Lesson 01 — Diagnostic: Rollout Readiness, On His Real Plan

Generation-time decisions (logged, per Playbook §3–5)

His real artifacts (what this whole lesson runs on)

1. rollout-plan.docx, Phase 1 Pilot Workflows (verbatim excerpt): "(1) Grants team will use Claude to help draft the Outcomes section of the quarterly grant report. (2) Development team will use Claude to help draft donor acknowledgment letters. (3) Program team will use Claude to help draft monthly program-outcome summaries for the board. All three pilot workflows will run in Cowork Auto mode to reduce staff friction. Roles: Devon T. = Owner; three team leads = Admin; program staff = User."
2. data-handling-policy-draft.docx (verbatim, currently one paragraph): "Staff should not upload sensitive donor or client information into Claude. IT will set a 90-day retention window so nothing lingers longer than necessary. Connectors to our shared drive will be enabled for convenience so staff don't have to re-upload files each time."
3. grant-report-template.docx, Q3 Family Support Outcomes Report, Outcomes section drafted by Claude from family_services_q3_export.xlsx: "Served 3,200 families in Q3, a 14% increase over Q2. 89% of enrolled families completed at least one case-management session. Distributed $145,000 in direct emergency assistance."

Learning objective

(Bloom's: Analyze) Classify your rollout plan, data-handling policy draft, and grant-report template against five readiness practices — specification, delegation, elicitation, verification, failure-anticipation — to locate, in your own real documents, where each currently stands.

SSD cycle 1 — the rollout plan, read for specification (AIHC.1.A1)

• SAY: A real specification states goal, inputs, constraints, and what "done" looks like before the work is handed off — and gets revised, not just re-rolled, when the output misses (AIHC.1.A1). Your rollout plan's line — "Grants team will use Claude to help draft the Outcomes section" — names a task, not a specification. It has no stated goal (what should the section do for the funder?), no named inputs (which file? family_services_q3_export.xlsx isn't mentioned), no constraints (word count? which claims need receipts?), and no done-criteria (what makes a draft ready to send to you for review?).
• SEE: A two-column static table — left, your plan's exact sentence for each of the 3 workflows; right, four empty checkboxes per row (☐ goal ☐ inputs ☐ constraints ☐ done-criteria). (Static; nothing moves — this is a document-annotation exercise, not a process to animate.)
• DO: For your Grants-team line, check off which of the four elements is actually present today. Recommended path: start with Grants, since it's the one about to touch the artifact you already have in hand (the grant-report template). (Alternatives — starting from Development or Program instead — are available on request; not listed as a menu here.) Expected finding: 0 of 4 checked. That gap count is your AIHC.1.A1 baseline reading, not a grade.

SSD cycle 2 — the "Auto mode for all three" line, read for delegation (AIHC.1.A2)

• SAY: AIHC.1.A2 asks you to decide whether Claude handles a task "from stated reasons about capability and the cost of being wrong — not novelty, habit, or hype." Your plan's line — "All three pilot workflows will run in Cowork Auto mode to reduce staff friction" — is a single blanket decision covering three workflows of very different sensitivity (a board summary, a grant report headed to a funder, and a donor-letter workflow that touches individual donor data). "Reduce staff friction" is a convenience reason, not a capability-and-cost-of-being-wrong reason (doc-set §7 names Auto's actual behavior: Claude "reviews each action for safety... and blocks anything it determines to be unsafe," but that is a safety check, not a stand-in for your own judgment about which of the three workflows can tolerate an unreviewed miss).
• SEE: A 3-row table — Workflow | Current plan says | Reasoned, or default? — with the middle column filled in (Auto, Auto, Auto) and the third column left blank for cycle 2's DO.
• DO: For each of the 3 workflows, mark the third column "reasoned" or "default" — is there a stated reason in the plan, or is "Auto for all" doing the thinking? Expected finding: all three marked "default." This is your AIHC.1.A2 baseline; Lesson 03 rebuilds this table with real reasoning per workflow.

