Tomkenco, Inc., borrowed money from First Wisconsin Financial Corp. against the security of Tomkenco's assets and receivables. The contract between Tomkenco and First Wisconsin created a "Loan Account Ledger". When Tomkenco presented First Wisconsin with evidence that it had sold goods and created receivables, First Wisconsin would advance 80% of the amount of the receivables. The ledger would be updated with a specific entry, identified by Tomkenco's customer. The loans would be summed, and each month Tomkenco would receive a statement of the debit balance. The whole balance was due on First Wisconsin's demand. Tomkenco sent the customers' checks directly to First Wisconsin, which credited the particular item in the ledger and applied any excess to remaining debt not reflecting customers' paper (for example, debt secured by Tomkenco's inventory and equipment). Thomas Yamaguchi and Kenneth Kuzmenko, the principal stockholders and managers of Tomkenco, personally guaranteed Tomkenco's debts to First Wisconsin.

First Wisconsin started lending money to Tomkenco in May 1979. During the fall of 1980 Kuzmenko began to fabricate documents showing that Tomkenco had sold goods and received customers' paper. First Wisconsin loaned money against this worthless paper. Although First Wisconsin sometimes checked with customers to learn whether the purported sales had occurred, Kuzmenko enlisted Edward Gavney, an employee of UNA Corp., to "verify" fictitious sales to Bohler Brothers, a division of UNA. First Wisconsin did not catch on until the end of June 1981, by which time Tomkenco had obtained about $213,000 in loans against the bogus documents. Tomkenco was liquidated, more than $250,000 in debt to First Wisconsin, which tried to collect on its guaranties. By then Kuzmenko had vanished, leaving Yamaguchi as First Wisconsin's logical target.

Yamaguchi, however, was no longer an officer of Tomkenco. He quit on January 31, 1981, and sold all of his stock to Kuzmenko. He informed First Wisconsin of the change and requested a "release" of any accrued liability on this guaranty. First Wisconsin declined to "release" Yamaguchi unless someone invested $50,000 more in Tomkenco, an additional "equity cushion" that would improve First Wisconsin's position. Neither Kuzmenko nor Yamaguchi produced the money. On April 2, 1981, Yamaguchi's lawyer sent this letter to First Wisconsin:

As you know Mr. Yamaguchi has resigned as an officer and director of Tomkenco, Inc., effective on January 31, 1981. We were under the impression that First Wisconsin was intending to release Mr. Yamaguchi from his guarantee upon delivery to you of certain documentation from Mr. Yamaguchi. I understand that now you are not able to do that. However, on behalf of Mr. Yamaguchi, I am requesting that you release Mr. Yamaguchi from all liabilities incurred by Tomkenco on or after January 31, 1981, the effective date of his resignation. This is, in effect, a revocation of the guaranty effective on January 31, 1981. Could you please acknowledge receipt of this letter by giving us a written acknowledgement of Mr. Yamaguchi's release.

Despite a reminder from Yamaguchi's lawyer, First Wisconsin never replied in writing to this letter or asked for clarification. First Wisconsin concedes, however, that it received the letter.

After Kuzmenko's fraud had been exposed, First Wisconsin filed this diversity action against Kuzmenko, Yamaguchi, and UNA, seeking to make good its loss. The parties agree that Wisconsin law applies. The facts concerning Yamaguchi are undisputed, and both parties moved for summary judgment, which Judge Aspen granted in 1983. Judge Aspen concluded that the letter of April 2 revoked Yamaguchi's guaranty, but as of April 2 rather than January 31. On April 2 Tomkenco owed First Wisconsin about $404,000. After that date, Tomkenco paid First Wisconsin some $554,000, and borrowed $407,000. First Wisconsin argued that these payments should be applied to the particular items of customer paper against which money had been lent. Under this approach Yamaguchi was liable for $404,000 less payments attributable to paper written before then. Judge Aspen disagreed, concluding that although First Wisconsin had lent against security, the parties' arrangement was a single open-account loan. Payments and new advances on a single loan usually are accounted for under the first-in, first-out (FIFO) method. The application of FIFO accounting extinguished Yamaguchi's debt. Tomkenco paid about $554,000 after April 2, more than enough to cover the balance of $404,000 on that date. The court therefore entered summary judgment for Yamaguchi. Other parties remained in the case, however, and Judge Aspen declined to enter a partial final judgment under Fed.R.Civ.P. 54(b).

