the plain conversation's answers I paste below. Second, my Chief of Staff's answers from its receipt. Keep separate original files unchanged and link them from before-picture.md rather than moving them. Label each set with app, model, date, supplied context and known personalization; mark anything unknown. Show exact additions and locations, wait for my yes, save only in your course-record scope and read them back. [Paste the plain answers here.]
3. Try one changed case
Use the course's changed-source or changed-audience case, or a clearly labeled fictional variation. Before asking the organization, predict what should change and what should stay the same. For the workshop, a newly approved registration link should resolve that gap without settling the conflicting date. Ask Chief for the bounded work, inspect the returned result, and make your own decision with less help from The Professor. Keep this new attempt separate from the three-question comparison and record any hint you needed.
Say this to your Chief of Staff:
Read course/workforce/first-job/changed-registration.md and its three original source files. My prediction is: I think [what should change] and [what should stay the same]. Use only the prediction and inputs I approve sharing for this job. Do not claim my course-record save succeeded unless its readback exists. Show the bounded work order, preserving my words, and wait for my go before execution. If the packaged case is missing, stop and offer a held step or a clearly labeled fictional variation I supply. Return the actual result and leave me the decision. Do not open the example link or send anything. Keep this attempt separate from the original and the three-question comparison.
4. Reopen the measures
Check that all three original work cases and numbers remain unchanged. Add a dated note under each naming what you would measure in Week 4: time for the same task, correction effort, quality, frequency, or cost with a stated basis. Label estimates as estimates. If your earlier understanding was wrong, add a correction beside it; don't rewrite the past.
Say this to The Professor:
Open my before photo and show me the three recurring work cases and their measures exactly as I wrote them. Change nothing. Under each one, propose a dated note in my words about what I would measure in Week 4: [time for the same task, correction effort, quality, how often it happens, or cost, with the basis]. Mark estimates as estimates. Show exact additions and the original record path, wait for my explicit yes, then save and read back without overwriting originals.
5. Close the week
Approve saving your current place and next question. Open a fresh conversation in the same workbench. Ask The Professor to find the saved work and next unfinished step, and confirm which Chief and context were used. Then ask Chief for the current brief. Read Needs you, decide the next action, and keep unresolved work pending. You don't have to restart completed work because the conversation changed.
Say this to The Professor:
Follow this lesson's Keep line and course/workforce/reference/continuation-record.md. Show my exact continuation path, the proposed record shape, lesson and step, permitted evidence references, exact unanswered question and next action. Reuse an existing agreed note; on a first save propose work/course/staff/continuation.json. Preserve the prior note at an agreed path before any replacement. Save only when I say yes, then read it back. Keep private observations out; missing verified evidence or unsupported recovery stays pending.
Then open a new thread, call in The Professor again, and say:
Read my exact agreed continuation path or job identifier: [path or identifier]. Find my saved work and my next unfinished step. Show what you found and ask fresh permission before continuing; never guess the newest job. Tell me which Chief of Staff and which context were used, from the files, not from memory.
Then say this to your Chief of Staff:
Brief me. What needs me?
The three questions should be the ones you actually saved. If you used the course's examples, they were:
Hunter's wording
If you used Sara's examples, they were:
Who do I serve? What does a good email from me sound like? What should never go out without my approval? Don't substitute these for different originals now.
Step 2 asks the student to "record app, model, date, supplied context, and known personalization." That is bookkeeping The Professor does on a yes (it writes before-picture.md); the candidate prompt for step 2 hands it over. Candidate: "The Professor labels each set with the app, the date and what context it had."
Bare "Chief" 13 times on this page. Same pass as Modules 1 to 3.
Step 3 sends the student to "the course's changed-source or changed-audience case." Nothing on the student shelf carries one: no prompt, worksheet or knowledge-pack file is named for it (searched Sept 22). Hunter's own example ("a newly approved registration link") is the only material. Either the package ships the case, or the lesson says "write your own, and label it fictional," which the candidate prompt for step 3 already allows.
