Compounding vs Additive Gains

A compounding gain leaves you more capable than you were before the tutor, the tool, or the extra hours arrived. The gain stays after those inputs stop.

An additive gain is real while those inputs run. When they stop, the gain goes with them, and you return to the earlier baseline.

Core takeaways

• The whole difference between the two kinds is whether each gain updates the base that the next gain is calculated on.
• Bought support (tutoring, extra resources, guided explanation, tools that do the work) raises performance above real ability. When the support is removed, the result drops back.
• In money the balance is the account. In learning the balance is you. Effort to improve belongs on the processes upstream of the grade.
• A skill is a balance of its own. Each 1% gain in a skill is taken on the raised base, inside one subject, with no transfer needed.
• The removal test tells the two kinds apart: close the tool or stop the support, then do the next comparable task. A compounding purchase needs less support next time. A purchase that keeps needing to grow is additive.
• Until a technique runs in your regular work, count yourself as not knowing it. The removal test has not started.

The arithmetic with money

With money the two kinds can be written out as arithmetic. Put $100 to work at 1% a cycle. The first cycle pays $1, the balance becomes $101, and the next 1% is taken on $101. The gap against an account still at $100 widens every cycle.

The additive version uses the same numbers. The first cycle still pays $1. That dollar is moved into a separate account, and the next calculation does not include it. Cycle two is 1% of $100 again, and so is cycle three. The money is real and it accumulates. The balance never moves, so no calculation is ever made on a higher number.

A 1% gain every day, added in a straight line, is 365% in a year. The same 1% compounded is roughly 37 times the starting amount. This is an arithmetic illustration. It is not a result from a study. In practice the rate is slower: 1% a week will not make a year 3,600% better, only substantially better. On any single day the two versions are $1.01 against $1.00, a difference too small to notice. Early on, the compounding curve sits below the additive line. Later it overtakes the additive line by several fold. That early lag is the cost of choosing the slower method. The entire difference between the two curves is whether each gain updates the balance.

Support you buy

Bought support behaves like the dollar moved into the separate account. Tutoring, extra resources, and guided explanation raise performance above real ability. Compare a game character at level one wearing level-six gear. It can clear level-five content. Take the gear off and the character is still level one. In learning, knowledge and results drop the same way when the support is removed. The research in WPW records this fade as a general pattern: effects fade when the environment that produced them ends, unless the person built a skill that would not have developed anyway.

Reliance on bought support grows over time. Course content gets harder and targets go up, so the borrowed support has to keep growing. Early on, the limit is how much support you can obtain. Past a point, the limit can no longer be bought. It becomes how much you can hold and use in the hours you actually have, alongside coursework, projects, and a life outside the subject. The limit is then your own capacity. Buying more support does not change that limit.

Buy 1% more support and you get 1% more of the same kind of result, added to the old base. You never get 1% on top of a raised base. A bought input cannot compound, because nothing about you changed between purchases.

The same pattern applies to hours of study. Imagine one hour of study gives an 80% result and two hours give 90%. These are worked numbers, not a measured rate. Drop back to one hour and the result drops with it. Past a threshold, added hours make the result worse: stress climbs, efficiency falls, and sleep, friends, and hobbies get less time. This threshold is a rough limit. It is not a fixed rule.

Some additive spend is unavoidable. A mix that is purely compounding does not exist. The useful habit is to know which kind you are buying before you spend. People keep buying the rented result anyway, because giving up a gain already in hand feels like a loss, whatever it costs to keep. That is the mechanism in Loss Aversion.

In learning, the balance is you

In money the balance is the account. In learning the balance is you. Wanting the outcome changes nothing. A person who wants a hole in a wall and punches it barehanded still works slowly and gets hurt, however much they want the hole. The odds improve when the method improves. A grade is a reading of studying skill, in the way a speedometer is a reading of speed. Tapping the speedometer dial does not move the car, and working on the grade directly does not raise the skill. Treat this as an instruction about where to put effort. It is not a full account of what grades measure: results also depend on prior knowledge, test design, and sleep. Effort to improve belongs on the processes upstream of the number. That is the wider rule in Locus of Control.

