AI Build Lab / Learn · Agent Workforce · Week 4 · Module 06

Build your proof page

What does my organization do now that it didn't before, and can I show it on one page?

Staff only · this page and its checks
Source
Internal's lesson, src/week4/internal/lessons/anw-w4-l11.md (Internal 60d90336, 2026-10-19 19:39), read by md_week4.py.
Copy
Internal's Markdown as published to the student shelf 10-12 (Internal 101e2662a = student 8275f21b; PR #215, receipt #218), vendored byte for byte by vendor_week4.py from the student-edition overrides; the published student tree is the same files without the "Staff only" blocks (checked on import). The drawings are the inline SVG figures in the lesson files, used as written.
Prompts
Word for word from the lesson file Internal ships (course/workforce/week4/). Not run end to end from Learn.
Checked in the app
Not checked in the app from Learn.
Open to students
Yes.
Other view
one scroll (one part at a time is the lesson; one scroll is kept to compare)
Start hereWhat this lesson is, and how to begin
Staff review only: full controlling prompt. Students paste only the short request below.

Use the native aibl-the-professor agent to guide the Week 4 lesson Build your proof page. Read course/workforce/week4/lessons/anw-w4-l11.md completely, its details page course/workforce/week4/reference/anw-w4-l11-details.md when asked for more, and the native aibl-course-guide and aibl-lesson-map methods for this app. You teach and read, and make one save that is yours: the student's two measures. You never build, edit or approve the proof page, run a test, supply or round a number, or make the PDF for the student. The student talks to the Chief of Staff and to the Evaluation Analyst (Evy) in their own threads. Do not call, spawn or route to any agent. Edit no file other than the save below.

Use these words exactly. Every golden case must pass before the page is built: the Chief of Staff reruns the golden set, the student asks Evy directly to grade the reruns (Evy never takes grading work from the Chief), and a failing test sends the student through the lessons Pick the problem and agree the fix and Make the fix, rerun the golden set, and keep or undo it once more for that one problem; after two rounds the page may show the test failing with the student's reason, never hidden. The two measures are the student's own minutes on one piece of the work: the before-number (from the starting point in the Week 3 design, or, for a student with only the email team, the student's own estimate written today and labelled as an estimate with its date) and the now-number (the student's own timing of the next three jobs). Both measure the same stretch: from opening the email (or whatever starts the job) to having a reply you'd send; the team's own working time is not counted. Your save, only after you show the exact entry and the student says yes in their own words, quoted word for word: one dated entry appended to work/course/staff/progress.md with both measures in the student's words, every earlier entry kept unchanged, read back after writing. With no timing yet, the page says "not measured yet". The proof page, put together by the Chief of Staff from the student's records, holds: the before-number beside the now-number; how much of the team's work went out without edits, lightly edited, or rewritten; how often the student and Evy agree, from the counting script; every golden case and whether it passes, from the latest rerun; and what changed and why, from the change record. Every number names its record; no agent types or estimates a number. The page can be printed as a PDF with the aibl-bangerpdf method. Name agents job title first, nickname in brackets on first use; Evy is "it". Say "your work", never "business"; refer to other lessons by title, never by number.

The Hattie B's practice inbox is the worked example on the page. Never reveal or paraphrase any answer key for the practice emails; use only what the lesson page shows.

Ask open questions only. Never word the "what changed and why" lines for the student, and never estimate a time for them.

Part 1. If asked, read the rerun table the student shows, read only, and say which tests fail. Explain the loop back through the two Module 3 lessons.

Part 2. Help the student name where their before-number comes from. Show the exact progress entry with both measures in their words, and save it on their yes.

Part 3. If asked, explain each part of the proof page and which record it comes from.

Part 4. If asked, help the student check one number at a time against the record it names: open the record, read only, and say whether the number matches. Never correct a number yourself; the student tells the Chief.

Part 5. If asked, explain how to keep the PDF and that it is the student's to share or not.

Otherwise, if asked, explain the step and what the student should see.

Staff only: status, sources, what the build supports. Not student-facing.

Status: second draft, 9 October 2026, lane "lessons" (dispatch 51bbe9da), after the Codex and Kimi reviews of PR #201. From spine v2 Module 6 (anw-w4-l11: the before-number from Week 3 against now; how much work goes out without edits; how often they and Evy agree; every golden case passing; what changed and why; the numbers come from their records, never typed by an agent).

What changed in this draft.

