The Valve You Can Turn

Opening Scene

The most convincing way to prove two pipes are connected isn’t tracing a schematic or reasoning about likely confounders — it’s finding a valve on one pipe, turning it, and directly observing whether water flow changes at the other end. Deliberate intervention, not observation, is the gold standard. This is exactly what a randomized controlled trial does for any causal question where it’s genuinely possible to run one.

In Plain English

A randomized controlled trial (RCT) deliberately assigns a treatment at random and measures the resulting effect, exactly as covered in depth throughout this content library’s dedicated experimentation and A/B testing series. It remains the strongest possible method for establishing causation, because random assignment specifically rules out confounding (Article 4) by ensuring the treatment is unrelated to any other factor that might also affect the outcome.

The Old Way

Before RCTs were standard practice, causal claims relied on the weaker observational reasoning covered in Articles 3 and 4:

Medical treatments historically judged by whether patients who received them recovered, without a randomized comparison group to reveal what would have happened anyway.
Policy interventions historically judged by before-and-after comparisons, without addressing the confounding factors that might also explain a change.
Business decisions historically judged by informal before-and-after observation, the exact trap covered at length in this content library’s experimentation series.

RCTs emerged specifically as the rigorous alternative to these weaker forms of causal reasoning.

What’s Changing (and Why AI Is the Reason)

Digital experimentation, covered in far more depth throughout this content library’s dedicated series, has made running genuine RCTs dramatically faster and cheaper than in earlier eras, expanding how many causal questions can actually be tested this rigorously.
As more decisions become automatable and testable at scale, RCTs have moved from a specialized scientific method to a routine business tool for establishing genuine causal effects.
Not every causal question can be tested with an RCT — ethical, practical, or logistical constraints often rule it out, which is exactly why the rest of this series covers the observational methods needed when the valve genuinely can’t be turned.

The Metaphor, Fully Extended

Behind the Wall	Randomized Controlled Trial Concept
Finding a valve and deliberately turning it	Deliberately assigning a treatment at random
Directly observing whether flow changes at the other end	Directly measuring the resulting change in outcome
The strongest possible proof a connection is real	The strongest possible method for establishing causation
Cases where there’s simply no accessible valve to turn	Cases where a randomized trial genuinely isn’t possible

For Beginners: What to Actually Do

Learn the core logic of randomized controlled trials in depth from this content library’s dedicated experimentation series — it’s the foundation this entire causal inference series builds on and departs from when necessary.
Practice recognizing when a causal question genuinely could be tested with an RCT but isn’t being, versus when an RCT genuinely isn’t feasible.
Get comfortable treating an RCT’s result as the strongest available evidence, even when it doesn’t match an intuitively appealing observational finding.

For Practitioners and Leaders: The Deeper Layer

Default to a randomized controlled trial whenever a causal question can genuinely be tested this way, rather than settling for weaker observational evidence out of convenience.
Recognize the real ethical, practical, and logistical constraints that sometimes rule out an RCT, and plan for the observational alternatives covered in the rest of this series.
Connect this series explicitly to this content library’s dedicated experimentation series — the two are deeply complementary, not competing bodies of knowledge.

Quick Recap

A randomized controlled trial deliberately assigns treatment at random, directly measuring the resulting causal effect.
Random assignment specifically rules out confounding, making RCTs the strongest available method for establishing causation.
Digital experimentation has made RCTs faster and cheaper to run than in earlier eras.
Not every causal question can be tested with an RCT, which is why the rest of this series covers observational alternatives.

Where This Fits in the Series

Article 6 covered the gold-standard method for the cases where it’s available. Article 7 covers what to do for the far more common cases where it isn’t.