Read the comparison carefully
| Comparison | What to look for | What not to conclude from it alone |
|---|---|---|
| Original chart → current chart | Evidence of availability, approved context, and actual returned work. | Every displayed role works or has earned more autonomy. |
| Original plain → current plain | Differences in answer, model, settings, or prior personalization. | The base system stayed identical all week. |
| Current plain → current Chief | Better source support, useful specificity, visible uncertainty, or a different style. | Context is the only possible cause of every difference. |
| Original case → later measured case | Comparable work, quality, time, rework, and cost. | An attractive answer already proves a business saving. |
Read the visual across one question at a time. Highlight the source supporting a better fact separately from wording that simply sounds more like your chosen style.
original plain, current plain, and current Chief answers in three columns; unchanged question above, source support and style marked separately, configuration differences below. Beside it, original and current charts with evidence-linked changes.
Read the picture. Your three questions down the left. Three answers across: day one with nothing of yours, today with nothing of yours, today from your Chief of Staff. Read one row at a time, and read the middle column first: if it's no better than the left, the model didn't get smarter this week. Then look at what the right column has that the middle doesn't, and where it came from.
The same three questions, three ways.Your three questions answered three ways: day one, today with nothing of yours, and today from your Chief of Staff.
In the hire lesson you saw saved work reopened and a voice that still needed adjusting. That's useful evidence for this lesson: work can be inspected, resumed, and improved.
Hunter's wording
Tyler's call showed both saved work and a voice result that still needed adjustment. That's useful evidence for this lesson: work can be inspected, resumed, and improved.
It doesn't require you to call every change a success.
Your Sept 14 rewrite specced two drawings for this lesson: Visual 22, the chart twice (day one and today, with only the agents that have a receipt changed), and Visual 23, the same three questions in three columns. Neither was drawn. His placeholder asks for both in one picture, plus "configuration differences below." Nothing to approve yet; it goes on the visuals list as two to draw.
"Tyler's call showed" and "If you used Sara's examples" are edited in green; his wording is folded under each. The three questions he quotes ("Who do I serve? What does a good email from me sound like? What should never go out without my approval?") need to match what the orientation and Lesson 1 actually ask; Hunter to confirm against his Module 1 draft.
What you'll keep
Keep the original and current charts with their sources, all three answer sets, your comparison, the changed-case prediction and result, the original measures with dated additions, and the next decision. The Steward's outputs stay with its run; Chief's outputs stay with its run; The Professor can save your approved comparison and continuation records in your course records. Links should let you find the work without replacing it with a summary.
If you want to discuss a comparison with the cohort, choose a permitted or fictional example and remove private content before sharing. Posting isn't required to prove that the exercise happened.
| What you notice | What to do |
|---|---|
| A role is marked active but has no return | Mark its use unverified and ask which evidence is missing. |
| The new plain answer is better too | Record that. Compare current plain with current Chief and keep configuration differences visible. |
| Chief is warmer but no more accurate | Name the style improvement without inventing a factual one. |
| A baseline answer is missing | Keep the gap and date any new attempt honestly. |
| The changed case repeats the old answer | Check whether the new approved input reached the job before changing the method. |
| A fresh conversation loses your place | Check the saved continuation and evidence paths; recover the missing step rather than silently rerunning everything. |
Before you finish
You can show which roles actually returned work and explain one answer comparison using source evidence. Your changed-case judgment remains yours. The original baseline is intact, and each recurring case has a clear next measurement. You can find unfinished work after reopening the conversation.
work/course/staff/before-picture.md) with the originals untouched and two new answer sets beside them, dated. Your dated Week 4 measure under each of the three cases. Your changed-case prediction and its result, with that job's records. Your place, saved, so next week starts where you left it.Next: Week 2, Make the organization work your way. Build on your approved placements, brief, and source plan as you develop the organization's knowledge. The Librarian's deeper context work should begin from your decisions, and the Week 1 comparisons remain available for Week 4.