Keep the stones
Everyone keeps their own stones. They can shield themselves from attacks, but we assume some danger remains.
A thought experiment / Research note
Who gets the
Infinity Stones?
Hand them to one person. Hope they destroy them. Now suppose the factories can make more.
We proved a small but surprising result: the average time until the stones return cannot, by itself, guarantee that a risky handover is worth it.
Try the three worldsThe scenario belongs to CYBRDELIC (@cybrdelic) ↗
31 August 2026. The model and its assumptions are ours.
The rules we added
Destruction is a pause.
It may not be an ending.
Risk the handover
One person gets all the stones. They might use that power against everyone. If they destroy the stones instead, the world is safe for a while.
Meet the factories again
Factories eventually make more stones. You face the same choice again. Every handover is a fresh gamble: a past success does not make the next one safe.
What “better” means here: more safe time, on average. These are made-up risks for a thought experiment.
Experiment 01 / Change the pattern
Twenty years on average.
Three different promises.
Every world below has the same 20-year average pause. Move the risk, then change how often the uneven world returns early. The other assumptions stay the same.
A probability per attempt, not per year.
Early means 1 year. The long pause adjusts to keep the average at 20.
Change the other assumptions
The example starts with about a 2% yearly danger while keeping the stones, and gives nearer safe years extra weight. Try changing either.
Chance of catastrophe in a year while everyone keeps their own stones.
At 3%, a safe year 23 years from now counts half as much as one today. At 0%, every year counts equally.
Chosen handover risk Highest handover risk worth takingCommon scale: 0–100%
AHandover is better
A reliable pause
Always returns after 20 years.
Risk worth taking, up to23.12%
Safety score23.96
BHandover is better
An unpredictable pause
The chance of returning stays the same each year. Average: 20 years.
Risk worth taking, up to20.00%
Safety score22.67
CKeeping is better
An uneven pause
90% return after 1 year; 10% after 191 years.
Risk worth taking, up to7.76%
Safety score13.90
Keeping the stones20.00safety score
Same average, opposite answers: handover helps in the reliable world but hurts in the uneven one.
The same average, in one line
90% × 1 + 10% × 191 = 20 years
For the uneven world, multiply each pause by its chance, then add. Rare long pauses can hide lots of early returns.
How to read this: if the black line is inside the purple bar, handover is worth its risk. The safety score adds up the safe years you can expect, giving nearer years extra weight if you chose that above. Higher is better; the green box shows the score for keeping the stones. Each card assumes you repeat the handover whenever new stones appear.
Experiment 02 / Change what you know
An average is a hint.
A limit is a guarantee.
The average is still 20 years. Now imagine you do not know the pattern—only how early or late the stones could return. Is that enough information to make a decision?
This uses the risk and safety settings you chose above.
Worst possible pattern7.76%
toBest possible pattern23.12%
0%50%100%
The known limits still allow opposite decisions.
The worst pattern mixes the earliest and latest returns. The best returns after exactly 20 years every time. Other patterns can give any answer between these two.
Green: handover is worth it for every pattern allowed by your settings. Stripes: you need more information. Beyond the stripes: keeping wins. Exactly at a boundary, a tie may be possible.
The proved result / Within this model
The average alone
cannot clear the handover.
If you give any extra weight to safety soon, an average pause alone cannot justify even a tiny handover risk. Some pattern with that same average will make keeping the stones better.
Make early returns very common. Offset them with rare, very long pauses. The average stays fixed while the tolerable handover risk approaches zero.
Knowing the stones cannot return for a certain time changes that. So does permanent removal of production. The theorem identifies the missing information; it does not say disarmament always fails.
Read the complete proof ↗One more twist
Perfect patience changes
what the average can tell you.
Count a safe year far in the future just as much as a safe year today, and the average becomes enough. All three cards agree. Move “How much nearer years matter” to zero to see it.
Give even a little extra weight to nearer years, and some extremely uneven pattern can still spoil that guarantee. As you become more patient, that troublesome pattern becomes more extreme. No single world suddenly jumps from one answer to another.
Show the work
Proof first.
Checks from another direction.
The written proof covers every comeback pattern allowed by the model. We also checked the calculator a different way: follow the world through time, adding up safe years and tracking failed handovers.
Verification record ↗8,820exact arithmetic checks
62checks by following time
1,000comeback-pattern checks
6examples that defeat the average
Read before generalizing
The assumptions do real work.
Fresh risk, every time
This is not one immortal person's hidden good-or-evil personality. If a successful first handover proves that the person is permanently trustworthy, the comeback pattern stops mattering. The proof explains that alternative too.
We only count safety
We do not price freedom, inequality, useful applications, or the incentives of an elite. The note addresses a narrower question than the original three-way dilemma.
The rules stay the same
We assume each round starts fresh, with the same rules. Factories and handovers do not influence each other's luck. Learning, better technology, or people deliberately rebuilding weapons could change the answer.
What is new here?
This note pins down exactly what an average and time limits can tell us in this scenario. The underlying maths and the problem of trusting someone during disarmament are older ideas. We have not established that nobody has derived this result before. External peer review is still pending.
Credit where it belongs
A tweet started it.
Older ideas made it precise.
- 01CYBRDELIC · The original Infinity Stones scenario ↗
31 August 2026. Original scenario credit; no coauthorship or endorsement implied. The follow-up invitation ↗
- 02Barbara F. Walter · The Critical Barrier to Civil War Settlement ↗
1997. Losing the ability to protect yourself during disarmament was already a serious research problem.
- 03Robert Powell · War as a Commitment Problem ↗
2006. Agreements that change future power can undermine their own enforcement.
- 04Robert Gallager · Renewal processes ↗
MIT notes, Chapter 4. The mathematical machinery for a system that starts again.
- 05Albert Madansky · Convex expectation bounds ↗
1959. The endpoint bound is classical; we use it rather than claim to have invented it.