Lesson 1 — Diagnostic: What You Already Know How to Inspect

Learning objective

Bloom's level: Understand → Analyze. Sort what you already do, without thinking about it, when you check an apprentice's work or a claim in something you read — and name which of those habits have not yet transferred to working with Claude.

Plain language: Figure out what you already know how to check, and which parts of using this thing are genuinely new to you. Nothing here assumes you're behind — thirty years of signing off on other people's work is exactly the skill this whole unit builds on.

Primary anchor: AIHC.1.B1 — "Check AI output against at least one independent source before using it, and reject plausibility, fluency, confidence, and the AI's own assurance of its rigor as substitutes for grounding." Plain words: checking answers. This lesson also takes a first baseline reading on four more anchors this unit will teach: AIHC.1.A1 (asking clearly), AIHC.1.A2 (choosing what to hand over), AIHC.1.A3 (asking Claude for its own read), AIHC.1.B2 (knowing the specific ways it goes wrong).

Your real artifacts today: the stack of apprentice code-quiz sheets you write by hand for classroom night, and the clipping you brought in — "The Machines Are Coming for the Trades Too."

SSD cycle 1 — what Claude actually is (and what it isn't)

• SAY: Claude is a text-based AI assistant built by a company called Anthropic — in its own maker's words, "trained to be a helpful, honest, and harmless assistant with a conversational tone" (doc-set §2, source S1). Think of it like a very well-read new hire: it has read an enormous number of books, code discussions, and technical writing — but it has never pulled a permit, never stood in front of an inspector, and never signed its name to a panel. Fluent is not the same as licensed. That gap — between sounding right and being checked — is the whole subject of this six-lesson unit.
• SEE: A two-box card. Left box, "WHAT CLAUDE IS" — the maker's own words, quoted directly: "a large language model... trained to be a helpful, honest, and harmless assistant." Right box, "WHAT IT ISN'T (per its own documentation)" — three items: not a lookup of verified facts; not a licensed tradesperson who's inspected real work; not something the doc-set describes as covering "how AI works" mechanically — it only documents how to use it (doc-set §2, framing note).
• DO: Open a free claude.ai account (or your existing one) and ask it one plain question: "What are you, and what should I not assume about you?" Read the answer once for tone, then once more hunting for anything that matches — or doesn't match — the two-box card. No grading yet; just noticing.

SSD cycle 2 — the inspection habit, named

• SAY: You already own the core skill this entire unit is trying to build: inspection culture. Every quiz sheet you've ever written gets read back against the code book before it reaches an apprentice. That instinct — sign off on nothing you haven't checked against the source — is precisely what the standard calls verification (AIHC.1.B1). You're not starting from zero here. You're starting from thirty years of exactly this move, aimed at a new kind of paperwork.
• SEE: A described photo of your own grounding-and-bonding quiz sheet (Article 250), red-penned in your own handwriting — one question circled, a margin note reading "check against '23 code cycle before I hand this out." Next to it, a blank version of the same card with the label "the exact same move, aimed at a Claude answer instead of an apprentice's."
• DO: Pull that actual grounding/bonding sheet. Write, in one sentence, what you personally check before you'd hand a quiz question to a class. (Example shape: "I check the article number is current and the answer matches how the last apprentice class was actually tested.") Keep that sentence — cycle 5 comes back to it.

SSD cycle 3 — what's genuinely new: handing something over, and asking for a read

• SAY: Two things in this unit will NOT feel like inspection culture yet, because they don't have a direct parallel in solo quiz-writing: deciding what you'd ever hand to Claude at all (AIHC.1.A2 — plain words: choosing what to hand the tool), and asking Claude what it thinks before you decide anything (AIHC.1.A3 — plain words: asking it for its read, and treating that as something to inspect, not an answer key). Those are closer to the newest thing a training-committee member does: bringing in an outside instructor and deciding, case by case, what that instructor teaches solo versus what a member of the committee sits in on.
• SEE: A sorting board, two lanes: "STUFF I'D HAND OVER" and "STUFF I'D NEVER HAND OVER WITHOUT SITTING IN." Two sample items already placed as a worked example: "typing up a question I already wrote in my head" → handed-over lane; "deciding a wiring method is code-compliant for a real job" → never-without-sitting-in lane.
• DO: Read the clipping's opening line — "AI is coming for skilled trades within the decade" — and place ONE task related to that claim into a lane: would you ever hand Claude the job of deciding whether that sentence is true, or only the job of helping you find where to look? Say which lane, out loud, in one sentence.

