Cut the Downside, Keep the Upside Open

Four lessons ago you met two traders, one winning 80% of her bets and going broke, the other losing 80% of his and getting rich. Since then you’ve made the payoff shape precise (the smile and the frown), watched a convex bet lose most of its battles and win the war, learned to manufacture asymmetry with optionality, and seen the barbell and Jensen’s inequality turn all of it into a strategy. This is the capstone. It compresses everything into a single decision rule you can say to yourself out loud before any risky choice — and then shows you the three ways the rule quietly turns against you if you’re not careful.

The rule is shorter than the course. The rest of this lesson is why it works, and how it breaks.

The heuristic

Here it is, the entire course folded into one line:

Cut the downside, keep the upside open — run many small reversible experiments, and let the winners run.

That single sentence has four moving parts, and each one is a lesson you’ve already taken, wearing work clothes. Let’s unpack them.

(a) Cut the downside. Cap every loss so that no single one can end the game. This is the margin of safety in one phrase: before you ask what you might win, make sure the worst case is survivable. A bet you can walk away from is a bet you get to keep making. A bet that can wipe you out is played exactly once, whether you meant it to be or not. Convexity starts here — the capped downside is the left half of the smile.

(b) Keep the upside open. Never clip the rare, huge win. The entire arithmetic of a convex bet lives in the one outcome that pays for all the others, so the instant you put a ceiling on the good case, you’ve amputated the thing that made the bet worth taking. Most decisions in life have hidden ceilings people install by reflex — “I’ll sell when it doubles,” “I’ll cap my exposure to any one idea” — and each one trades away the fat right tail for a little comfort. Keep it open.

(c) Run many small reversible experiments. One convex bet is a lottery ticket; many convex bets is a strategy. Pay a small, known premium — of money, time, or attention — for each of a large number of shots, so that the eventual winner has room to appear. This is optionality bought in bulk. Small enough that any single loss is a rounding error; numerous enough that the rare payoff is nearly certain to show up somewhere in the batch.

(d) Let the winners run. Kill losers fast, ride winners long. Convex portfolios are lopsided by design — a few positions carry everything — so the discipline is asymmetric on purpose: cut what isn’t working before it drains you, and refuse to trim what is working just because it feels prudent. Averaging your winners down to your losers is how you manufacture a concave payoff out of convex parts.

A worked example: ten cheap bets

Put real numbers on it. You place ten small convex bets, each risking one unit. Eight die, one breaks even, one runs to +30.

Your win-rate is 10% — you’d have to explain to anyone watching that ninety percent of your bets “failed.” And you finished up +22 units on a 10-unit stake, a +220% return, entirely because the downside on each loser was capped at −1 and the upside on the one winner was left open to +30. Now watch how fragile that result is to the two things this course keeps warning about:

• Clip the winner at 2× (sell it the moment it doubles) and bet 10 becomes +1 instead of +30. Total: 8 losers (−8) + break-even (0) + a clipped winner (+1) = −7 units. You turned a +220% strategy into a loss by capping the one payoff that mattered. That’s failure mode (b).
• Uncap one loser — let a single “small” bet run to −25 because you didn’t bound it — and even with the +30 winner intact you’re at 7 losers (−7) + a runaway loser (−25) + break-even (0) + winner (+30) = −2 units. One uncapped downside erased the entire edge. That’s failure mode (a).

The +22 wasn’t luck. It was geometry: capped left, open right, run enough times, let the tail arrive. Break either rule and the geometry collapses.

When to use it

Reach for the heuristic whenever a decision is repeatable, its losses can be bounded, and its upside is genuinely open-ended — new ventures, experiments, skills, introductions, creative shots, research directions, hiring. It is not the rule for one-shot decisions with symmetric stakes (there’s no batch for the tail to show up in) or for anything whose downside you cannot actually cap (see the traps). When the shape fits, the rule replaces the anxious question “will this work?” with the useful one: “is my loss capped, my upside open, and am I taking enough shots?”

Reversibility: prefer two-way doors

Notice the word reversible smuggled into the heuristic. It’s doing more work than it looks. Jeff Bezos calls the two kinds of decisions one-way doors and two-way doors. A two-way door is a choice you can walk back cheaply — try it, and if it’s wrong, return to where you were at little cost. A one-way door slams behind you: undoing it is expensive, slow, or impossible.

Here’s the connection that ties reversibility straight back to the course: a reversible experiment IS an option. Recall from lesson 3 that an option is the right but not the obligation to do something — you keep the upside and can decline the downside. A two-way door is exactly that. You pay a small premium to try, and you retain the right to walk away if it disappoints. The reversibility is the optionality. That’s why cheap-to-abandon experiments are the raw material of a convex strategy: each one caps your downside by construction, because you can quit.

An irreversible move throws that right away. Betting the whole company on one launch, taking on debt you can’t unwind, burning a relationship, making a public commitment you can’t retract — each converts a two-way door into a one-way door, which converts an option into an obligation, which uncaps your downside. The design principle falls right out: structure your bets as two-way doors wherever you can, and spend your scarce irreversibility only where you’re most sure.