First Wisconsin eventually caught up with Kuzmenko, and the case was set for trial in May 1986 before Judge Leinenweber, to whom the case had been transferred. The parties settled on the eve of trial. First Wisconsin then tried to amend its complaint to add a claim of fraud against Yamaguchi, contending that not until the deposition of Kuzmenko in November 1985 did First Wisconsin have reason to believe that Yamaguchi had participated in the deceit. Judge Leinenweber declined to allow the amendment, and at last the court entered a judgment "final" and appealable under 28 U.S.C. Sec. 1291. First Wisconsin contests the denial of leave to amend, but this was within the district judge's discretion. See Bohen v. City of East Chicago, 799 F.2d 1180, 1184-85 (7th Cir.1986). By its own account, First Wisconsin did not try to amend the complaint for six months after learning of Yamaguchi's complicity. First Wisconsin explains that it waited because it did not want to risk a postponement of the trial scheduled for May 1986. This damns rather than justifies the delay, however; First Wisconsin concedes that it tried to split this case and, for its own convenience, obtain two trials. District courts are entitled to protect themselves, and other litigants whose cases would be affected, against such strategic conduct.

This leaves two questions of substance: the effect of the letter of April 2, and the choice of the accounting method. We agree with the district court that the letter of April 2 revoked the guaranty but do not believe that FIFO accounting is appropriate in a case of this character.

After Yamaguchi left Tomkenco, he was exposed to liability based on the debt outstanding on January 31. He promptly opened negotiations for relief. These foundered, and the letter of April 2 deals with liability for future transactions. It asks for "release" from liability for all transactions "after" January 31 and continues: "This is, in effect, a revocation of the guaranty effective on January 31, 1981." First Wisconsin contends that it interpreted this letter as another effort to negotiate a release from the liability accrued as of January 31, but the subjective interpretations of one party to a communication do not detract from the force it objectively carries. The letter acknowledged that First Wisconsin was not going to let Yamaguchi avoid any accrued liability; Yamaguchi then asked to be let out of any new liability effective January 31. To achieve this goal Yamaguchi needed two things: a "release" of the exposure created between January 31 and April 2, and no liability thereafter--in other words, a revocation of the guaranty. The letter used both words.

Under Wisconsin law any clearly communicated revocation of a guaranty is effective. Bremer v. Rufener, 186 Wis. 195, 202 N.W. 206 (1925); Klatte v. Franklin State Bank, 211 Wis. 613, 618, 248 N.W. 158, 160 (1933). Clarity is important to commercial transactions. The beneficiary of a guaranty relies on the undertaking in supplying credit. Unless the lender can determine the extent of its protection through objective criteria, the risk of doing business will go up, and with it the rate of interest charged for loans. Uncertainty thus injures honest, reliable debtors. Wisconsin's law honors only "definite and unequivocal notice" of revocations of guaranties. Bremer, 186 Wis. at 198, 202 N.W. at 207. Just as it is undesirable to expose lenders to needless risk of loss by allowing revocations that do not alert lenders to the change of liability, it is also undesirable to allow foibles of expression to expose guarantors to continuous liability. Guarantors need certainty, too, to plan affairs or decide what to charge for the guaranty, cf. United States Shoe Corp. v. Hackett, 793 F.2d 161, 162 (7th Cir.1986). It will be harder or more costly to obtain guaranties (and therefore more difficult to launch new businesses) if every guaranty carries the risk that a jury subsequently will decline to let the guarantor off. Perfection is too much to demand in any notice, just as opacity offers too little warning to support revocation.

There are three possible sources of ambiguity in the letter. First, the letter was sent while the parties were negotiating for release of the liabilities predating January 31; second, the letter seeks to "revoke" effective two months before it was sent and therefore combines elements of revocation and release; third, the critical sentence contains the modifier "in effect". None of these, however, so clouds the picture that summary judgment is inappropriate. The existence of other ongoing negotiations could not have obscured (in the mind of an objectively reasonable reader) the significance of this letter; Yamaguchi explicitly put pre-January 31 obligations to one side and addressed subsequent liabilities. First Wisconsin now says that it is sure that Yamaguchi did not want to revoke his guaranty, because had First Wisconsin thought Yamaguchi was doing any such thing it would have stopped extending credit to Tomkenco, bringing the fraud to light at once. As we must take the case, however, Yamaguchi had no knowledge of the fraud and no interest in Tomkenco, so he would not have cared whether First Wisconsin made further loans. The existing customer paper appeared to cover Tomkenco's debts. If First Wisconsin had any doubt whether Yamaguchi was pulling out, it had only to ask; yet First Wisconsin did not ask for clarification. The letter was designed to its purpose. It requested a release, a remedy not available unilaterally, and declared a revocation, a result within Yamaguchi's sole control. The muddle created by the statement that the revocation was "in effect" created as of January 31 does not obscure Yamaguchi's desire to be liable for no new debts of Tomkenco. No modestly careful reader of this letter could fail to appreciate that Yamaguchi was not going to accept liability for Tomkenco's future debts.

