What you are splitting
Result
Each rule's split
The same splits, side by side
Does any subgroup do better on its own?
Stretch by stretch
Example: the Mishnah's estate problem, solved 1,800 years before game theory
Mishnah Ketubot 10:4 has a man die owing his three wives 100, 200 and 300 zuz under their marriage contracts. If the estate is 100, each gets 33⅓; if 200, they get 50, 75 and 75; if 300, they get 50, 100 and 150. For centuries the middle case looked arbitrary. Aumann and Maschler (1985) showed that all three are the nucleolus of the claims game: no one gets more than half their claim until everyone has half, and beyond that the losses are shared the same way. Load the example and change the estate to 100 or 300: the tool's Talmud column reproduces all three, and its nucleolus, found by linear programming with no knowledge of the rule, agrees. Proportional division (33⅓, 66⅔, 100 at 200) and random arrival (33⅓, 83⅓, 83⅓) are shown beside it.
Insolvency law in most places divides an estate among creditors of the same class in proportion to their claims, after secured and priority creditors. The Talmud rule is a different, older answer, shown for comparison.
Questions
What is the Shapley value, in plain words?
Imagine the members join the shared project one at a time. Each one adds some cost: what the group costs with them minus what it cost without them. That extra depends on who was already there, so the Shapley value averages it over every order in which the group could have formed. It is the only rule that divides the whole cost, treats members with the same effect the same, charges nothing to a member who adds nothing, and adds up across separate projects.
What is the core, and why does it matter?
A split is in the core when no subgroup would pay less by leaving and doing the project on its own. A split outside the core gives that subgroup a reason to walk away, which is how shared projects fall apart. The tool checks every subgroup and names the one each split overcharges most.
What is the nucleolus?
Of all splits, the nucleolus is the one that makes the worst-treated subgroup as well off as possible, then the next worst, and so on. When the core is not empty the nucleolus is always inside it, so it is the split to reach for when keeping every subgroup on board is the priority. When the core is empty, it keeps the largest overcharge as small as it can be.
How should neighbours split the cost of a shared private road?
A common rule splits each stretch of road equally among the households that drive on it: everyone shares the first stretch, and only those further out share the stretches beyond. That is the Shapley value of the road (Littlechild and Owen 1973). Splitting in proportion to distance looks similar but can charge the nearest households more than a road of their own would cost; the tool says when it does. Your road maintenance agreement, deed or local law may set a different rule, and it governs.
What if I do not know the cost of every group?
Price each member alone and the whole group, then any groups you can. A group you leave out is priced as the cheapest way to split it into groups you did price, since it could always do that. The tool says how many it filled in. It also flags a priced group that costs more than splitting up would, since that group would rather split.
How does the Talmud divide an estate that cannot pay every claim?
The Mishnah (Ketubot 10:4) has a man die owing his three wives 100, 200 and 300. With 100 in the estate each gets 33 1/3; with 200 they get 50, 75 and 75; with 300 they get 50, 100 and 150. Aumann and Maschler (1985) showed each of these is the nucleolus of the claims game: split equally, but no one gets more than half their claim until all have half, and past half the claims, split the losses the same way. The tool reproduces all three.
Is my data uploaded anywhere?
No. Every split is computed by JavaScript in this browser tab. There are no accounts, cookies or analytics vendors.
Method and citation
Shapley value. Member i pays ΣS∌i |S|!(n−|S|−1)!/n! · (c(S∪i) − c(S)), computed exactly over every subgroup (up to 14 members). On a road, where a group costs what its furthest member needs, this is Littlechild and Owen's closed form: each stretch split equally among those who reach it, at any size.
Nucleolus and core. The excess of a subgroup is what it pays minus what it would cost alone. The tool solves Maschler, Peleg and Shapley's (1979) sequence of linear programmes: find the split whose largest excess is smallest (the least core; the core is empty when that excess is above zero), fix the subgroups that must sit at it, and repeat until the split is unique. Each programme is solved through its dual by a two-phase simplex written for this tool (up to 10 members). Shares are kept at or below each member's stand-alone cost.
SCRB and ENSC (Young, Okada & Hashimoto 1982). A member's separable cost is c(N) − c(N without i); the rest is non-separable. ENSC splits that rest equally; SCRB splits it in proportion to each member's stand-alone cost minus their separable cost.
Claims. The claims game gives a subgroup what is left after paying everyone outside it in full. Talmud rule and its equality with the nucleolus: Aumann & Maschler (1985). Random arrival, the Shapley value of the same game: O'Neill (1982).
Checked against an independent Python implementation (Shapley by every ordering of the members; nucleolus by scipy's HiGHS solver, finding fixed subgroups a different way) on 40 random games of 3 to 7 members, to within 10−11; against the Mishnah's three divisions, the glove game's known Shapley value and nucleolus, and a textbook linear programme. Implemented from the papers; no third-party code was ported.
@incollection{shapley1953value, author={Shapley, Lloyd S.}, title={A Value for n-Person Games}, booktitle={Contributions to the Theory of Games II}, series={Annals of Mathematics Studies 28}, pages={307--317}, year={1953}, doi={10.1515/9781400881970-018}}
@article{schmeidler1969nucleolus, author={Schmeidler, David}, title={The Nucleolus of a Characteristic Function Game}, journal={SIAM Journal on Applied Mathematics}, volume={17}, number={6}, pages={1163--1170}, year={1969}, doi={10.1137/0117107}}
@article{maschler1979geometric, author={Maschler, Michael and Peleg, Bezalel and Shapley, Lloyd S.}, title={Geometric Properties of the Kernel, Nucleolus, and Related Solution Concepts}, journal={Mathematics of Operations Research}, volume={4}, number={4}, pages={303--338}, year={1979}, doi={10.1287/moor.4.4.303}}
@article{young1982cost, author={Young, H. Peyton and Okada, Norio and Hashimoto, Tsuyoshi}, title={Cost allocation in water resources development}, journal={Water Resources Research}, volume={18}, number={3}, pages={463--475}, year={1982}, doi={10.1029/WR018i003p00463}}
@article{littlechild1973simple, author={Littlechild, S. C. and Owen, G.}, title={A Simple Expression for the Shapley Value in a Special Case}, journal={Management Science}, volume={20}, number={3}, pages={370--372}, year={1973}, doi={10.1287/mnsc.20.3.370}}
@article{aumann1985game, author={Aumann, Robert J. and Maschler, Michael}, title={Game theoretic analysis of a bankruptcy problem from the Talmud}, journal={Journal of Economic Theory}, volume={36}, number={2}, pages={195--213}, year={1985}, doi={10.1016/0022-0531(85)90102-4}}
@article{oneill1982problem, author={O'Neill, Barry}, title={A problem of rights arbitration from the Talmud}, journal={Mathematical Social Sciences}, volume={2}, number={4}, pages={345--371}, year={1982}, doi={10.1016/0165-4896(82)90029-4}}
Limits: the coalition table takes up to 14 members, the nucleolus and full core check up to 10 (the road's core check uses its structure at any size), claims up to 14. A split says what is fair by one definition; it does not say what a contract, deed or court requires.