The Wall Everyone's Afraid to Open
two pipes run side by side behind a wall, and only one of them is actually connected to the leak. Guessing which one is expensive and often wrong.
Part 1
Part 2
What the Plumber Is Actually Tracing
the real difference between two pipes simply running near each other and one pipe actually feeding into the other.
Part 3
Guessing by Proximity
why 'it's probably the nearest pipe' is exactly the reasoning that leads to cutting into the wrong wall, and why folk causal reasoning fails the same way.
Part 4
The Third Pipe
how a single hidden pipe feeding two separate fixtures can make them look causally connected to each other when neither one actually is.
Part 5
Tracing the Real Connection
how causal graphs let you draw the entire plumbing system on paper before deciding which wall is actually worth opening.
Part 6
The Valve You Can Turn
why deliberately turning a valve and watching what happens remains the single strongest way to prove a causal connection, whenever it's actually possible.
Part 7
When You Can't Turn the Valve
why most real causal questions can't be tested with a randomized trial, and why that doesn't mean the question is unanswerable.
Part 8
A Pipe That Only Sometimes Carries Water
how instrumental variables use a natural, unrelated source of variation to mimic a randomized trial even when nobody actually ran one.
Part 9
Matching Similar Pipes
how comparing naturally similar cases that happen to differ in one specific way approximates a controlled comparison without a true randomized trial.
Part 10
Before and After the Leak
how comparing the change in one affected pipe against the change in a similar, unaffected pipe over the same period isolates a treatment's real effect.
Part 11
The Threshold Where the Water Starts Flowing
how a sharp, arbitrary cutoff — like a pressure valve that trips at exactly one number — creates a natural experiment right at that threshold.
Part 12
Two Pipes That Look the Same From Outside
how selection bias quietly distorts a comparison before any statistical analysis even begins, by determining who ends up in the data at all.
Part 13
A Leak That Fixes Itself
why a pipe that seems to cause a symptom might actually be responding to it, and how getting the direction of flow backward flips the whole diagnosis.
Part 14
Tracing the Whole System
how mediation analysis reveals whether a cause reaches its effect directly, through an intermediate step, or both.
Part 15
When the Plumbing Changes Mid-Repair
why a confounder that shifts over time needs a genuinely different approach than one that stays fixed, and why treating it as fixed produces a biased estimate.
Part 16
The Inspector's Blind Spot
how to reason honestly about a confounder you can't measure or don't even know exists, rather than pretending the problem away.
Part 17
Reading the System Diagram
why every causal claim rests on a specific set of assumptions, and why stating them explicitly is what separates rigorous analysis from a confident guess.
Part 18
When AI Agents Trust the Wrong Pipe
why an AI system acting on a learned correlation, rather than a verified causal relationship, can make confidently wrong decisions at real scale.
Part 19
The Cost of Getting It Wrong
why a mistaken causal claim isn't just an academic error — it's a decision made on the wrong pipe, with real consequences in medicine, policy, and business.
Part 20
The Whole House, Traced End to End
reassembling the whole plumbing system, from a wall nobody wanted to open to a fully traced, honestly documented account of what actually connects to what.