On to the accounting method. Wisconsin follows the method for attribution of payments described in Restatement of Security Sec. 142 (1941) and Restatement (Second) of Contracts Secs. 259-60 (1981). See Moser Paper Co. v. North Shore Publishing Co., 83 Wis. 2d 852, 857-58, 266 N.W.2d 411, 415 (1978); Crown Life Insurance Co. v. LaBonte, 111 Wis. 2d 26, 330 N.W.2d 201 (1983). Under this approach a payment is applied as the debtor directs; if the debtor does not direct, it is applied as the creditor chooses, provided the creditor promptly informs the debtor; if neither debtor nor creditor communicates a choice, the court selects an equitable method of application. Both the Restatement (Second) of Contracts Sec. 260(2) and the Wisconsin cases express a preference for the FIFO method when the court must choose. See State v. J.C. Penney Co., 48 Wis. 2d 125, 147, 179 N.W.2d 641, 652 (1970); Marshall & Ilsley Bank v. Hackett, Hoff & Thiermann, 213 Wis. 426, 431, 250 N.W. 866, 868 (1933); Hannan v. Engelmann, 49 Wis. 278, 5 N.W. 791 (1880). All of this is qualified in the Restatement (Second) of Contracts Sec. 260(1) by a concern that the method recognize the interests of third parties--often guarantors, who may furnish funds for the payment of specific accounts. Funds furnished by or on behalf of guarantors therefore must be traced and applied to extinguish the guarantied items, Sec. 260(2)(a); Sorge Ice Cream & Dairy Co. v. Wahlgren, 28 Wis. 2d 220, 137 N.W.2d 118 (1965), although this rule is of uncertain scope in Wisconsin in light of Crown Life, 111 Wis. 2d at 35-36 n. 7, 330 N.W.2d at 205-06 n. 7.

The district court concluded that there was a single open-account debt, that neither Tomkenco nor First Wisconsin had expressed to the other a method of application, and that FIFO was therefore the appropriate method. Yet both the Restatements and the Wisconsin cases treat FIFO accounting as the presumptive or usual method, not the universal method. In selecting a plan of attribution, the court must consider the nature of the parties' dealings and the effect of the accounting method on the achievement of the parties' objectives.

The purpose of Yamaguchi's guaranty was to provide First Wisconsin a source of payment in the event Tomkenco sold to unreliable customers or committed fraud. First Wisconsin expected to be repaid out of payments from Tomkenco's customers. But if Tomkenco had lax credit standards, or forged customer paper, there would be a risk of shortfall, and First Wisconsin would not have the security for which it had bargained. That's where the guaranty came in. The guaranty induced Yamaguchi to take care in the operation of Tomkenco and to monitor Kuzmenko as well; in the event Yamaguchi failed to take enough care or deadbeats failed to pay, Yamaguchi would be liable. When Yamaguchi rescinded the guaranty on April 2, Tomkenco had $404,000 in outstanding debt. To the extent the customers paid their debts, Yamaguchi would be excused. But if there were problems, Yamaguchi would make up the excess.

Had Tomkenco dissolved on April 2, the legitimate customers would have paid up, and the bogus paper would have been Yamaguchi's responsibility. The shortfall that would have been revealed by an immediate liquidation or sale of the business was Yamaguchi's contingent liability, the amount of First Wisconsin's risk Yamaguchi was covering. Suppose the $404,000 debt on April 2 represented $100,000 in spurious paper and $304,000 in honest receivables. Then Yamaguchi was on the hook for $100,000 and should be liable unless Tomkenco found some other way to cover its debts. Whatever happened later was none of Yamaguchi's business. The accounting rule should produce a result as close as possible to the result of a hypothetical liquidation or sale of the business to a fully informed buyer on April 2, for only such an approximation holds Yamaguchi to the amount of risk he assumed.

