Expedition 2 — Compass Calibration

After Session 2 | ~40-50 minutes

---

You just experienced RCCE and meta-prompting in Session 2 — Role, Context, Constraints, Examples. You saw what happens when you rebuild a prompt with all four elements. Now you practice both techniques independently, on your own work, at your own pace.

This expedition is the most technique-heavy homework in the program. Six steps, three required artifacts. Work through them in order. Time estimates are guides — if you're running short, prioritize Steps 1-4 and do Step 5 quickly. Never skip Step 6.

---

What You'll Produce

Artifact	Steps	Feeds Into
Before/After Prompt Pairs (x2)	1, 2	Session 3 Growth Reveal
Evaluation Log (3 outputs assessed)	4	Trail partner comparison at S3 opening
Cross-Domain Transfer reflection	5	Session 3 Transfer Challenge

Write everything in this conversation so you can scroll back to find it.

---

Step 1 — RCCE Practice, Rep 1 (10 min)

Paste or describe a real prompt you've used recently — something from actual work, not a hypothetical. Don't polish it. Show exactly what you'd normally type.

1. Run your original prompt (or confirm you already saw the output).
2. Rate the output 1-5: how close was it to what you actually needed?
3. Walk through RCCE. For each element, ask yourself:
   • Role — Did I tell AI who I am and who it should be?
   • Context — Does AI know my audience, purpose, and situation?
   • Constraints — Did I specify format, length, tone, scope, and what NOT to do?
   • Examples — Did I show AI what good looks like?
4. Rebuild the prompt yourself, adding the missing elements.
5. Run the rebuilt version. Compare outputs.

Record Before/After Pair #1: The original prompt, the RCCE-rebuilt version, and one sentence on what changed most.

---

Step 2 — RCCE Practice, Rep 2 (8 min)

Same process, different task. This one goes faster because you've done the moves once.

Pick a different type of work — if Rep 1 was writing, try planning or analysis. Push yourself toward a task where you're less certain AI can help.

Walk through RCCE again, but this time try to identify what's missing before checking each element. Intervene on yourself. After running the rebuilt prompt, notice: which RCCE element do you still skip by default?

Record Before/After Pair #2: Original prompt, rebuilt version, and one sentence on what you applied differently from Rep 1.

---

Step 3 — Meta-Prompting (7 min)

Different technique. Instead of rebuilding a prompt after the fact, you ask AI what it needs before you write it.

1. Pick a new task — not one from Steps 1 or 2.

2. Send this to Claude:

   "I need to accomplish [brief description of your task]. Before I give you the full prompt, what questions do you need answered to do this exceptionally well?"

3. Read the questions Claude asks. Answer them.

4. Write your full prompt incorporating those answers. Run it.

5. Compare how this felt versus Steps 1 and 2. What was different about having the questions first?

Meta-prompting is planning before acting — you're letting AI help you figure out what RCCE elements you need before you write anything.

---

Step 4 — Evaluation Practice (10 min)

Your expertise is your most powerful AI tool. This step teaches you to use it.

Evaluate three AI outputs using the quality checklist:

1. Your best output from Steps 1-3. Read it critically as the domain expert you are. What did AI get right? What's missing, oversimplified, or subtly wrong?
2. Ask Claude to generate a substantive piece in your area of expertise — an explanation, analysis, or recommendation (~150-200 words). Evaluate it.
3. Ask Claude for a second piece — different task type, same domain. Evaluate it.

Quality Checklist (apply to each output):

• Accurate? Is every claim correct and verifiable?
• Complete? Are important considerations missing?
• Actionable? Could someone use this to make a decision or take a step?
• Appropriately scoped? Does it fit the actual request — not too broad, not too narrow?

Record your Evaluation Log: For each of the three outputs, write one sentence on the most significant quality issue you found and what fix you would request.

---

Step 5 — Cross-Domain Transfer (10 min)

You've built two RCCE prompts. Now find out what's transferable.

1. Pick your stronger RCCE prompt from Steps 1 or 2.
2. Identify a completely different work area — different function, different type of task, genuinely different terrain. (If your prompt was for drafting communications, try data analysis, project planning, or research synthesis.)
3. Adapt the RCCE prompt for the new domain. You can't just swap words — rethink what Role, Context, Constraints, and Examples mean in this new context.
4. Run the adapted prompt.
5. Write a brief reflection: What transferred directly? What had to change? What broke or produced a worse result than expected?

This is the hardest step. That's intentional. The transfer is where you discover which parts of RCCE are universal principles versus domain-specific techniques.

---

Step 6 — Self-Grade (5 min)

Honest self-assessment. Re-read your best output from this session — the one you're most proud of.

Grade yourself on Execution Fidelity:

• Level 1: I accepted AI output without evaluating it
• Level 2: I noticed some issues but accepted a "good enough" result
• Level 3: I applied specific quality criteria, caught issues, and iterated with clear goals
• Level 4: My final output is polished, verified, and ready to use — not a draft

Name your level. Give one sentence of evidence from this conversation. Then answer: What would you need to do in the next conversation to earn one level higher?

---

Trail Partner Sharing

Before Session 3, share with your trail partner:

• Your best Before/After Pair — the one you're most proud of
• Your most interesting finding from the Evaluation Log

You don't need to share everything. The Session 3 conversation will be richer if you both come with something to show.

---

Feed-Forward

These artifacts feed Session 3's Growth Reveal. Your Before/After Pairs become evidence of prompting growth. Your Evaluation Log becomes a comparison point with your trail partner. Your Cross-Domain Transfer reflection seeds the Transfer Challenge. Come with all three.

---

RANGE Dimensions in This Expedition

• Execution Fidelity (primary) — Steps 1-2 and 4 build quality-focused output behavior
• Navigation (primary) — Steps 3 and 5 develop planning-before-acting and strategic cross-domain thinking
• Generalization (emerging) — Step 5 is the first explicit cross-domain transfer in the program
• Autonomy (developing) — Step 6 builds independent quality standards through self-assessment