Three-Deaths Resilience Test
Treat repeated project deaths as prompts to investigate, not automatic verdicts.
- Difficulty
- Advanced
- Time to result
- ~ongoing to results
- Steps
- 5
- Confidence
- 94%
The Three-Deaths Resilience Test is a discipline for interpreting repeated setbacks around unconventional ideas. When a project appears dead, do not count the failure and continue mechanically. Document what specifically failed, distinguish the core hypothesis from the experiment or implementation, and ask whether the result genuinely falsifies the idea. If the explanation reveals a correctable flaw or a meaningful alternate condition, design a new test that addresses it. Repetition matters because many valuable ideas look weak through several early trials, but persistence only earns its place when each death creates new causal knowledge. The framework therefore combines resilience with a stopping rule: continue while curiosity produces credible explanations and better experiments; stop when the evidence attacks the core idea or no plausible improvement remains.
Origin
Nobel-winning drug discoverer Sir James Black told Bahcall, after a discouraging laboratory result, that a drug was not good unless it had been killed three times.
Core principles
- 01Important ideas often encounter multiple apparent deaths.
- 02Early rejection does not distinguish a bad idea from a bad test.
- 03Persistence should be investigative rather than blind.
- 04Each failure should produce a sharper causal explanation.
How to run it
- 1
Name the death
State exactly what result appears to kill the project. Avoid broad labels such as the market hates it or the science failed.
Pro tip Write the failed prediction and observed result side by side.
Watch out Do not soften a genuinely negative result.
- 2
Locate the failure
Determine whether the evidence attacks the core idea, the test design, the implementation, the species, the channel, or another condition.
Pro tip Ask what would have to be true for this to be a false fail.
Watch out A merely possible excuse is not a causal explanation.
- 3
Demand new learning
Extract a specific insight from the failure and use it to define a changed experiment. Repeating the same attempt does not qualify.
Pro tip State how the new test discriminates between explanations.
Watch out Persistence without a changed hypothesis or test is stubbornness.
- 4
Run the revised test
Execute the smallest credible experiment that addresses the suspected flaw. Compare its result with the predicted outcomes.
Pro tip Choose a test capable of proving your rescue explanation wrong.
Watch out Do not design a test that can only confirm the idea.
- 5
Continue or close
Continue if the new evidence preserves the core and reveals a better path. Close the project if the core is falsified or no credible improvement remains.
Pro tip Use curiosity and causal clarity as signals, not the literal number three alone.
Watch out Three failures do not automatically make an idea valuable.
In the wild
Akira Endo continued after cholesterol-lowering approaches were broadly doubted, after the drug failed in mice, and after later negative data drove others away. The mouse result had a species-specific explanation, and further work showed the drug worked in species with relevant cholesterol biology.
→ The investigation preserved a drug category that went on to save millions of lives.
Common mistakes
Counting deaths without diagnosis
The principle is not a punch card that rewards any idea after three failures; each setback must produce causal learning.
Protecting the core from evidence
Calling every negative result a flawed test makes the idea unfalsifiable and wastes resources.
Is it for you?
Best for
Novel research, products, and strategies where uncertainty is high and failure mechanisms can be investigated.
Not ideal for
Projects that violate hard constraints or persist without learning, evidence, or a plausible revised test.
From the transcript
“it's not a good drug unless it's been killed three times”
“all of the really good projects have failed several times before they succeeded”
“sometimes you will get a failure that's not because your idea is bad but because there's a flaw in the experiment”
From the episode
Safi Bahcall: Shoot Your Loonshot
Safi Bahcall