Lesson 01 — Diagnostic Baseline On Her Real Workflow

Learning objective (Bloom's: Analyze): Analyze your own research-to-draft workflow — how you specify requests, delegate work, elicit Claude's input, verify claims, and handle mistakes — against five AI-collaboration practices, to establish your starting band and mastery on each before any lesson scaffolds you.

Standards diagnosed: AIHC.1.A1 (Specification) · AIHC.1.A2 (Delegation) · AIHC.1.A3 (Elicitation) · AIHC.1.B1 (Verification) · AIHC.1.B2 (Failure literacy) — all Band 1 (Assisted Use). (AIHC.1.C1, provenance, already has a tracked baseline from prior work (mastery 0.25) and stays out of this diagnostic's scope so the five practices this lesson set covers get full attention — not a gap, a scoping choice.)

Structure: PBL — her actual three artifacts (the longform draft "The Last Mill," her interview notes, and her pitch tracker) ARE the diagnostic instrument; there is nothing to simulate. Band 1; timing-tolerance accommodation honored: each cycle below is a self-contained pause point — stop after any cycle and pick back up later with the state summary intact.

Trust boundary, named up front (R13): this diagnostic has you re-reading your own past chats and files. The doc-set states Projects keep a separate knowledge base per project and that "context is not shared across chats within a project unless the information is added into the project knowledge base" [S04] — so if your three artifacts live in different places right now, that scatter is itself diagnostic data, not a bug to apologize for.

Cycle 1 — A1 Specification: spec'd ask or vague ask?

• SAY: A spec'd ask names four things: the audience, the job the text must do, what's fixed, and what's open. The doc-set's own advice for a non-developer user is blunter: "Explain your ask simply and clearly" (Clarity), give Claude "as much context as possible… pretend you are giving these instructions to someone with no background knowledge" (Context), "Break down complex requests into substeps" (Structure), and treat a weak first answer as a cue to add context, not abandon the prompt (Iteration) [S17]. Those four rules ARE the four-part spec, renamed for a newsroom.
• SEE: A described two-column rubric: left, a representative (labeled fictional, not one of her real messages) vague ask — "make this sound better" — scored against the four rules (0 of 4 present); right, the same ask rewritten spec'd, scored 4 of 4.
• DO: Open your own real chat history and pull your last three actual asks about "The Last Mill." Score each against the four-rule rubric (0–4). (Evaluation: on-demand — the tutor checks your scoring against the rubric's own four named elements [S17], not against how the ask "felt.")

⏸ Pause point. Banked: three real asks, scored. Safe to stop here — nothing downstream needs this cycle re-run to resume later.

Cycle 2 — A2 Delegation: named split or "whatever's fastest today"?

• SAY: Delegation is a deliberate split — named Claude-side tasks, named Priya-side tasks, and a named way you'll check the handoff — not just routing work to whichever of you is free. Confusing "I could ask Claude to do this" with "I should" runs both directions: handing off editorial judgment you alone hold fabricates a boundary nobody set; sitting on legwork Claude could safely run burns your scarcest resource, your own attention.
• SEE: A described split-table for a one-person newsroom: Claude-side — organizing transcripts, drafting boilerplate pitch follow-ups; Priya-side — which anecdote leads, which source to trust more, final wording of the nut graf; check-method — how she'll confirm the handoff before it ships.
• DO: Think back to the last time you handed Claude a chunk of your interview notes to "make sense of." Did you name the split (what Claude decides, what you decide, how you'd check) — or just hand it over? Write down what actually happened. (Evaluation: presence/absence of a named split, not whether the output was good — this diagnoses the practice, not the result.)

⏸ Pause point. Banked: one real delegation moment, classified. Come back anytime.

Cycle 3 — A3 Elicitation: did you ever ask Claude what IT thinks?

• SAY: Elicitation means asking Claude for its own read, its proposal, or its uncertainty before you finalize anything — treating "what would you do?" as a working question, not a courtesy. It's the practice most people skip entirely, because it feels like the AI's job is to answer, not to be asked.
• SEE: A described transcript contrast: a session with zero elicitation turns (every message is Priya asking, Claude answering, nothing asked back the other way) next to one elicitation turn added at the end — "what's the weakest part of this section?"
• DO: Scroll your most recent real drafting session on "The Last Mill." Search for any moment you asked Claude something like "what would you cut," "what are you least sure of," or "what would you do here." Count them. (Evaluation: a literal count — zero is a valid, honest diagnostic result.)