  • Every golden case passing (spine outcome): Part 1 makes it a step. A failing golden case sends the student round the Module 3 loop once more before the page is built. The page is built from the latest rerun only after every golden case passes, or the student says plainly that one still fails and why (the page then shows it failing; it never hides it).
  • Before and now. A designed team has its starting point in the Week 3 design. An email-team-only student has none: Part 2 has them write their own estimate today, dated, labelled as an estimate, saved by the Professor as a course-progress note in work/course/staff/progress.md (the Professor's existing may_write). "Now" is measured the same way, by the student timing their own review of the next three jobs, saved the same way. With no measure, the line says "not measured yet".
  • The PDF: the Chief of Staff holds aibl-bangerpdf (PR #205, Chief's skills in the registry). If it is missing, the student prints the page from the app.
  • A certificate is spine open item 1 (Sara with Tyler). The page doesn't mention one.
  • The Hattie B's page below is a made-up worked example.
Figure IDLessonWhat it showsCaptionAlt text
w4-l11-proof-pageanw-w4-l11 Part 3A drawn one-page document with five labelled blocks: before and now; how much went out without edits; you and Evy agree; golden cases passing; what changed and why. A small tag beside each block names the record it comes fromOne page, every number read from a record that's named beside it.Your proof page has five parts: your starting point beside the same measure now, how much work went out without edits, how often you and Evy agree, whether every golden case passes, and what changed this week and why. Each number names the record it came from.
In plain English

This week you graded your team, fixed a problem, and set your organization up to check itself. Today your Chief of Staff puts it all on one page: what your work took before, what it takes now, and what changed.

First you make sure every golden case passes, and you measure the two numbers the page leads with. Every number on the page comes from your records, and names the record it came from. No agent types or guesses a number. If there's no record for something yet, the page says so.

What you get

One page showing what your organization does now, compared with before, which you can print as a PDF.

Why it matters

"It's working" is a feeling. A page with numbers you can trace is proof, for you and for anyone you show it to.

How it works

The Course Companion (The Professor) guides you and saves your two measures on your yes. Your Chief of Staff reruns your golden set and builds and prints the page.

Your part

You fix what still fails, time your own work, check every number against its record, and decide whether the page is right.

Before you start

  • This week's records. Your grades, Evy's grades, the comparison, your fix and its rerun, and your shelves.
  • Three jobs your team will do soon, for timing yourself in Part 2. If you're doing the week on your email team, practice emails count.
  • In Codex, start your first message to an agent in a thread with "Ask the <agent> to do this, and show me its reply word for word:". Start every message after that with "Tell the <agent> this, word for word, and show me its reply word for word:".

Start or resume this lesson

Open your prepared workbench and call The Professor by its role name, aibl-the-professor. Paste only this.

Use the native aibl-the-professor agent to start the Week 4 lesson Build your proof page. Read course/workforce/week4/prompts/anw-w4-l11.txt and course/workforce/week4/lessons/anw-w4-l11.md. Ask me one question at a time and wait. I will talk to my Chief of Staff in its own thread.

The full instructions your agent follows for this lesson are in course/workforce/week4/prompts/anw-w4-l11.txt.

Part 1 of 5Make every golden case pass· 10 min, plus agent time

You check the latest rerun of your golden set. You get every one passing before the page is built.

On the Hattie B's practice inbox

After the first fix, two golden cases still failed: the cold-food complaint, whose draft offered a refund, and Marcus and "Saturday", whose draft signs off in a different voice (the mistake the student wrote down in "Save your golden examples, and turn mistakes into regression tests"). The student went round the loop for the refund: Gigawatt's plan was one line for the writer, the remedy stays a blank for the owner. The complaint passed, nothing else broke, and the student kept it. Then they went round twice for the sign-off. Both plans fixed the sign-off but changed the greeting too, so a test that passed before failed, and both were undone, each written down in the change record. Four of five golden cases pass, and the page says which one doesn't, and why.

Do this now: check your latest rerun

In a new thread with your Chief of Staff, say this, with your team's name:

Rerun the golden set for my <team name> team, and run the one it must refuse three times, and tell me the job folders when they're done. Don't send or publish anything.

When it's done, ask Evy in Evy's own thread: "Grade every job from my latest rerun of the <team name> golden set against my newest rule sheet." Then ask your Chief: "Show me every golden case and whether it passes, from Evy's grades."

If every test passes, go to Part 2. If one fails, do "Pick the problem and agree the fix" and "Make the fix, rerun the golden set, and keep or undo it" again for that one problem, then come back here. If you've been round twice and it still fails, you can build the page anyway: it shows that test failing, with your reason, and never hides it.

Part 2 of 5Measure before and now· 8 min

You write down how long the work took you before, and time how long it takes you now. You get the two numbers the page leads with, from your own records.

The same measure, twice

The page compares like with like: your own minutes on one piece of this work.

Both numbers measure the same stretch: from opening the email (or whatever starts the job) to having a reply you'd send.

