Plan your email team's schedule, and how you can entrust your team to do more without you

Current Outlook route, 22 October 2026. Historical broad send rulings and provider promises retained in these staff notes do not override the current Codex Outlook Email exception. No agent sends on that route, including by browser. Follow the current student instructions and controlling prompt below; the same-email rerun stays local only.

Sara-standard pass (5 October, evening). Status lives only here. Student text states each supported path as a plain rule, with no probe ids and no "built, not tested" wording; where a step is not open yet, the student text says what the student does instead. The lesson is safe to follow with every row NOT RUN: the student plans the interval with the Chief, and nobody creates a scheduled task until the course's own test (8b) has passed for that app. "At release" is the change to make when the named probe row is PASS for that app and provider; rows with no probe row cannot change this release.

The one gate (AR6). The only place that turns a gated step on is the "What the course has tested in your app" table under Before you start on the previous lesson's page, course/workforce/week3/lessons/anw-w3-l03.md, which the student edition ships and the Chief already reads for the sending gate. This page points at that table by name (Part 1, "Why it doesn't switch on today"); it carries no table of its own. The schedule skill reads PROBES.md for its 8b line tonight and must be re-pointed at that table at the release edit (STAFF-NOTES, arbitration pass). The ladder page carries no test table in its student text; its test status sits in a staff block at the top of the page, a reading copy that turns nothing on.

The edit class (AR8). The person never grades a send. When the Chief sends on their words, the send check records how much they changed the draft, and anything the record cannot tell counts as rewritten; the Chief never decides light or rewritten itself (the sweep skill and the Chief body, "Sending, supervised"). Part 2 and its outcome table say so, and the Light and Rewritten rows are the ladder page's rows, word for word. One sending path only (Tyler's ruling, 5 October, 5:15 pm CT): the Chief sends on the student's yes; nothing on this page tells the student to send by hand.

Open mismatch carried from the trim pass: the trimmed page body said the full stop events (6d1 to 6d8) and the off switch (6e, 6f) were tested in Claude Code, while rows L4-9 and L4-10 and the ladder page said NOT RUN. The student text no longer states either; check PROBES.md at the go/no-go.

Patch notes applied: no Haiku named; the ladder page's test status (WS11) lives in that page's staff block and is pointed at, not copied; the Chief does not hold the app's scheduled-task tools (WS10) and this week does not guide task creation (orchestrator ruling B). The details page (course/workforce/week3/reference/anw-w3-l04-details.md, source overrides/course/week3/reference/anw-w3-l04-details.md) has its source-plan entry and renders to student/agent-workforce/course/workforce/week3/reference/anw-w3-l04-details.md; the render resolves every link from this page to it, anchors included.

What has been reviewed, and what has not

DRAFT, built against the Exercise 1 branches on the night of 4 October. Nothing on this page had passed a live probe when it was written. Per orchestrator ruling B (fix round 2), the page does not instruct creating a scheduled task: the student plans the interval with the Chief, and the course turns the create step on for an app after probe 8b passes for it, with the exact steps 8b actually ran. The by-hand steps from aibl-inbox-schedule (ws10) and its references/app-mechanisms.md are kept in the staff block below for the release edit, not on the student page. The ladder (Part 3) is WS11's autonomy-ladder.md, whose test status sits in a staff block at the top of that page (8 October: the student text states plain rules, with no test table, probe ids or hedges); this page points at it instead of copying it, so one status block is updated at release.

Not run against a student workbench. No paste block has been run end to end. The controlling prompt is a draft for Hunter's prompt pass.

Rulings this page is built on

Plan section 3.1 step 8 (schedule after three clean rounds), 3.11 (earned autonomy, per category), 3.12 (scheduling), rulings R2, R12, R27, R30, R37; build addendum item 9 (off switch path). Interface page 1.6 (off switch), 5.1 and A1.11 (schedule.json), 9.8 (the task table, filled by probe 8b). WS10's ruling: the Chief does not hold the app's scheduled-task tools; when creation turns on, the owner creates the task by hand, guided one step at a time. WS10 and WS3b's ruling on WS11's open question: a scheduled wake drafts only for categories at rung 2 or higher; rung 1 emails are listed under Needs you as "waiting for a manual run". The page says that plainly, because on day one every category is on rung 1, so a schedule only gathers and lists mail until the owner moves a category up.

