Lesson 04 — A3 Elicitation: Asking Claude What It Would Cut

Learning objective (Bloom's: Apply): Apply the elicitation move — asking Claude for its own read, proposal, or uncertainty before you finalize anything — to a real research-synthesis and drafting decision in "The Last Mill."

Standard: AIHC.1.A3 (Elicitation), Band 1, verbatim: "Ask the AI coworker for its read, its proposal, or its uncertainty before finalizing an approach, and treat 'what would you do?' as a working question rather than a courtesy." [P2 · CI1 · X10]. Touches X10 (bidirectional knowledge flow — Claude's own diagnosis is first-class input, not a nicety).

Incoming mastery (from Lesson 01): AIHC.1.A3 = 0.20 — your lowest starting practice; today does the most scaffolding of the set.

Structure: PBL. Band 1; timing-tolerance honored: 3 cycles, each a full pause point.

Referent (flavor only, not load-bearing): a true-crime podcast host doesn't just read the verdict at the end of an episode — they ask the detective "what's the one lead you're still not sure about?" before closing it out. Elicitation is asking Claude that same question about its own draft.

Trust boundary named up front (R13): the elicitation question in Cycle 3 runs on a Research output. Research is paid-plan-only and requires web search on first [S11]; its citations are "easy to check back against sources" [S11] — meaning elicitation here is a lens on Research's synthesis, not a replacement for actually opening the cited sources (that's Lesson 05).

Cycle 1 — What elicitation actually sounds like

• SAY: Not "does this look okay?" — that's a check-in, not elicitation. A real elicitation question names a target: confidence ("how sure are you"), proposal ("what would you do"), or omission ("what would you cut"). AIHC.1.A3 treats the answer as working input, not politeness.
• SEE: A described contrast: a non-elicitation check-in next to three elicitation phrasings, each labeled by which target it hits.
• DO: On your actual current draft section of "The Last Mill," ask Claude directly: "What's the weakest sentence in this section, and why?" Log the answer. (Evaluation: checks the question named a specific target — omission — not a vague "thoughts?")

⏸ Pause point. Banked: one real elicitation exchange, logged.

Cycle 2 — Before finalizing, not after

• SAY: The standard specifically says "before finalizing an approach" — asking after you've already sent the pitch is a postmortem, not a working input. Timing is the whole difference between elicitation and a nice-to-have.
• SEE: A described timeline: the same elicitation question moved from after-ship (nothing changes, it's too late) to before-ship (it can still change the outcome), on one of her real workflow moments.
• DO: Before you finalize your next real pitch-tracker follow-up message, ask Claude: "What would you do here, and what's your least confident part of this recommendation?" Do this before sending. Log the answer, and note whether it changed your final move. (Evaluation: checks the elicitation happened before the send, not retroactively.)

⏸ Pause point. Banked: one real before-finalizing elicitation, with an outcome note.

Cycle 3 — Eliciting on a Research synthesis

• SAY: Because Research pulls together multiple sources [S11], asking "which source here are you least sure about?" turns a black-box synthesis into a checkable one — and sets up next lesson's verification work directly.
• SEE: A described Research transcript on "The Last Mill" background, with an added elicitation turn asking about source confidence, one source highlighted as weak.
• DO: On one real Research run about your piece's background, add the elicitation question and log which source got flagged. (Evaluation: checks a specific source was named, not a general "it seems fine.")

⏸ Pause point. Banked: one real Research run with an elicitation pass. Three cycles complete.

Independent at-bat

Take your next real drafting decision — a whole section, not one sentence. Elicit twice: once for confidence, once for an alternative. Decide the exact wording of both questions yourself; no template given this time.

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

1. (Remember) In your own words, what does AIHC.1.A3 ask you to treat "what would you do?" as?
2. (Understand) Why does elicitation have to happen before finalizing to count as working input rather than a courtesy? One sentence.
3. (Apply) Here's a Research result with three sources. Name the one elicitation question you'd ask before trusting any of it.
4. (Apply — objective level) On your real next section of "The Last Mill," ask Claude both a confidence question and a "what would you cut" question before you finalize it. Report what changed, if anything.

Ledger write

ledger_append:
  learner_id: L3-CONS-PRIYA
  lesson_id: L3-cons-priya-04-A3-elicitation
  standards:
    - {standard: AIHC.1.A3, mastery_before: 0.20, mastery_after_simulated: 0.50}
  bloom_reached: apply
  auto_score: ""
  self_score: ""
  structure_used: PBL
  referents_used: ["media — true-crime podcasts"]
  journal_prompt: >
    What did asking "what would you cut" catch that you would have missed by just reading the
    section again yourself?
  next_lesson_seed: "elicitation habit now present; next (B1) can lean on it directly — contrast re-asking Claude vs. checking a primary source"

RUBRIC SELF-AUDIT

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