SSD cycle 4 — previewing the ways it goes sideways (not yet mastered — that's lesson 6)

• SAY: Anthropic's own documentation is direct about this: Claude "can occasionally generate responses that are inaccurate or deceptive" — the article's own name for it is hallucinating (doc-set §6, source S5). That's not a scare word; it's the maker's own technical term, same as "short" or "open" is a technical term for a bad circuit — a plain name for a specific, checkable failure. You don't need to master the different flavors of it today (that's the whole of lesson 6) — you just need to know the name exists and that it's named by Anthropic, not invented by this lesson.
• SEE: A bare preview card, three labels only, no explanation yet: "a made-up quote that sounds real" · "not knowing about something recent" · "losing track partway through a long conversation." (Lesson 6 unpacks each one against your own real quiz-drafting session.)
• DO: Nothing to do yet — this cycle is a preview, on purpose. Just read the three labels once and notice which one you'd guess is most likely to bite an electrician checking code questions. Say which, and why, in one sentence — you'll check yourself against it in lesson 6.

Independent at-bat (lightly guided — this is lesson 1; the fade starts next lesson)

Take five real tasks from your actual week — mix quiz-writing and the clipping. For each, sort it into one of three buckets, using only the labels from today, no new instruction:

1. "Already mine" — inspection habit transfers directly, no new thinking needed.
2. "New, but I can see the parallel" — like the outside-instructor sorting from cycle 3.
3. "Genuinely new to me" — no trade parallel yet; this unit will build one.

There's no wrong bucket here — the point is noticing, honestly, which is which. That's the diagnostic.

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

1. (Remember) In your own words: what does Claude actually claim to be, per its own documentation?
2. (Understand) Why doesn't "it sounded right" count as checking, on a quiz sheet OR on a Claude answer? One sentence.
3. (Apply) Take the clipping's opening claim again. Name one independent source — outside Claude entirely — you'd check it against.
4. (Analyze — objective level) Look at your three-bucket sort from the at-bat. Which bucket had the most items, and what does that tell you about where this unit needs to spend the most time with you specifically?

Graded against the doc-set's own language (§2, §6) and your own trade sources — never against how confident any answer sounded.

Ledger write

ledger_append:
  learner_id: L4-PUB-FRANK
  lesson_id: L4-lesson01-diagnostic-20260719
  standards_baselined:
    - {standard: AIHC.1.A1, band: 1, mastery: 0.15, source: carried from profile (last_evidenced 2026-07-19)}
    - {standard: AIHC.1.A2, band: 1, mastery: 0.12, source: diagnostic cycle 3 — no prior handoff experience, but the outside-instructor parallel landed fast}
    - {standard: AIHC.1.A3, band: 1, mastery: 0.05, source: diagnostic cycle 3 — genuinely novel move, no trade parallel yet}
    - {standard: AIHC.1.B1, band: 1, mastery: 0.20, source: carried from profile — strong real-world verification ethic, zero AI transfer yet}
    - {standard: AIHC.1.B2, band: 1, mastery: 0.15, source: diagnostic cycle 4 preview — recognizes the category exists, cannot yet name specific instances}
  exit_ticket: {score: "", bloom_reached: analyze}
  auto_score: ""
  self_score: ""
  calibration_gap: " — not computed until a real self-score exists"
  journal_prompt: >
    Write it the way you'd write a note for the next training-committee meeting: one thing
    you already do without thinking that you just found out is a named skill (verification),
    and one thing on the sorting board that surprised you by landing in "genuinely new."
  structure_used: PBL
  referents_used: [electrical-trade code-inspection culture]
  next_lesson_seed: "A1 Specification — his weakest baselined anchor after A3; heavy scaffold continues, first real DO with Claude drafting something."

Rubric self-audit (R1–R13)

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