The Ascent / Learn

Level 1 · Ask — the lesson

Ask (Chat)

One expert who's read almost everything. It reasons, you verify, you act.

The line Chat drafts and advises — it can't touch your files, inbox or calendar. That's Level 2.

Read the full referencethe deep, complete version of this level

Play the models below — each one you try climbs a floor of your tower.

Tap a card — watch a lazy prompt become a sharp one

WritingVague
Write something about our project update.
Write a 3-sentence update for our client (non-technical): civil works 2 weeks behind on permits, MEP on track. Keep it reassuring.

✗ No reader, tone, length or facts.

✓ Audience, length, facts and tone — all set.

AnalysisVague
What do you think about these numbers?
Here are Q1 power costs vs budget [table]. Name the top 2 variances, a likely cause and one action each. Under 150 words, for a board pack.

✗ Which numbers? What decision?

✓ Data, task, format, limit, audience — locked in.

SummaryVague
Summarise this.
Summarise this transcript in 5 bullets — decisions made and who owns each action. Ignore the chit-chat.

✗ How long? What angle? For whom?

✓ Format, length, focus and what to drop.

Build a prompt — every field rebuilds it live

Load a preset
R
Role
O
Objective
C
Context
C
Constraints
O
Output
Your prompt, liveOpen Claude ↗

      

Claude answered confidently. Sort each claim by how you'd check it

Claude

You should reforecast every 2 weeks. SaaS firms reforecast every 14 days on average, and a rolling forecast beats an annual budget when revenue is volatile.

Each claim below — how would you check it?

SaaS firms reforecast every 14 days on average⚠ hallucination

A precise, sourceless number — the classic tell. Search it and you'll find nothing. Check facts outside the chat; never let the model cite itself.

A rolling forecast beats an annual budget when revenue is volatile

A recommendation, not a lookup. Stress-test it: what has to be true for this to hold — and when would it not?

You should reforecast every 2 weeks

Advice tuned to nothing about you. Pressure-test it against your actual situation before you act.

Do it now — 4 small steps

One step = one paste into Claude (the mobile app is perfect). Your progress saves on this device.

  1. Step 1 of 4 · Fix a vague prompt

    Half of 'bad answers' are really a bad ask.

    Tap the cards above — feel a lazy prompt become a sharp one.

    You should see: The same request, reshaped so the answer can't miss.

  2. Step 2 of 4 · Build with ROCCO

    Role · Objective · Context · Constraints · Output.

    Use the builder above: load a preset, edit a field, copy the prompt, run it in Claude.

    You should see: A fully-dressed prompt — and a much sharper reply.

  3. Step 3 of 4 · Make it show its work

    Confident isn't correct.

    Answer, then name the one assumption that, if wrong, most changes your answer — and how I'd check it.
    Open Claude ↗

    You should see: A claim you can test, not just trust.

  4. Step 4 of 4 · Ship an artifact

    If nothing left the chat, you only talked.

    Turn this into a one-page brief: title, 3-bullet summary, one next step.
    Open Claude ↗

    You should see: Something you can paste anywhere.

The ideas behind the steps — tap a takeaway to unpack it.

Your three hats

Takeaway Frame the intent, test the reply, steer the thread.

Navigator — you own the goal. Verifier — you demand evidence. Conductor — you steer across turns. At L1 the Verifier leads.

The wrong-answer cost test

Takeaway Match your caution to what a wrong answer would cost.

Ask: if this is confidently wrong and I act on it, what does it cost? Low → move fast. High → demand sources, verify everything.

Verify by type

Takeaway Check facts against the world; pressure-test opinions.

Facts (numbers, dates, cites) → check outside the chat. Judgments → stress-test: what has to be true? Never let it cite itself.

Is this the right level for your task?

You want Claude's help right now. Which fits?

Level 1, right here. Run it through ROCCO.
That's Level 2 — connect a tool. Open the Connect lesson →
That's Level 3 — delegate the job. Open the Delegate lesson →

Pick the closest match — you'll get a verdict.

The capstone — the whole level in one prompt

Your capstone — the whole level in one prompt. Fill the brackets (they're the five ROCCO fields) and run it.

Role: [who Claude should be]. Objective: [the outcome, one line]. Context: [2–3 facts only you know]. Constraints: [limits — length, tone, what to avoid]. Output: [the shape you need].

Then: show your reasoning, flag general advice vs specific to me, and name the assumption most likely to be wrong. I'll push back on your weakest point.
Open Claude ↗

Self-check

0 / 4

You've leveled up when…

You're ready when you frame before you fire, verify by type, and keep wishing Claude could just *see and act on your real stuff*.

Next: Connect (L2) — grant Claude one real tool, and own its permissions.

Next · Level 2: Connect →