The evidence ladder grades what you know about a belief by what people did, not by what they said. Six rungs, from praise to pays. A belief climbs only when someone does something that costs them: time, a name, money. Praise never moves it. Your own words never move it. This guide explains each rung, why the rules are strict, and how to run a test that can actually fail.
The six rungs
Start with a belief: something that must be true for your product to work. "Solo designers chase late invoices every month." Then ask what you actually know about it. The ladder is how you answer.
- PraiseNever counts as proof
- Past painThey did something about it
- Asks for accessWants it before it exists
- Gives timeA call, a test, a trial
- Makes an introPuts their name on it
- PaysMoney, before or after
Praise. They said something nice. "That's a great idea." "I'd totally use that." "I'd pay for that." It costs nothing and predicts nothing. It stays in the first column.
Past pain. They described a specific time the problem hurt, without being led to it. "Last month I spent a Sunday chasing three invoices." They did something about it, even if the something was suffering. The key word is unprompted.
Asks for access. They asked to get in before it exists. "Can I try it when it's ready?" "Put me on the list." They want something they cannot have yet. That is a small cost: their attention and their inbox.
Gives time. They showed up. A follow-up call. A trial they actually used. A test they took part in. Time is the first thing people guard, so time given is real evidence.
Makes an intro. They put their name on it. They introduced you to a colleague, a client, a community. Reputation is at stake now. People do not do this for things they only politely like.
Pays. Money changed hands, before or after the product exists. A pre-payment, a deposit, a subscription. The only rung that cannot be faked by goodwill.
What moves a belief
One rule: a belief moves right only when someone did something. Words move nothing. Actions move it as far as the action goes.
- A conversation can carry a belief as far as Asks for access. Talk ends there.
- Gives time and Makes an intro need a record you keep: who, what they did, when.
- Pays is reached only through a test whose pass line was set before it started.
- Contradicting evidence is shown beside the belief but never lowers it on its own. Calling a belief refuted is your decision, written down.
- A compliment in answer to your leading question is praise, even if it sounds like pain. "Isn't it frustrating to chase invoices?" "Yes, so frustrating." That is you talking.
The rules are strict on purpose. Loose rules let the belief you most want to be true climb fastest, because that is the one you ask about most and hear agreement on most.
Why praise never counts
Praise is the most common thing a founder hears and the least informative. Three reasons.
It is free. Saying "I'd use that" costs the speaker nothing. Nothing that is free tells you about willingness to pay, to switch or to change habits.
It is social. People are kind, especially to someone excited about an idea. A compliment ends an awkward moment. It is also the fastest way to close a conversation politely.
It is often your fault. Most praise follows a leading question. Name your product, describe the feature, then ask "would you use it?" and you will hear yes. The yes is a reflection of your question. The customer interview questions guide covers the six question shapes that produce praise and how to rewrite them.
"I'd pay for that" deserves its own line. It sounds like Pays. It is Praise. Nobody has paid. Until money or a locked test moves it, it sits in the first column with the rest.
Why your own words never count
You are the least reliable witness to your own belief. You wrote it. You want it to be true. Every pitch you give, every description you offer in a call, every summary you type into a notes app is your voice, not the customer's.
So the rule is absolute: your lines, and your teammates' lines, never count as evidence. In a transcript, only the other person's turns count. In your notes, only what they said. Your summary of what they meant is a draft of a belief, which goes back on the left of the ladder to be tested.
This feels harsh the first time. It is also the only thing that stops a founder from validating an idea by talking about it.
Why your tools agree with you
Founders now check their beliefs with AI tools, and those tools agree. Cheng, Jurafsky and colleagues, writing in Science in March 2026, tested eleven models and found that AI affirmed users' actions 49% more often than people did (science.org, Mar 2026). Ask a general model whether your idea is good and it will tend to say yes.
This matters for the ladder because a model's agreement is praise with extra steps. It is not a person, it did nothing, and it costs nothing. A tool that reads your beliefs should show you what argues against them before it agrees, and should say "nothing in your data on this" when that is true. If it cannot, treat its agreement as you would treat a friend's compliment.
How to run a pre-sale test
Pays is the top rung, and a pre-sale is the cleanest way to reach it before the product exists. The test is simple. The discipline is in the order.
Write the belief and the metric
"Solo designers will pay for a pay page." Metric: pre-payments, counted from a payment link or a payments export.
Set the pass line, with a date
"Ten pre-payments by 15 October." Write what a pass moves the belief to (Pays) and what you will do if it misses. Decide this before anyone sees the offer.
Lock it
From now on the line, the metric and your call cannot change. Only the deadline can move, and the move is logged with its reason. A line you can edit afterwards is not a line.
Make the offer
A page, an email, a message to the people who climbed to Asks for access. State the price and what they get. Collect payment, not promises.
Record results as they come
Each pre-payment with its date and amount. A result can be voided with a reason, never deleted.
Read the verdict on the deadline
Passed or missed against the line you set. A pass moves the belief to Pays. A miss does not, whatever else happened. Then log the decision you said you would make.
Two traps. Counting sign-ups as payments: a sign-up is Asks for access. Moving the line after the fact: "eight is basically ten". Eight is a miss. Learn from it and run the next test.
Examples of each rung
Generic examples, none from any customer. The belief in each case is the same: "small agencies will pay to automate client invoicing".
| Rung | What happened | Why it lands here |
|---|---|---|
| Praise | "Oh that's smart, agencies would love that." | Nice words, no action, no cost |
| Past pain | "We lost a client in March over a late invoice nobody noticed." Said before you described the product. | A specific past event, unprompted |
| Asks for access | "Send me the link when it's live. Actually, can I see it now?" | Wants something that does not exist yet |
| Gives time | Took a second 45-minute call and walked you through their invoicing spreadsheet. | Gave the thing people guard most |
| Makes an intro | Introduced you to two other agency owners by email, with a line about why. | Put their reputation beside yours |
| Pays | Pre-paid for three months before launch, through a test with its line set first. | Money, through a locked test |
Notice what is missing from the right-hand column: your opinion. Every row is something a person did, with a date you could write down.
Related guides: Customer interview questions · The product-market fit survey · Spec-driven development for founders
Do this in MindBacklog
Beliefs on the ladder, with the rules enforced. Track what you believe and how strong the evidence is: praise never counts as proof. Your own words never count as evidence. Drop a transcript and record consent. Mind pulls their exact words, leaves out yours, and marks what you prompted.
Set the pass line before the test starts; results can't be edited afterwards. Mind shows you what argues against you before it agrees.
Questions, conversations and tests · Problems ranked by your users
