1Fast overview
If you only have 30 seconds, read this.
The idea: Your brain doesn't react to "the facts." It reacts to a small slice of facts it happened to notice, filtered through your past experience, and turned into a story — and that story feels like the truth.
The Ladder of Inference is a simple map of that process: from raw data, up through selection, interpretation, and assumption, to conclusions, beliefs, and finally action. Each rung you climb feels automatic and obvious in the moment. It usually isn't.
The fix isn't to stop climbing — you can't think without inference. The fix is noticing when you've climbed, and being willing to climb back down and check your footing before you act, speak, or write the email you'll regret.
The seven rungs, bottom to top:
2Mental Model Compass
A quick gut-check before you reach for this tool.
Reach for this when…
- You're suddenly certain about someone's motive, competence, or intent
- A disagreement keeps escalating and no one can say exactly what the other person did wrong
- You're about to send an email, give feedback, or make a call while frustrated
- A team keeps arguing about conclusions without ever comparing the data each person started from
- You want to coach someone else's reasoning without telling them they're "wrong"
Skip it when…
- The situation needs a fast, low-stakes call and the cost of overthinking exceeds the cost of a wrong guess
- You already have complete, verified, shared data — the disagreement is about values or priorities, not facts
- Someone is in genuine danger and needs immediate action, not a reasoning audit
- You're using it to relitigate a decision that's already been made and acted on
3Climb the ladder
Click each rung, bottom to top, to see what's actually happening at that stage — and the question that catches a problem there.
Start at the bottom
Click rung 1 to begin climbing. Each step shows how reasonable people quietly turn the same raw reality into very different actions.
0 of 7 rungs explored
Quick check: where did this go wrong?
Priya emails her manager: "Per my last email, see below." Her manager thinks: "She's being passive-aggressive because I didn't reply fast enough," and replies curtly. Where did the manager's reasoning first go off track?
4How it shows up
The same climb, in very different rooms.
An engineer's camera is off in every standup. A manager selects that one data point, interprets it as disengagement, concludes the engineer has "quiet quit," and skips them for a stretch assignment — never learning their laptop camera broke two weeks ago.
A prospect doesn't reply to a follow-up for four days. A rep assumes the deal is dead and stops pursuing it, when the prospect was actually waiting on budget approval and still intends to buy.
A teenager is monosyllabic at dinner. A parent concludes "they're hiding something" and interrogates them, when the teen is simply exhausted from exams — and now also defensive.
A support agent reads a terse, all-caps ticket as "hostile customer" and responds formally and coldly, when the customer was typing from a phone with a broken shift key, mid-emergency at work.
An investor sees a single bad quarter, concludes the company's thesis is broken, and sells — skipping the step of checking whether the miss was a one-off supply issue the market had already priced in.
A user sees one wrong AI answer, concludes "this model can't be trusted for anything," and abandons a tool that would have worked fine with a clearer prompt and a chance to check its reasoning.
5Worked application
Walking one real situation all the way down the ladder and back up again, deliberately.
-
1
Name the action you're about to take. "I'm about to tell my co-founder I think they're not pulling their weight, in front of the rest of the team."
-
2
State the belief underneath it. "I believe they've checked out of this company."
-
3
State the conclusion that produced the belief. "I concluded they don't care anymore."
-
4
Surface the assumption. "I'm assuming that if someone cared, they'd be working the same hours and channels I do."
-
5
Trace the meaning you added. "I decided their shorter replies and later logins meant disengagement, not a different work style or a personal issue they haven't mentioned."
-
6
Identify the data you actually selected. "Three late logins and two short Slack replies, over two weeks, out of hundreds of interactions."
-
7
Check it against the fuller pool of data. "They also shipped the hardest feature this quarter, and told me a month ago their kid was having a rough time at school — which I filed away and forgot."
-
8
Climb back up on purpose. New conclusion: "I don't actually know what's going on — I noticed a pattern and built a story around it without checking." New action: ask a direct, private, curious question instead of making a public accusation.
6Common mistakes & when to hold off
Common mistakes
Using it as a weapon
"You're just climbing the ladder" is not a rebuttal. Pointing out the model at someone, rather than walking it with them, turns a thinking tool into a conversation-ender.
Treating "more data" as the whole fix
Climbing down to the data rung helps, but you also have to question the meaning and assumptions you added — more facts fed into the same flawed interpretation still produce the same wrong conclusion.
Only applying it to other people
It's tempting to diagnose a colleague's "leap up the ladder" while treating your own read of the situation as simply correct. The tool only works pointed at yourself first.
Doing it after the damage is done
Reconstructing your reasoning is useful for learning, but the real value is catching the climb before you act — not narrating it afterward as a polished excuse.
When not to use it
Time-critical, low-stakes calls
If a wrong guess costs you five minutes and a right one saves an hour of deliberation, just decide. Save the full climb-down for decisions that are hard to reverse.
Genuine emergencies
If someone is in physical danger or a system is actively failing, act on the best available read and audit the reasoning afterward, in the postmortem.
Disputes that are actually about values
If both sides agree on the facts and simply want different things, this model won't resolve it — that's a negotiation problem, not an inference problem.
As a stalling tactic
Endlessly "checking your assumptions" can itself become an avoidance strategy. At some point you have enough data to act and should.
7Related models in this series
The Ladder of Inference pairs especially well with:
8Try it yourself
Describe a real situation and get a personalized set of ladder questions to ask yourself before you act.
9Save as Skill
Turn this model into something an AI assistant can apply to your actual problems, any time.
This is the portable skill file. Copy it into any AI assistant's custom instructions, or download it as Markdown and attach it to a project.