Lesson 02 — A1 Specification: Four-Part Briefs On Notes And Pitches

Learning objective (Bloom's: Apply): Apply a complete four-part specification — audience, job, what's fixed, what's open — to a real organizing request on your interview notes and pitch tracker, so Claude's first answer needs no rescue rewrite.

Standard: AIHC.1.A1 (Specification), Band 1. Touches X9 (the ask/act boundary): naming what's fixed vs. open up front is exactly how you avoid Claude silently guessing at requirements only you hold.

Incoming mastery (from Lesson 01's diagnostic): AIHC.1.A1 = 0.60 — "good asker, unstructured" — you already ask reasonable things; today makes the four parts explicit and repeatable.

Structure: PBL — the messy asks are the ones already sitting in your real workflow this week. Band 1; timing-tolerance honored: four short cycles, each a full pause point.

Referent (flavor only, not load-bearing): in the IPL, a captain sets the field before the ball is bowled — naming exactly the shot he expects, not describing it after the fact. A spec'd ask sets the field before Claude "bowls" its answer.

Trust boundary named up front (R13): this lesson has you write standing project instructions. The doc-set states these apply "across all chats in that project" [S04] and that Claude "does not have access to" a project's description field, only its instructions and uploaded knowledge [S04, S03] — so what you write there is the only thing that persists; nothing about Anthropic's training-data policy on your notes is stated in this doc-set, and this lesson does not claim one.

Cycle 1 — The four parts, named

• SAY: Every spec'd ask names: audience (who this is for), job (what the text/output must do), fixed (non-negotiable), open (Claude's call). The doc-set's own four rules for a non-developer map directly onto this: "Explain your ask simply and clearly" (Clarity), "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 answer as a cue to add context rather than abandon the ask (Iteration) [S17].
• SEE: A described before/after: a representative (labeled fictional) messy ask — "help me figure out which of these pitches to follow up on" — with none of the four parts named, next to it rewritten with all four labeled.
• DO: Take one real messy ask you're about to make about your pitch tracker this week (which of your open pitches needs a follow-up, and to whom). Rewrite it into the four-part frame and run it for real. (Evaluation: checks all four labels are present and non-generic — "audience: me" doesn't count if the real audience is an editor.)

⏸ Pause point. Banked: one real four-part ask, run once.

Cycle 2 — Naming fixed vs. open on your interview notes

• SAY: "Fixed" and "open" is the part people skip. On a research file, some content is non-negotiable (a direct quote must stay verbatim) and some is genuinely Claude's call (how to group themes). Naming which is which before you ask protects the non-negotiable part without you having to re-check it after.
• SEE: A described tagging of her interview-notes file's content types: direct quotes → fixed; named-source attributions → fixed; thematic grouping → open; which anecdote is most vivid → open (her call to review, not Claude's to decide silently).
• DO: On your real interview-notes file, tag three content types as fixed or open, then write the spec for "organize this by theme" using those tags. (Evaluation: fixed items must include at least one "verbatim quote" tag — the one that protects you as a byline journalist.)

⏸ Pause point. Banked: fixed/open tags on one real file, one spec written.

Cycle 3 — Making the spec standing, not repeated

• SAY: You shouldn't have to restate your fixed rules every single message. The doc-set's project instructions are exactly this: click "Set project instructions," add guidelines, save — they "apply across all chats in that project" [S04]. A standing project instruction turns "always keep quotes verbatim" from a thing you remember to say into a thing that's simply true from now on.
• SEE: A described view of the "Set project instructions" panel [S04] populated with one real line: "Direct quotes from interview notes are fixed — never paraphrase them. Flag any sentence that's inference, not fact."
• DO: Write that one standing instruction — in your own words, naming your own real fixed rules — and save it as your actual project instruction for "The Last Mill." (Evaluation: the instruction must be testable — checkable in Claude's next reply, not a vague aspiration.)

⏸ Pause point. Banked: one standing project instruction, saved for real.

Cycle 4 — Testing whether the spec survives contact

• SAY: The real test of a spec isn't whether it feels complete — it's rewrite-distance: does Claude's first answer need a full rescue rewrite, a light edit, or nothing at all? That's your self-check going forward, and it's specific to you, not a universal score.
• SEE: A described three-point scale (none / light-edit / rescue-rewrite) applied to two of her own past real outputs, one vague-ask-origin and one spec'd-ask-origin, side by side.
• DO: Run Cycle 1's four-part ask again if you haven't already, and self-rate the answer's rewrite-distance. (Evaluation: compares her stated rewrite-distance against how much text she actually changes — an honest self-check, not a guess.)

⏸ Pause point. Banked: one rated real output. Four cycles complete.

Independent at-bat

Take your next real pitch-tracker decision — no template shown this time. Write its four-part spec cold, run it, and self-rate the rewrite-distance before you decide whether to send anything based on the answer.

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

1. (Remember) Name the four parts of a spec'd ask.
2. (Understand) Why does naming "what's fixed" specifically protect a byline journalist's verbatim quotes, in a way that a generally "clear" ask doesn't guarantee? One sentence.
3. (Apply) Here's a vague ask: "clean this up for me" about your pitch tracker. Rewrite it as a four-part spec.
4. (Apply — objective level) Take your real next interview-notes organizing task. Write its four-part spec, predict the rewrite-distance before you run it, then run it and compare.

Ledger write

ledger_append:
  learner_id: L3-CONS-PRIYA
  lesson_id: L3-cons-priya-02-A1-specification
  standards:
    - {standard: AIHC.1.A1, mastery_before: 0.60, mastery_after_simulated: 0.78}
  bloom_reached: apply
  auto_score: ""
  self_score: ""
  structure_used: PBL
  referents_used: ["sports — cricket (IPL)"]
  journal_prompt: >
    What's one ask on your pitch tracker or notes that you know you'll never send vague again —
    and what's the fixed element you almost forgot to name before today?
  next_lesson_seed: "A1 scaffolding can fade fastest next (already her strongest anchor); shift heavy scaffolding to A2 (0.35)"

RUBRIC SELF-AUDIT

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