The create steps, for the release edit when 8b passes (from aibl-inbox-schedule, not student-facing tonight). Prompt line for both apps: Check my email (scheduled wake); run the inbox sweep skill; if the ledger says busy, stop. Claude Code desktop: Code tab, Routines, New routine, Local; name check-my-email; the prompt as instructions; the workbench folder; worktree off; the picker offers Hourly, Daily, Weekdays, Weekly, and any other interval is set by asking a Claude Code session in the workbench with the exact schedule rule inbox_schedule.py check printed; then Run now once, approving read and draft tools only. Codex desktop: in the project's chat, ask for a standalone scheduled task, local folder (not a worktree), new chat each run, with the zone-bound rule (fallback: the bare rule in the computer's clock); check the Scheduled section. Both: inbox_schedule.py record with the owner's words and the exact rule; the ledger must show one wake.start with mode scheduled. Use only the path 8b actually ran for that app, with the labels 8b recorded.

The Professor part. Plan WS16 row: "Professor routing". Part 4 is written from the Professor's own body (aibl-the-professor: explains, defines, checks evidence, prepares a blocker; hands anything that must be done to the Chief). No new behavior is promised for the Professor.

Sources

aibl-inbox-schedule with references/app-mechanisms.md (ws10), autonomy-ladder.md and autonomy_report.py (ws11), the Chief body (ws3b: "Scheduled wakes", "Earned autonomy and the off switch"), aibl-inbox-sweep (ws3b), the Professor's body (merged), work-orders/standing/README.md (ws11).

Dependencies

Release (WS17): autonomy-ladder.md must render to course/workforce/week3/autonomy-ladder.md (the Chief copies it to the owner's records on the first wake); the lesson map titles; the prompt files. Probe 8b decides the exact create steps per app and whether a scheduled Claude session behaves as the Chief with the one-line prompt (INTERFACES 9.8); until it passes for an app, the Chief should decline to guide task creation for that app (a line for WS10's skill or the Chief body: see STAFF-NOTES open question 12).

Arbitration pass (Monday 5 October, early morning CT). AR6 and AR8 are as stated at the top of the promise register. AR9: the off switch paragraph says a name starting autonomy-off counts (the ledger's rule); the ladder page now says the same (checked at the trim pass).

Sara-standard pass (5 October, evening). Sara's review: every part opens with what the student does and gets; machine internals, safety mechanics and edge cases move to the details page. Parts are now: 1 the schedule, 2 the trust ladder, 3 the off switch, 4 who to ask. No probe ids or test status in student text; the rounds are named as observational evals with TOAST; no lesson numbers in student text or the paste prompt (the Professor routes by the lesson map's id and title and the prompt names the files by path); "working hours", "your work", no "owner" or "business"; "the record", "category" and "rung" explained where first used; every "say this" step names its conversation. Every student step and safety rule is kept: the seven text boxes are the same requests (the first Chief request says "observational eval rounds", the second "working hours"), and the off switch's two-step stop and two-step start are unchanged. The controlling prompt follows the new part order and drops "owner"; its routing lines are unchanged.

• New title (Sara, 6 October): "Plan your email team's schedule, and how you can entrust your team to do more without you". The start prompt still quotes the old title, because prompts stay word for word: Gigawatt, please update the start prompt, the prompt file and the lesson map to the new title.
• "Autonomy ladder" (Sara, 6 October): the page uses the agents' own name (the file is autonomy-ladder.md, titled "Your autonomy ladder"), with rungs. The Part 3 prompt to The Professor still says "trust ladder", and so does the controlling prompt file prompts/anw-w3-l04.txt (line 3: "A rung is one of the four levels on the trust ladder"), so The Professor will say "trust ladder" to students. Gigawatt, please change both to "autonomy ladder", and update the old lesson title in the prompt file's first line.
• Copy-paste prompts unchanged, all six, word for word.
• Framing: the new-hire comparison in the opening; every part opens with what you do and what you get; part titles say what the part is about on their own.
• Before you start: no longer says Gigawatt confirmed the rounds; Part 1's prompt has the Chief check the record, so the page now says that once.
• Part 1: "clean" defined as "went well" in plain words (mix of three to five, nothing unexplained). The cloud-schedule sentence cut (later step, not needed here).
• Part 2: "Kansa checks drafts for five older emails" explained as a spot check; tables in plain words; "Full stop" removed from Words you now own (the term is no longer used on the page).
• Part 3: "your app's own stop" explained as the app's stop button and its question before every send. CHECK with Gigawatt that this is what "your app's own stop" means.
• Part 4: "the weekly model check" removed from the student table (unexplained); the who-to-ask table trimmed; the password rule now says tell The Professor (a course question), not Gigawatt.
• Times: parts total 40 minutes, as before.