Iterations Over Hours
Mastery comes from completed learning loops, not from accumulated time on task.
- Difficulty
- Moderate
- Time to result
- ~ongoing to results
- Steps
- 7
- Confidence
- 84%
The mechanism reframes the unit of mastery from hours to loops. One iteration has four parts: you do something, you look at and test the result against an external grader such as a free market, nature, or physics, you ask what part of the experiment worked and what did not, and you make a new creative guess about how to improve it. Then you run it again. The rate at which you can complete that rotation is the learning curve you are actually on, which is why two people with identical hours logged can be at completely different levels of skill. Two consequences follow. First, you must tolerate throwing away most of your work, because learning necessarily involves failure and the discarded attempts are what produced the insight. Second, insights compound: each loop distils something that builds on the last, which is why a fast-iterating team stays ahead of a larger one that copies the current state.
Origin
Extracted from the Naval podcast episode Curate People, offered as a correction to the popularised ten thousand hours rule and paired with Balaji Srinivasan's idea of wandering through the idea maze.
Core principles
- 01Ten thousand hours is directionally correct but measures the wrong variable.
- 02The learning curve is driven by iteration count, not elapsed time.
- 03An iteration is only real if the result is tested against something external.
- 04Free markets, nature, and physics are the honest graders.
- 05Learning necessarily involves failure, so most work will be thrown away.
- 06All new information starts out looking like misinformation.
How to run it
- 1
Define the loop
Write down what one complete iteration of your work looks like, from attempt through to an externally tested result. Vague work has no loop and therefore no learning rate.
- 2
Run the attempt
Do the thing rather than deliberating about it. The output of this step is an artefact that can be tested.
- 3
Test against an external grader
Put the result in front of a free market, nature, or physics rather than your own judgement or an internal committee.
Pro tip Choose the harshest available grader; a soft one flatters you and stops the loop from teaching anything.
Watch out Internal opinion feels like a test but only confirms what you already believed.
- 4
Ask what worked and what did not
Decompose the result rather than scoring it pass or fail. The point is to isolate which part of the experiment carried the outcome.
- 5
Make a new creative guess
Form a fresh hypothesis about how to improve based on what the test showed, then immediately run the loop again.
- 6
Distil and keep the insight
Capture what the iteration taught so that later loops build on it. Great operators distil an insight from every single iteration.
- 7
Compress the cycle time
Attack whatever makes each loop slow, because rotation speed, not effort, sets the curve you are on.
Watch out Pride is the biggest impediment; staying locked into the original vision stops you backtracking when the loop says to.
In the wild
A startup breaks out and the assumption is that a large incumbent will simply copy it and crush it. In practice the startup has been taking left turns, right turns, and backtracks through the idea maze for years, so by the time the incumbent arrives at the startup's visible position, the startup has moved on. The incumbent then cannot resist exploring side hallways the startup already mapped and knows are dead ends. The advantage was never the current feature set; it was the accumulated iteration count.
→ The smaller team stays permanently ahead despite being copied, because copying transfers state but not loops.
Two people each spend a year on the same craft. One produces work continuously without ever testing it externally, accumulating hours. The other ships smaller pieces to a real audience, examines what landed, forms a new guess, and repeats. At the end of the year the hours logged are identical and the skill levels are not, because only one of them completed a large number of full loops.
→ Iteration count, not time served, explains the gap in capability.
Common mistakes
Counting hours instead of loops
Time on task without an external test produces volume, not learning. The rotation is the unit that moves you down the curve.
Grading yourself
Testing a result against your own judgement rather than a market, nature, or physics closes the loop with the same bias that produced the attempt.
Refusing to discard work
Pride keeps people locked into an original direction. Learning requires failure, so a process that throws nothing away is not a learning process.
Is it for you?
Best for
Founders, engineers, and operators trying to get down a learning curve faster than a better-resourced competitor.
Not ideal for
Domains where each attempt is irreversible or catastrophically expensive and rapid iteration is not survivable.
From the transcript
“It's not just hours put in, it's iterations. How many learning loops do you have that drive the learning curve?”
“Iteration is when you do something and then you look at the result, you test the results somehow, ideally against a free market, nature or…”
“Great people will distill insights from every iteration.”
From the episode
Curate People