The road is worth building, the report says, because those who gain from it gain enough to repay those who lose. The road is built. The payment is not. The report remains on the shelf, its arithmetic undisturbed, while the household beside the traffic lives with the difference between a payment that was possible and a payment that arrived. This is the compensation principle's central difficulty, and also its central distinction. A project can open the possibility of an improvement for everyone without delivering that improvement to everyone.
Kaldor's proposal in 1939 addressed a real impasse. If one person gains and another loses, their separate rankings cannot alone establish that the gain is worth more than the loss. His test asks a different question: could the beneficiaries compensate the people harmed and still retain an advantage? That question does not require us to declare one person's pleasure greater than another's pain. It asks whether a feasible arrangement exists in which each person, judging for himself, is at least as well off as before and someone is better off. The possibility matters. It distinguishes a change that could be made beneficial to all from a simple transfer of advantage that cannot. But the test describes the change together with a possible rearrangement of its gains. Remove the rearrangement, and the conclusion about everyone no longer follows.
Take an invented case whose numbers are deliberately simple. A change gives its beneficiary a gain equivalent to $120 and imposes a loss equivalent to $80 on someone else. Assume these are complete money equivalents at the relevant starting position, stable over the transfers considered, with no omitted effects and initially no cost of making the payment. The sum is positive forty. A payment of ninety dollars from the winner to the loser leaves the first thirty dollars ahead and the second ten ahead. Without that payment, the winner is a hundred and twenty ahead and the loser eighty behind. The positive sum is identical; the two actual arrangements are not. One passes the Pareto comparison with the original position. The other does not, because somebody has been made worse off.
Hicks's contribution is useful precisely because it does not let the analyst stop at this easy arithmetic. In closing his 1939 account, he insists on examining the effects of the reform and of the compensation together. Transfers may require administration, taxation, or other changes with consequences for production. Put a real cost of twenty-five dollars into our invented case, borne by the beneficiary in addition to the payment. The same ninety-dollar transfer now leaves that person five ahead and the other ten ahead: fifteen remains. Raise the real cost to forty-five, and the original forty-dollar opportunity is exhausted. The beneficiary could give up at most seventy-five while retaining the old position, but the loser needs eighty. A calculation that approved the original proposal has not thereby approved this more costly package.
Before even that calculation, someone must identify the loss. Replacing a paycheck need not replace a job's hours, its location, its companions, or the freedom a person has lost with it. Kaldor himself made room for such consequences; they are not objections that his arithmetic has already dismissed. If eighty dollars restores income but leaves the affected person worse off on the terms the test professes to respect, eighty was the wrong compensation figure. If no available payment can restore the relevant position, the asserted possibility has not been shown. If the proposal includes risks, the timing and reliability of the payment belong among the things being evaluated. A promise whose value depends on its being kept cannot be appraised as though its fulfillment were already a fact.
The word could therefore demands more than a large total printed beside a small one. The means of transfer must exist; the people to be compensated must be within its reach; the compensation must answer the harm actually counted. A proposal can be useful despite uncertainty on these matters, but its uncertainty remains part of its description. The technical achievement is to identify a possible improvement. The political achievement would be to select and carry out the arrangement that realizes it. When a beneficiary invokes the first achievement to excuse declining the second, he has exchanged one argument for another. He is no longer saying that everyone can be left unharmed by the package he offers. He is saying that the gain he keeps is a sufficient reason for the harm another bears.
There is another difficulty, preserved in the earlier treatment of The Standard. Scitovsky showed that a one-way compensation test can approve a change and then approve its reversal. The opportunities for redistributing welfare can cross: the new arrangement can contain a distribution that would improve on the old actual position, while the old arrangement contains a different distribution that would improve on the new actual position. The two hypothetical payments are being imagined from different starting points. A favorable comparison in one direction does not establish that the new arrangement offers a superior set of opportunities throughout. Scitovsky's double test asks both whether the new situation could improve on the old and whether redistribution within the old could improve on the new. The technical repair limits the recommendation. It still does not put a payment in anyone's hand.
Does this make compensation the only defensible basis for reform? No. The starting position may itself be unjust. A policy that ends an exclusion or removes a privilege need not promise the former beneficiary an equivalent privilege elsewhere before its merits can be argued. To insist on such a rule would give the status quo a protection that arithmetic cannot confer. But this is precisely why the account must name the reason being used. A reform may be defended because it corrects an injustice, because its benefits justify its burdens under an openly stated standard, or because its actual package makes everyone better off. Those are different arguments. The existence of the first two does not make the third true when its losers remain unpaid.
Nor does a positive net benefit prove that the beneficiary is the party who should bear the cost of redress, that a money payment is the best remedy, or that a particular arrangement of rights is just. Those questions require further premises. The compensation test's narrower contribution is nevertheless substantial. It can show that a dispute is not forced upon us by a shortage of gain: the resources may exist to leave the harmed person whole while retaining a benefit elsewhere. It can expose the difference between an unavoidable sacrifice and a refusal to share. Keeping that distinction visible strengthens the accounting. It makes the party defending nonpayment explain why the possible repair should be withheld, instead of letting the surplus total answer a question it was never designed to decide.
The possibility of repaying a loss is an opportunity to choose a better settlement. It is not the settlement itself.
A change that could repay its losers but does not is a potential improvement of a particular kind, not an accomplished improvement for all. Whether it is justified despite its losers is a further question, answerable only by bringing distribution and the reasons for the change into view. The ledger can help us find the terms on which a better bargain could be struck. To call the terms agreed when no payment has been offered is to mistake the room for an agreement for the agreement itself. The next plate follows the loss that may not even have reached the ledger — the value with no market price beside its name.