And here is where reversibility meets “let the winners run.” The failure of clipping winners early is really a failure of walking back through a door you should have left open. Selling the rare 30× at 2× isn’t caution — it’s closing your own option. You paid the premium (eight losers) precisely to own the right to the tail, and then you sold that right for pocket change the moment it started to pay. It destroys the whole convex edge because the edge was the tail. Reversibility should protect you from losers, not tempt you out of winners.

Drive it one more time

You’ve driven the payoff explorer before; drive it once more, now as a test of the whole heuristic. The claim of this entire course is that a bet with a capped downside can be worth taking even when it loses most of the time — and that the cap, not the odds, is what keeps it alive. Prove it to yourself.

Try this, in order:

1. Build a convex bet. Set a tiny downside cap (lose a little when you lose), a big upside, and a low win-rate — make it lose most of the time. Run 100 trials.
2. Watch the total. It climbs, despite the bet losing on the majority of trials. That’s the smile working: the rare wins are large enough that a handful of them overpays for the many small losses. Win-rate is not the scoreboard.
3. Now break it — but change only ONE thing. Leave the win-rate exactly where it is. Raise the downside cap — let each loss hurt more and more. Keep re-running. At some point the same win-rate that just made money turns the total negative.

That last step is the punchline of the course. You didn’t change how often you win. You changed how much you lose when you lose — and that alone flipped a winning strategy into a losing one. The cap, not the odds, is what keeps a convex bet alive. Cut the downside and a 10%-win bet can print money; uncap it and even a 40%-win bet can bleed out.

Three traps

The heuristic is powerful, which means it’s dangerous when misapplied. Three traps turn “cut the downside, keep the upside open” against the person using it. Each one looks like the model working and is actually the model failing. Learn to smell all three.

Trap 1: Paying too much for optionality

Optionality is not free — lesson 3 was emphatic about this. Every option carries a premium, and a convex payoff is only good if the premium you pay is small relative to the upside you’re buying. Forget that, and you fall for the seductive lie that anything convex is worth doing.

Consider a lottery ticket. Its payoff is beautifully convex — lose $2, maybe win $10,000,000, capped downside and enormous upside. It is also, in expectation, a terrible bet, because the price of the ticket vastly overpays for the microscopic odds. Convex and overpriced is still bad. The shape being right does not excuse the price being wrong. Overpaying for stock options at a nosebleed valuation, buying out-of-the-money contracts everyone else has already bid up, spending a year of your life chasing a “moonshot” whose true odds don’t justify the year — all convex, all potentially awful, because the premium ate the edge. Keep the upside open, yes — but only when you’re paying a fair or cheap price to hold it open.

Trap 2: A “capped” downside that isn’t

This is the deadliest of the three, because it wears the model’s own uniform. You tell yourself the downside is capped — that’s the whole point of the strategy — and you’re wrong. The loss you called bounded is secretly open-ended, and you’ve built a concave frown while congratulating yourself on a convex smile.

The classic example is selling put options (or writing any insurance): you collect a small premium and, in exchange, promise to eat someone else’s loss if things go bad. It feels like income with a floor. It is a capped upside sitting on an open-ended downside — the exact frown. Same trap: unhedged leverage (“I can only lose what I put in… unless a margin call forces more”), a currency peg that “can’t break” until it does, a supplier who “would never” fail. The tell is always a sentence with a hidden unless: “it can’t go below zero… unless it can.” Before you trust a cap, ask what event would blow through it — and whether you’ve actually excluded that event or just declined to imagine it. A margin of safety you assumed is not a margin of safety you built.

Trap 3: Confusing variance with asymmetry

The subtlest trap. Asymmetry — the lopsided shape — is what you want. Variance — mere variability, a wide range of outcomes — is not the same thing, and the two are easy to mix up because both feel “risky” and “exciting.”

Take a symmetric coin flip: win $100 on heads, lose $100 on tails. It is enormously volatile — huge swings, a wide spread of results — and it is worthless, with an expected value of exactly zero and no edge whatsoever. Widening it to win/lose $1,000 adds variance and still has zero edge. High variability is not an advantage; a wide symmetric distribution just means you’ll be jerked around more for the same nothing. What you actually need is the lopsided shape — small on one side, big on the other — not a big shape. Convexity is a statement about the asymmetry between the two tails, not the width of either. Chase the shape, not the volatility: a quiet bet that loses a little and rarely wins a lot beats a wild one that swings symmetrically around zero every time.

Asymmetry & optionality, in one picture

You’ve now walked the whole ladder. Here is the entire course as a single map — five lessons, each a rung, ending at the rule you can carry out the door.

Next up

That’s the whole ladder climbed: from the shape of a single payoff to a decision rule you can say out loud before any risky choice. One step remains, and it’s the one you can’t undo — the Final Exam. It’s graded and one-way: questions come one at a time, submitting locks your answer for good, and your pass/fail score appears only at the end. No back button, no retries, no restart. Everything you need is in the five lessons you’ve just finished — so before you open the door that only swings one way, make sure you can cut the downside, keep the upside open, and spot all three traps without hesitating.