SSD cycle 3 — the retention-window sentence, read for verification (AIHC.1.B1 + a flagged gap)

• SAY: AIHC.1.B1 asks you to check a claim against an independent source before acting on it — and to reject your own (or IT's) confident phrasing as a substitute for that check. Your policy draft states, as settled fact: "IT will set a 90-day retention window." Before this sentence governs your team, it needs to be checked against what the admin console actually offers — not assumed from how confidently it's written.
• SEE: The sentence, annotated with one flag: ⚠ unverified against doc-set §5 (Data controls: retention, audit logs, "Covered Models" policy).
• DO: Don't resolve this yet — that's the point of a diagnostic flag, and it's the load-bearing move Lesson 06 completes in full. Mark the sentence "flagged — verify before it governs staff behavior" rather than either trusting it or guessing. (Preview, not yet taught: the doc-set records custom retention controls as an Enterprise-only feature (§5) — your org is on Team. Hold that thought for Lesson 06.)

SSD cycle 4 — the grant-report Outcomes section, read for elicitation (AIHC.1.A3)

• SAY: AIHC.1.A3 asks you to solicit the AI coworker's own read — its proposal, its uncertainty — before finalizing, treating "what would you do?" as a working question, not a courtesy. Your current grant-report workflow is: Claude drafts → you review → it ships. There is no step where you ask Claude which of its own claims it's least sure about.
• SEE: A simple pipeline diagram of your current workflow — draft → review → submit — with a dotted box "ask Claude what it's unsure about" shown not yet present between draft and review.
• DO: Mark whether that step exists in your workflow today (it doesn't). That absence is your AIHC.1.A3 baseline; Lesson 04 builds the step.

Independent at-bat

Run the same four-probe pass — specification, delegation, verification-flagging, elicitation — on the Development team's donor-acknowledgment-letter workflow, which cycles 1–4 mentioned but did not detail. Produce your own gap table (4 elements × 1 workflow) and a one-line delegation-reasoning verdict, unscaffolded.

Exit ticket (Bloom's climbs to Analyze; graded against the doc-set + your own artifacts)

1. (Remember) Name the four elements a real specification states before work is handed off, per AIHC.1.A1.
2. (Understand) In one sentence, why is "Claude will help draft the Outcomes section" not yet a specification?
3. (Apply) For the donor-letter workflow (your at-bat), write one missing specification element it currently lacks, stated the way a real spec would state it.
4. (Analyze — objective level) Across your three real artifacts, which of the five practices (A1 specification, A2 delegation, A3 elicitation, B1 verification, B2 failure-anticipation) is your weakest starting point right now, and what is the one-line evidence from your own documents that proves it — not a guess, the actual sentence or absence you found?
5. (Analyze) Rank your five baseline readings from strongest to weakest, in one line each.

Ledger write

ledger_write:
  learner_id: L2-ADMIN-DEVON
  lesson_id: L2-admin-devon-01-diagnostic-20260719
  anchors: [AIHC.1.A1, AIHC.1.A2, AIHC.1.A3, AIHC.1.B1, AIHC.1.B2]
  exit_ticket: {score: "", bloom_reached: analyze}
  baseline_mastery_established:
    - {standard: AIHC.1.A1, band: 1, mastery: 0.50, source: "carried from profile, reconfirmed by cycle 1 finding (0/4 elements present)"}
    - {standard: AIHC.1.A2, band: 1, mastery: 0.20, source: "new — cycle 2 finding (all 3 workflows: default, not reasoned)"}
    - {standard: AIHC.1.A3, band: 1, mastery: 0.10, source: "new — cycle 4 finding (elicitation step absent from workflow)"}
    - {standard: AIHC.1.B1, band: 1, mastery: 0.30, source: "new — cycle 3 finding (retention claim asserted, not checked)"}
    - {standard: AIHC.1.B2, band: 1, mastery: 0.25, source: "new — no incident vocabulary yet observed in his artifacts"}
  auto_score: ""
  self_score: ""
  journal_prompt: >
    Of the five gaps you found today, which one worries you most as the person whose name is on
    this rollout — and why that one, not the others?
  structure_used: PBL
  referents_used: []   # diagnostic is artifact-only; referents enter starting Lesson 02
  next_lesson_seed: "Lesson 02 rebuilds the Grants-team specification from scratch, element by element, on this exact plan."

RUBRIC SELF-AUDIT (against Exemplar_and_Rigor_Rubric_Cowork_x_Nonprofits_2026-07-19_v01_I.md)

Escalation check: no load-bearing indicator fails → auto-ships per the versioned escalation policy (Playbook §5).