Physics of Progress
Turn every failure into a sample that improves your next prediction
- Difficulty
- Moderate
- Time to result
- ~ongoing to results
- Steps
- 5
- Confidence
- 99%
Treat progress as the construction of a prediction engine. Begin with a desired result, take an action, and observe what the world returns. That result is a sample: evidence about the relationship between behavior and outcome. A miss is therefore not a verdict on the person; it narrows the possible rules of the game. Reflect on what the painful result reveals, adjust the behavior, and run another trial. Across repetitions, the accumulated samples make predictions more accurate, so you can see what to try next even though a changing world can never guarantee arrival. The emotional sting still matters because it prompts reflection, but it should not be allowed to break the learning loop. Progress emerges from pain plus reflection, not from avoiding failure.
Origin
Tom compared human trial and error with an AI learning a game from samples, then named the resulting prediction-building loop the physics of progress.
Core principles
- 01Failure is data before it is identity
- 02Every attempt yields a sample
- 03Reflection converts pain into a revised prediction
- 04A better prediction engine reveals the next experiment
How to run it
- 1
Name the target
Specify the outcome your prediction engine is trying to reach.
Pro tip Choose a result you can observe rather than a feeling you cannot measure.
- 2
Run one trial
Take a bounded action and note what you expected it to produce.
Watch out Changing many variables at once weakens the sample.
- 3
Capture the sample
Record what actually happened and separate that evidence from the story you tell about yourself.
Pro tip Describe behavior and outcome in neutral language.
Watch out Treating the miss as identity stops the loop.
- 4
Reflect on the gap
Ask what the mismatch teaches about the rules, environment, or your current skill.
Pro tip Let pain trigger inquiry instead of avoidance.
- 5
Predict and retry
Change the next behavior using the new information, then repeat the cycle.
Pro tip Judge improvement by prediction accuracy, not by never missing.
Watch out No prediction remains guaranteed in a changing world.
In the wild
Manufacturers said Quest's protein bar could not be made on standard equipment. Instead of accepting the failed production path as final, the team examined viscosity, stickiness, pressure, and equipment assumptions. The sample revealed a tooling constraint rather than a physics constraint, leading them to engineer equipment with different tolerances.
→ The bar became manufacturable and the capability contributed to Quest's success.
Common mistakes
Personalizing the sample
Turning an unsuccessful attempt into proof that you are a failure discards its information value.
Reflecting without changing
Insight only improves the prediction engine when it changes the next trial.
Is it for you?
Best for
People learning a skill or building in conditions where no reliable playbook exists.
Not ideal for
Irreversible experiments where a failed attempt creates unacceptable harm.
From the transcript
“So, you try a thing, it gives you a piece of data. This is a sample. Now, you know a little bit more about the…”
“And if people understood that you're running trial and error as a way of building up a prediction engine, so that you know, oh, when…”
“And so when you fail, it hurts, but that sting causes you to reflect and say, "Okay, I don't want to feel this way again.…”
From the episode
Tom Bilyeu: The Billion-Dollar Entrepreneur Mindset That Turns Failures into Success
Tom Bilyeu