Lesson 8 — C1 Provenance: Whose Stamp Is on This Work

Honesty note on this lesson's grounding: claude.ai has no built-in "label this as AI-drafted" feature — that gap is real and this lesson doesn't paper over it (doc-set has no source describing one). What the doc-set does give is Anthropic's own disclosure requirement for certain uses (S11, cycle 1) — this lesson teaches a personal labeling habit that satisfies that requirement, not a product feature.

Learning objective

Bloom's level: Apply. Label a real apprentice-training document so any reader can tell exactly which parts are your own words and judgment and which parts Claude drafted or suggested.

Plain language: Stamp the work the way you'd stamp a job — mark what's yours and what's Claude's, plainly, right on the page, before it ever reaches a training-committee member or an apprentice.

Standard: AIHC.1.C1 — "Label work they hand off or publish with what the AI did and what they did." Plain words: keeping straight who did what.

Your real artifact today: last week's finished grounding-and-bonding quiz sheet, and — for the independent at-bat — the toolbox-talk handout you're prepping for next week's safety meeting.

SSD cycle 1 — why the stamp matters more at the union hall than anywhere else

• SAY: Anthropic's own Usage Policy names legal, financial, and employment-related uses as requiring extra safeguards when the output is consumer-facing: human review, and disclosure that AI helped produce it (doc-set §9, source S11). Apprentice-training material sits squarely in that employment-adjacent lane — it feeds a real certification path. Real term: AI disclosure ↔ trade version: the job card that shows which stamp — journeyman's or apprentice's — is on which task, because when the inspector or the union asks, you need the record, not your memory.
• SEE: A described job-card layout next to a "content provenance card" mapped onto an apprentice worksheet: a header block with three categories — Frank-authored / Claude-drafted, Frank-checked / Claude-drafted, not yet checked.
• DO: Pull last week's finished quiz sheet. Go line by line and mark, retroactively, which lines were your own wording from the start and which (if any) started as a Claude draft you then edited.

SSD cycle 2 — three tags, not a guessing game

• SAY: A workable system, three tags: [MINE] · [CLAUDE-DRAFT, CHECKED] · [CLAUDE-DRAFT, NOT YET CHECKED]. That third tag is the one that matters most — it's the flag that says "this does not leave my desk labeled this way." It ties straight back to last lesson's rule sheet: anything still carrying that tag is, by definition, in your never-without-review lane (AIHC.1.B3).
• SEE: A sample paragraph with all three tags applied inline, and a callout arrow pointing at the "NOT YET CHECKED" tag: "this one gates release — nothing else does."
• DO: Draft one new quiz question live, with Claude's help, and apply all three tags to the resulting text as you go — in real time, not after the fact.

SSD cycle 3 — the stamp travels with the document

• SAY: A stamp that only lives in your head doesn't help the training committee. It has to be on the artifact — a footer line a committee member reads without having to ask you. Trade version: the permit posted at the job site, not the permit sitting in your truck.
• SEE: A finished worksheet with a footer line: "Prepared by Frank O.; Claude (claude.ai, free plan) assisted with question wording, reviewed and corrected by Frank before use."
• DO: Write that exact footer line for the quiz sheet you tagged in cycle 2, and attach it to the document.

Independent at-bat (unscaffolded — a second, different real document)

Take the toolbox-talk handout you're prepping for next week's safety meeting. Using only the three-tag system and the footer convention from memory — no cycle text to reference — label it and attach its footer line, start to finish, on your own.

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

1. (Remember) What are the three tags in your labeling system?
2. (Understand) Why does the "NOT YET CHECKED" tag matter more than the other two?
3. (Apply) Here's a paragraph Claude drafted that you haven't reviewed yet — tag it correctly.
4. (Analyze) Compare your cycle 2 quiz question and your at-bat toolbox talk — which one carried more Claude-drafted content, and does that change how carefully you'd stamp it?
5. (Apply — objective level) Write the exact footer line for your toolbox-talk handout.

Graded against your own tagged documents and the S11 disclosure language — never against whether the labeling merely looks thorough.

Ledger write

ledger_append:
  learner_id: L4-PUB-FRANK
  lesson_id: L4-lesson08-c1-provenance-20260719
  standard: AIHC.1.C1
  exit_ticket: {score: "", bloom_reached: apply}
  auto_mastery: "0.0 ->  (projected ~0.40 — first lesson on this anchor; a genuinely new habit, no direct trade parallel until the job-card framing landed in cycle 1)"
  self_score: ""
  calibration_gap: " — not computed until a real self-score exists"
  journal_prompt: >
    If a first-year apprentice asked why there's a stamp line on a quiz sheet now,
    what would you tell them in one sentence?
  structure_used: PBL
  referents_used: [electrical-trade job-card and permit-posting culture]
  next_lesson_seed: "C2 Governance — his own free-tier privacy settings plus a training-committee 'never paste this' list; builds directly on the never-without-review lane this lesson's tags surface."

Rubric self-audit (R1–R13)

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