Lab 9: put one AP job on a schedule

In this lab you take the role of Maria, the office manager at Brightline Wholesale Supply, the book's example company. You rebuild her Thursday job the right way: the review of open invoices before each Friday payment run. AP means accounts payable, the bills a company owes. The pre-run note is the summary that Dave, the controller, reads before he approves the payment run.

On Monday, November 9, 2026, Maria set two jobs running on their own. On Thursday, November 19, the Thursday run listed a Tri-County invoice as ready to pay while a request to change Tri-County's bank details was still open. Maria caught it just before Dave approved the payment run. On Friday, November 20, Dave asked to keep the Thursday review, but with a proper design. You test the new design on November 12 and 19 again, and on the next Thursday, November 26.

Why it matters. Work that runs without you is easy to trust and easy to get wrong. In this lab you predict what a run with no record misses, then build the spec, the record and the rules, and test them on three Thursdays.

Where you work. Most of the lab is in a folder on your computer. Download brightline-lab-ch09.zip from the Labs companion, unzip it, and open the files in any text editor, such as Notepad or TextEdit. Steps 2, 3 and 4 also use fresh chats in Claude or ChatGPT, and step 6 uses a scheduled task on one AI vendor, or on both if you do its optional port. No code or server is needed.

You can also do the lab with the Claude or ChatGPT desktop app. Open the folder in the app, and ask it to read LAB.md, the lab's instruction file, and start. You write the gates, the map, the spec and the rules, and you decide. The app writes your answers down. The runs still happen in fresh chats.

How long. About six and a half hours in total: the seven steps below add up to 380 minutes. Take them in three or four sittings. The optional parts of steps 6 and 7 add about 95 minutes. Each step ends with a file saved, so you can stop after any step.

What the lab gives you.

• Five AP jobs that Maria wants to run without her
• Maria's review as she does it today, step by step, with times
• Dave's request, and Maria's original instructions for the Thursday job
• The five approved AP concepts, short files that each hold the rules for one topic
• The open-invoice export, called the register, for each Thursday
• The SSoR record of AP matters as of November 12, and the entries people added between runs. SSoR, the State System of Record, keeps the dated history of each piece of work.
• Eight things a standing worker might do when nobody asked
• Answer keys, to open only when a step says so

What you do. Do the steps in order. LAB.md gives the full directions, one Part for each step. Read each Part when you reach its step, not all at once.

1. Predict (LAB.md Part A, 40 minutes, in the folder). Run the five jobs through the three gates, the three questions of Concept 9.1. Then predict what a fresh run will miss on November 19 if it reads only the register and the concepts. You save: results/gates.md.
2. Map the review (Part B, 45 minutes, in the folder and one fresh chat). Map each step of Maria's review: who does it next, and what supports it. Ask the AI to interview you about the review, one question at a time, and use its questions to check your map. You save: results/delegation-map.md.
3. Write the standing spec (Part C, 60 minutes, in the folder and one fresh chat). Start the SSoR record from the November 12 file. You keep it by hand from now on. Then write instructions that a run with no memory of earlier runs can follow. Ask the AI where such a run could go wrong, and fix up to three points you agree with. You save: workspace/standing-spec.md and ssor/ap-matters.md.
4. Run three Thursdays (Part D, 60 minutes, in five fresh chats). Run each Thursday in a fresh chat, as a scheduled run would start. On November 19, run three versions: first Maria's original spec, then your spec without the record, the stop test, which should stop, and last your spec with the record. Add each run's proposed entries to the record by hand. You save: results/week-run-log.md and the five notes.
5. Investigate (Part E, 40 minutes, in the folder). Only now, open the answer key for the runs. Score each note, and name the cause of each miss. Save the November 12 run as the spec's first example with a known answer. You save: the scores in results/week-run-log.md, and workspace/examples/2026-11-12/.
6. Modify (Part F, 90 minutes, in the folder, then a scheduled task on one AI vendor). Write the initiative rules. Put the job on a real schedule on one AI vendor, and set a deadline check, a check outside the run that its note arrived on time. If you choose, test that check, then port the job: put the same spec on the other AI vendor, run the stop test there, and record which settings had to change. You save: workspace/initiative-rules.md and results/schedule-settings.md.
7. Make (Part G, 45 minutes, in the folder). Write the six-part note to Dave. Add its triggers, what starts each job, and its initiative rules to the AP Worker's Role Contract, the one-page definition of its job, as Draft 6. If you choose, run one job from your own field through the three gates. You save: results/note-to-dave.md and role/ap-worker-role-contract.md, and results/my-job.md if you do the optional part. The lab is not finished until the first two are saved.

A good set of runs shows the design works when you give it the right files. It does not show that a real schedule will always get them. That is why the spec names its inputs, and why a missing input is an exit, not a guess.

Only one AI vendor? Do the runs on it, and skip the port. If you want to, write the transfer plan instead: what the other AI vendor would need.

Exam notes

These notes are not part of the lab. CCAO-F, Anthropic's Claude Certified Associate: Foundations, follows an exam guide written in July 2026, and its questions are about Claude only. Where this chapter teaches today's products and the guide expects a different answer, the note gives the answer that scores. Learn both. Certification and Portfolio Roadmaps has more.

• Work that repeats. The exam guide expects a project with saved instructions and knowledge, where Claude drafts and a named person reviews before anything is sent. For continuity between conversations, it expects memory or the project's knowledge. This chapter teaches scheduled and event-triggered runs, routines, standing workers and the SSoR record. The exam does not cover them. On the exam, choose the project and the human review. (D-007)

Artifact checklist

• results/gates.md, five jobs through the three gates, with a reason for each
• results/delegation-map.md, the Thursday review with an owner and a test for each step
• workspace/standing-spec.md, inputs, rules, output, exits and the SSoR read and write
• results/week-run-log.md, five runs: three scored, the stop test, and Maria's original spec, with a cause for each miss
• ssor/ap-matters.md, the record after November 26, with every entry's origin
• workspace/initiative-rules.md, eight cases graded and a rule for each kind of act
• results/schedule-settings.md, the real scheduled task's settings and its deadline check, and, if you did them, the check's test and the port
• results/note-to-dave.md, the six-part note
• role/ap-worker-role-contract.md, Draft 6, with triggers and initiative filled in
• results/my-job.md, one job from your own field through the three gates (optional)