What a skill gain raises

A skill is itself a balance. Take it at 100. Improve it 1% and it stands at 101. The next 1% is taken on 101, which puts it at 102.01. This holds inside one subject with no transfer anywhere. Each gain in a skill raises the base the next gain is taken on.

Improving recall, conceptual understanding, problem-solving on routine and unfamiliar questions, or execution at speed improves the person doing every future hour of study. Changing how the hour is spent lifts the result of that same hour, 80 to 90 to 100, with no added time and no rented support. These numbers are again an illustration, not a finding.

The same skill gain can apply across everything you study at once. This cross-subject reach is a second, separate effect. It is not the compounding mechanism, and it is less certain than the within-subject gain. A gain in how you learn has a reach that subject-specific help does not have. Science tutoring does nothing for maths, unless the tutor is also training a shared study skill. One picture for this: subjects differ as blue paint differs from yellow, and learning them is the same technique with the brush. That picture claims more than transfer research supports. Keep the reach as a reason to prefer skill gains. Do not treat it as a measured multiplier.

The processes a result actually reads are exam technique, accurate recall of detail, conceptual understanding, problem-solving across routine and unfamiliar questions, all of it executed at speed under a clock, plus calm under pressure and time management. Aim the gains at becoming that person. Exams still have variance. Becoming that person is the aim, and it is not a guarantee.

Tools that do the work and tools that train you

A tool that performs the work delivers output. A tool that trains a capability you keep delivers a change in you that stays. The removal test separates them on the spot. Drafts, code, and analysis produced for you are output, the same as the gear in the game example: they raise the result without raising you. The spend compounds only when the interaction leaves you better at the work once the assistant is closed.

A compounding purchase shows up as the same result at lower cost: the next comparable task takes less support, not more. A purchase that keeps needing to grow has already failed the test.

Techniques you have not practiced yet

A technique you have not practiced counts as a technique you do not know. Knowing about a technique gives a false sense of security, and the details only show up when you use it. Until it is running in your regular work, count yourself as not knowing it. The diagnostic for this is in Are You Learning, or Just Using Techniques. The removal test assumes there is something to remove. Until a technique runs regularly in real work, the test has not started.

How to practice this

1. Before you pay for a tutor, a tool, or extra hours, name which kind of gain you are buying. Ask whether the result would stay if the input stopped. If the answer is no, the gain is additive.
2. Run the removal test on a tool you use. Close it and do the next comparable task without it. Notice whether the task takes less support than last time or more.
3. Watch a purchase over several rounds. Notice whether the support has to grow to hold the same result. If it has to grow, it has failed the test.
4. Move your effort from the grade to the processes upstream of it: recall, understanding, problem-solving, and speed. Notice whether the same hour of study gives a better result with no added time.
5. For a technique you have read about, check whether it runs in your regular work. If it does not, count yourself as not knowing it. Notice that the removal test has not started for it.
6. Track added hours of study against how you feel. Notice the point where stress climbs and efficiency falls. That is the threshold where more hours make the result worse.

Related pages

• Marginal Gains: owns stacking, REDO and rate limiters, the machinery for accumulating gains once the removal test has decided which ones are worth stacking.

Open questions

How much of a study-skill gain actually crosses subjects, when far transfer is the exception and the reach is what makes skill gains worth preferring?

Where does the additive floor sit for a given person? How much bought support is the correct purchase rather than a rented result?

Sources

• Bailey, Duncan, Cunha, Foorman & Yeager 2020, Psychological Science in the Public Interest 21(2): educational-intervention effects commonly diminish once the intervention ends. Supports fade-out as a class; not a tutoring-hours study.
• Dunlosky, Rawson, Marsh, Nathan & Willingham 2013, Psychological Science in the Public Interest 14(1): how an hour is spent moves retention more than adding hours of rereading.
• Plant, Ericsson, Hill & Asberg 2005, Contemporary Educational Psychology 30(1): quantity of study poorly predicts results once quality and prior achievement are controlled.
• Barnett & Ceci 2002, Psychological Bulletin: far transfer is the exception; the bound on the cross-subject reach claim.
• Clear, Atomic Habits: the popular home of the 1%-a-day / 37× illustration (1.01^365 ≈ 37.78).