Pick your year
Home by Dark
×

Prove it, don't believe it

⏱ 15 min4 steps

Claude is usually right, and it sounds every bit as sure when it's wrong. So the rule in my projects is: a thing isn't done until someone has seen it work. Not "it should work." Not "I made the change." Seen it.

I learned this from a check that told me everything was fine. My daily health check stayed green while the nightly job it was meant to watch had failed. Everything looked fine until I went and looked at the real result.

Still sounds too sure? The fix is under the step where you ask.

Before this, you'll want:

×

Steps

Step 1 of 40 done · about 15 min left

Steps for
  1. Step 1 of 4, not done yet
  2. Step 2 of 4, not done yet
  3. Step 3 of 4, not done yet
  4. Step 4 of 4, not done yet

  1. Pick up your last conversation in my-idea

    1. Open (Cmd+SpaceWindows key, type Terminal, press Enter).

    2. Go to your my-idea . Type this and press Enter:

      cd ~/Backyard/my-idea

      cd ~\Backyard\my-idea

    3. Check the says my-idea.

    4. Pick up where you left off. Type this and press Enter:

      claude -c
    Terminal An app on your Mac where you type instructions instead of clicking on things. Claude Code works inside it. All wordsfolder A place on your computer that holds files and other folders, like a drawer in a filing cabinet. All wordsprompt line The line in Terminal that ends with a % sign. The word right before the % is the folder you're in. All words

    You should see:

    opens, with your last conversation above the line where you type.

    Claude Code Claude, working inside Terminal. It can make files and run things on your Mac, with a safety checker watching what it does. All words
    That's not what I see
    It says "cd: no such file or directoryCannot find path … because it does not exist".
    The my-idea folder isn't there. It comes from Check the idea first and The three files, so do those first, then come back.
    It says "zsh: command not found: claudeThe term 'claude' is not recognized".
    This Terminal window doesn't know about Claude Code. Close the window with Cmd+WCtrl+Shift+W, open a new one and try again. If it still says that, go back to Claude Code and .install Put a new program on your computer so you can use it. All wordssign in Tell Claude Code which Claude account is yours, so it knows you've paid for it. All words

  2. Ask Claude to show you how it checked

    You've been training for this since fifth grade, when you got stuck with the group-project kid who said they did their part. You learned to say "show me." Same thing here.

    This asks Claude to show its for everything it told you is done, and to say plainly what it never checked.

    1. Copy the prompt below.
    For everything you've told me is done in this project, show me how you checked it: what you ran or opened, and what you saw. If you didn't check something, say so plainly. Then list what you have not checked at all.
    1. Paste it into Claude Code with Cmd+VCtrl+V and press Enter.
    2. Read the answer.

    Anything it didn't check is a guess, however confident it sounded.

    prompt What you type to Claude. A good one says what you want, how to work and what never to do. All wordsproof Something you or Claude actually saw, like the page loading or the line in the result. Not "it should work". All words

    You should see:

    Claude goes through what it said was done, with what it ran or opened and what it saw. Then it lists what it hasn't checked.

    That's not what I see
    Claude says it checked, but doesn't show what it saw.
    Ask: Show me the exact line, page or result. "It works" isn't proof. What you saw is.

  3. Look at the real thing yourself

    Whatever you built, look at it the way the people it's for would. Check the live result, not the report about it.

    1. Pick one thing Claude said is done.
    2. Open it the way they would: the page on your phone, the button clicked, the document read as a stranger.
    3. Compare what you see with what Claude said.
    That's not what I see
    There's nothing built to look at yet.
    Check the files instead. Open and and read them the way a stranger would.PRD.md One page on the problem, who it's for, and what "done" looks like. PRD stands for product requirements document. All wordsREADME.md The front door of a project. It says what the project is and how to run it, for you in six months and anyone you hand it to. All words

  4. Write down what's proven and what isn't

    This prompt has Claude write a : what's working, what's broken and what nobody has checked, as of today.

    1. Copy the prompt below.
    Create a file called status.md with four headings. As of: today's date, the day we checked, not the day you edit the file. What I checked and saw: only things one of us actually looked at today, and what we saw. Open problems: what's wrong, worst first. Not proven yet: what nobody has checked. This one is never empty. Fill it in from this conversation, and ask me about anything you'd be guessing.
    1. Paste it into Claude Code with Cmd+VCtrl+V and press Enter.
    2. Answer any question Claude asks about something it would be guessing.

    The "Not proven yet" list is never empty. There is always something nobody has checked. Writing it down is how you stop it surprising you later.

    status file One page that says what's working, what's broken and what nobody has checked, as of a date. All words

    You should see:

    Claude makes status.md with the four headings, from As of down to Not proven yet, and tells you what it put under each.

    That's not what I see
    Claude put things under "checked" that I never saw.
    Move them to Not proven yet. Tell Claude: Only put things there that we looked at today. A status file is only useful if you can trust every line.
×

Your challenge

For one week, every time Claude says something is done or fixed, ask "show me how you checked." Keep a tally of how many times the answer changed what you believed.

Last checked 2026-10-17

Start