Lesson 10 — C3 Measurement: Is It Actually Helping Your Apprentices

Learning objective

Bloom's level: Analyze. Compare an estimate of how Claude will help — time, and quality by your own definition — against what actually happened on a real quiz-writing task, in writing, and draw a conclusion.

Plain language: Same as bidding a job by the hour and then checking the actual hours against your bid: write down what you expected before you start, then check it after — not just "did that feel faster."

Standard: AIHC.1.C3 — "Estimate before delegating and compare the estimate with what actually happened, in writing." Plain words: checking your guess against what really happened.

Your real artifacts today: this week's real grounding-and-bonding quiz topic, and — for the independent at-bat — next week's overcurrent-protection quiz.

SSD cycle 1 — bid the job before you start

• SAY: Anthropic's own product overview names "learning" as one of Claude's stated capability areas, specifically calling out "exam and interview prep" (doc-set §2, source S9) — a doc-set-grounded reason to expect help on quiz-writing specifically, not an assumption this lesson invents. Bidding a job before you start it is already your trade — you're just aiming the same habit at Claude. Before touching the tool, write three things: how long a ten-question quiz takes you solo, how long you expect with Claude's help, and what "better" would actually look like if it helped (fewer typos, sharper wrong-answer choices, faster — your own definition, not a vague feeling).
• SEE: A described "bid sheet": Solo estimate | Assisted estimate | What "better" means — filled with one worked example from a different topic (bonding versus grounding, so his own stays fresh) so the shape is clear without giving away his own numbers.
• DO: Fill out your own bid sheet for this week's real grounding-and-bonding quiz, before opening Claude at all.

SSD cycle 2 — run the two jobs, keep the clock

• SAY: Anthropic's own best-practice guidance says to "start by planning your conversations," "be specific and concise," and for writing tasks specifically, "send the entire text to be edited in one message" (doc-set §5, source S3). These aren't etiquette — sloppy prompting inflates your actual time past the estimate the same way a badly-read print inflates a job's actual hours past the bid.
• SEE: A described stopwatch icon next to the bid sheet's "assisted estimate" column, with a blank "actual" column waiting to be filled for both the solo and assisted runs.
• DO: Write the quiz solo, timed. Then write an equivalent quiz with Claude's help, timed, following S3's practices — one clear, complete first message, not five trickled-in follow-ups. Fill in the "actual" column for both.

SSD cycle 3 — what the numbers actually say

• SAY: Compare, in writing: the time delta, and the quality delta against your own "what better means" definition from cycle 1 — not against a vague feeling. Same as comparing a studio cut to a bootleg: you don't grade it by which one feels more energetic, you check it against what you were actually going for. (Referent: classic rock — flavor for the comparison point only; the grounding stays your own bid sheet.)
• SEE: A described side-by-side of your two finished quizzes, solo and assisted, with your cycle-1 quality checklist ticked off against each.
• DO: Write one paragraph: did Claude help, by how much, and on which of your own criteria — citing your actual estimate-versus-actual numbers, not an impression.

Independent at-bat (unscaffolded — a full repeat, alone, on a different topic)

Next week's class needs an overcurrent-protection quiz. Run the entire loop yourself, with no cycle text to reference: bid sheet before starting, timed solo run, timed assisted run, comparison paragraph against your own criteria.

Exit ticket (Bloom's climb: Remember → Understand → Apply → Analyze → Analyze, objective level)

1. (Remember) What three things did you estimate before starting, back in cycle 1?
2. (Understand) Why compare your actual time against your own estimate, instead of just asking whether it felt fast?
3. (Apply) Your estimate said twenty minutes assisted; it took thirty-five. Name one S3 practice you might have skipped that could explain the gap.
4. (Analyze) Between your cycle 2 quiz and your at-bat quiz, which showed a bigger help-versus-solo gap, and on which of your own criteria?
5. (Analyze — objective level) Write your verdict: is Claude worth the time for this kind of task specifically — cite your own numbers.

Graded against your own bid sheets and actual outcomes — never against how the session merely felt while you were in it.

Ledger write

ledger_append:
  learner_id: L4-PUB-FRANK
  lesson_id: L4-lesson10-c3-measurement-20260719
  standard: AIHC.1.C3
  exit_ticket: {score: "", bloom_reached: analyze}
  auto_mastery: "0.0 ->  (projected ~0.50 — first lesson on this anchor; estimate-versus-actual is a direct transfer from his own trade practice of bidding jobs)"
  self_score: ""
  calibration_gap: " — not computed until a real self-score exists"
  journal_prompt: >
    Same as after a real job — did the numbers come in where you bid them, and if
    not, what's the one thing you'd change about the estimate next time?
  structure_used: PBL
  referents_used: [classic rock — studio cut vs. bootleg (comparison-discipline point only)]
  next_lesson_seed: "D1 Orchestration — running a separate thread for quiz-writing and a separate thread for news-claim checking in the same week; his bid-sheet habit carries over as one of the things worth timing across two threads."

Rubric self-audit (R1–R13)

Escalation: all load-bearing indicators pass → auto-ship (per playbook §5 policy).