  • Before. If you designed your team with the Architect (Archie), your design says how long the work took you; check it measures that same stretch, and if not, say so when you save it. If you only built your email team, there's no before-number on file yet: write it now, from memory, for that stretch, and the page labels it as your estimate, with today's date.
  • Now. Time yourself on the next three jobs your team hands you, over the same stretch: start the stopwatch when you open the email, and stop it when the team's reply, with any edits you make, is one you'd send. Your team's own working time isn't counted, because you're free while it works.

On the Hattie B's practice inbox

The student wrote: "Before the team, one customer email took me about 15 minutes from opening it to a reply I'd send." Then they timed three emails the same way, from opening each one to the team's reply being ready to send: 5, 3 and 4 minutes. The Professor saved both, with the date.

Do this now: save your two measures

In your Professor thread, say:

Save my before-number for my <team name> team: <minutes, and where it comes from: my design, or my estimate today>. And my three timings now: <minutes for each>.

Say yes when the Professor shows you what it will save. If you haven't timed three jobs yet, save the before-number now and come back with the timings; until then the page says "not measured yet".

Part 3 of 5Have your Chief build it· 9 min, plus agent time

You see the five parts of a proof page, then ask your Chief for yours. You get a first version, built from your records.

One page, every number read from a record that's named beside it.
Your proof page has five parts: your starting point beside the same measure now, how much work went out without edits, how often you and Evy agree, whether every golden case passes, and what changed this week and why. Each number names the record it came from. What my organization does now 1. Before, and now 2. How much went out without edits 3. You and Evy agree 4. Golden set: all passing? 5. What changed, and why your two measures your review records the counting script the latest rerun the change record

On the Hattie B's practice inbox

What my email team does now

1. Before: about 15 minutes per email, opening it to a reply I'd send (my estimate, written 9 October).
Now: 4 minutes on average, measured the same way over three emails (my timings, 10 October)
2. This week's drafts I approved: 6 untouched, 3 lightly edited, 1 rewritten (my review records)
3. Evy and I agree: whole job 19 of 20 rule grades; on the ones I failed, 4 of 4 (the counting script)
4. Golden set: 4 of 5 pass. Failing: Marcus and "Saturday", the sign-off. Two fixes tried and undone:
each one broke the greeting. Next round after the course. (the latest rerun; the change record)
5. What changed: the writer leaves food-fact gaps and remedies for me. Why, in my words:
"a wrong allergen fact could make someone sick; refunds are my call." Both kept. (the change record)

Do this now: ask for your proof page

In your Chief thread, say this, with your team's name, or "every team":

Build my proof page for <team name>: my before-number beside my timings now, how much work went out without edits, how often Evy and I agree, every golden case and whether it passes, and what changed this week and why. Read every number from my records and name the record beside it. Don't type or estimate any number: if there's no record, say "not measured yet".

Part 4 of 5Check every number against its record· 8 min

You check each number on the page against the record it names. You get a page you can stand behind.

Do this now: check each line

Take the page one line at a time. For each number, in the same Chief thread, say:

Show me the record behind line <number>, and the exact place the number comes from.

Does the number match? If it doesn't, tell your Chief which one, and ask it to read it again from the record. Never accept a number that isn't in a record. Check the "why" is in your own words, as you said it when you kept the fix.

Part 5 of 5Save it as a PDF· 5 min

You save the finished page. You get a PDF you can keep, print or show.

Do this now: make the PDF

When every line checks out, say in the same Chief thread:

The page is right. Save it as a PDF and tell me where it is.

What you should see

The page as a one-page PDF, saved in your workbench, and the folder it's in. It's yours: nothing is sent or posted anywhere.

If you get stuck

What you noticeWhat to do
A golden case still fails after two roundsBuild the page anyway. It shows the test failing, with your reason. Come back to it after the course with the same loop.
A line says "not measured yet"That's honest. Time three jobs, save the timings with the Professor, and rebuild the page.
Your before-number is from memoryThat's fine. The page labels it as your estimate, with the date you wrote it.
A number doesn't match its recordTell your Chief which line. The page reads from records, so a mismatch is a fault to fix.
You want to round a number or leave one outLeave it as the record says. You can choose not to show the page to anyone; you can't change what it says.
The PDF doesn't saveAsk your Chief what stopped it. You can also print the page from your app as a PDF.

What each line reads, and how the PDF is made: more detail on this lesson.

What you'll keep

Every golden case passing, your before-number and your timings now, and your proof page, with every number traced to a record, saved as a PDF.

Before you finish

Before you finishFour checks
Self-assessed. No gate.

Self-assessed. No gate.

Words you now own

  • Proof page: one page that shows what your organization does now compared with before, every number traced to a record.
  • Before-number: how long one piece of the work took you before your team, from your design or your own dated estimate.
  • Not measured yet: what the page says when there's nothing to read a number from.

Next: Move your organization from class to work.

AI Build Lab · LearnAgent Workforce, Week 4, module-6, lesson-1