⏸ Pause point. Banked: an elicitation count for your most recent session.

Cycle 4 — B1 Verification: checked against a source, or against a feeling?

• SAY: Verification means checking a claim against your actual interview notes or an outside source — not against how confident Claude's sentence sounds. Confidence is a writing style; it is not evidence.
• SEE: A described sampling box: five claims pulled from a real passage, each with a verified / unverified / no-check-attempted tag.
• DO: Pull five factual claims from your current real draft of "The Last Mill." For each, answer honestly: did you already check this against your interview notes, or not? Tag each. (Evaluation: tags checked against whether a matching note actually exists — not against your confidence that you "probably" checked it.)

⏸ Pause point. Banked: five claims, tagged.

Cycle 5 — B2 Failure literacy: diagnosed into a rule, or deleted and forgotten?

• SAY: When Claude gets something wrong, there are two responses: delete the bad sentence and move on, or diagnose why it happened and write a standing rule that catches it next time. Only the second one changes your risk going forward.
• SEE: A described three-column log format: mistake → cause → rule, with one row filled in as a worked example (label: representative, not her real mistake).
• DO: Recall the last time Claude produced something wrong in this project — a misattributed detail, a fact that turned out stale, anything. Do you have a written rule from it, anywhere? If not, that's your honest answer. (Evaluation: does a rule exist in writing, yes or no — memory of "I'll be more careful" doesn't count as a rule.)

⏸ Pause point. Banked: your fifth and final diagnostic data point.

Independent at-bat

Pick one upcoming, real task from your pitch tracker (for example, drafting a follow-up to an editor who's gone quiet). Run it once, start to finish, self-observing across all five practices at once — where did you notice yourself specifying, delegating, eliciting, verifying, diagnosing failure, and where did you not notice yourself at all? No template this time; you built all five lenses in the cycles above.

Exit ticket (climbing to Analyze — the objective's level)

1. (Remember) Name the five AI-collaboration practices this diagnostic just checked.
2. (Understand) Why does "verifies sporadically" carry a different risk for a byline journalist than for someone using Claude to plan a dinner party? One sentence.
3. (Apply) Here's a description of a typical ask you make: you say what you want changed but not who it's for or what must stay fixed. Classify it against the four-rule A1 rubric and name the missing element(s).
4. (Apply) Of your five scores from today, which is weakest — and what's the very next real task where you predict you'll notice it?
5. (Analyze — objective level) Looking across all five practices, what's the one pattern connecting your two lowest scores? Ground your answer in something you actually found in a cycle above, not a guess about yourself.

Ledger write

ledger_append:
  learner_id: L3-CONS-PRIYA
  lesson_id: L3-cons-priya-01-diagnostic
  standards:
    - {standard: AIHC.1.A1, mastery_before: 0.60, mastery_after_simulated: 0.60, note: "confirmed, not moved — diagnostic only"}
    - {standard: AIHC.1.A2, mastery_before: null, mastery_after_simulated: 0.35, note: "NEW baseline established this lesson"}
    - {standard: AIHC.1.A3, mastery_before: null, mastery_after_simulated: 0.20, note: "NEW baseline established this lesson — her lowest"}
    - {standard: AIHC.1.B1, mastery_before: 0.45, mastery_after_simulated: 0.45, note: "confirmed, not moved — diagnostic only"}
    - {standard: AIHC.1.B2, mastery_before: null, mastery_after_simulated: 0.30, note: "NEW baseline established this lesson"}
  bloom_reached: analyze
  auto_score: ""
  self_score: ""
  structure_used: PBL
  referents_used: []   # diagnostic kept referent-neutral; flavor referents begin Lesson 02
  journal_prompt: >
    Of the five practices you just measured yourself on, which result surprised you most — and
    why do you think that one, specifically, is the one you didn't already know about yourself?
  next_lesson_seed: "scaffold heaviest on A3 (0.20) and B2 (0.30); fade fastest on A1 (0.60)"

RUBRIC SELF-AUDIT

Escalation: all load-bearing indicators PASS → auto-ship.