The FIFO approach does poorly by this standard. Suppose Tomkenco wrote no more bogus paper, sold $600,000 worth of goods, borrowed $480,000 against the receivables, and repaid the whole $480,000 (using the other funds for the operation of its business). Under FIFO accounting the whole debt of April 2 has been retired and Yamaguchi is discharged. Yet the risk to which First Wisconsin is exposed--that it will never recover the $100,000 lent against the security of the pretended receivables--is exactly the same after these transactions as it was on April 2. There is accordingly no reason to excuse Yamaguchi. To put this differently, the application of the FIFO method will discharge guarantors as time goes by and new loans revolve, even though risk is unchanged--perhaps even, as in this case, the rate of presentation of spurious paper increases. Allowing the normal (or fraudulent) operation of the business to discharge the guarantors would give people a reason to hide their fraud as long as possible, just the opposite of the appropriate incentive. Until First Wisconsin has been relieved of the risks that came into being on Yamaguchi's watch, Yamaguchi must remain responsible.

The problems with FIFO accounting do not suggest the use of LIFO (last-in, first-out). The LIFO method would prevent the discharge of a guarantor even though all of the risks created during his association with the business had been eliminated--perhaps because all of the accounts due on April 2 had been paid, and Kuzmenko or Yamaguchi had covered the bogus paper out of his own pocket. Once the affairs through April 2 have been wrapped up, the guaranty must be released; LIFO accounting does not achieve this.

Two rules leave Yamaguchi bearing approximately the risk that was in place on April 2. One is the "lowest intermediate balance" approach frequently applied to determine interests in commingled funds. See Restatement (Second) of Trusts Sec. 202 & comment j (1959); Restatement of Restitution Sec. 212 (1937). When credits and debits are made to a fund (here an open loan account), the interest of the first contributor (here the guarantor of the first loans) is the lowest balance the account had at any time. If $100 is deposited in an account, and later transactions occur, how much of the "original" $100 is left? If subsequent transactions are $200 in, $150 out, and $100 in, the answer is $100, for that amount was always there; but if subsequent transactions are $100 in, $150 out, and $100 in (actual balance $150), the answer is only $50, for that was the lowest balance at any time. This method, which has been applied to problems as abstruse as the tracing of proceeds from drug transactions, see United States v. Banco Cafetero Panama, 797 F.2d 1154, 1158-60 (2d Cir.1986), seems a natural here. On April 2 the debit balance was $404,000. If Tomkenco later reduced this to $50,000 and then borrowed another $500,000, Yamaguchi would be answerable for $50,000 at the most, for that is the maximum risk First Wisconsin retains from the transactions before April 2. If Tomkenco borrowed the $500,000 first and then repaid that money, however, Yamaguchi would remain liable for as much as $404,000, the lowest balance at any instant; Tomkenco could borrow and repay $500,000 without reducing by a dime First Wisconsin's risk, and therefore Yamaguchi's exposure would not be reduced, either.

The remaining method is a matching of payments against the paper that secured the debt. As the district court emphasized, there is only one "account", and the balance was reconciled at the end of each month. But it is also undisputed that First Wisconsin kept a detailed ledger in which advances were matched against security (receivables, inventory, or equipment). First Wisconsin cancelled particular items in this ledger as customers paid Tomkenco; the account stated at the end of the month was derived by summing the open items in the ledger. Customers' checks were delivered directly to First Wisconsin. The use of the ledger enabled First Wisconsin to identify non-performing debts, which helped it determine whether it was prudent to lend as much as 80% of the face value of the paper; the receipt of the original checks helped First Wisconsin detect fraud on Tomkenco's behalf and reduced the moral hazard that Tomkenco might hold onto money received from customers. Both the keeping of the ledger and the receipt of customers' checks could be construed as an explicit method of applying the proceeds. An agreement to have one open-account debt, the balance to be stated monthly, is not inconsistent with a more detailed method of posting payments to the account. If First Wisconsin communicated this method to Tomkenco, a subject on which the district court was silent, then this method prevails; a court is not authorized by Wisconsin law to supply an attribution mechanism that conflicts with the parties' own.

Although the facts about how First Wisconsin kept its ledger are undisputed, Tomkenco's knowledge of this method may be an open question. Inferences about the extent to which the parties have specified their own method also are open. Disputes about material facts and inferences should not be resolved on summary judgment. See United States Shoe, 793 F.2d at 166. It may be that the parties can supply enough evidence to permit the district judge to make a confident accounting choice without holding a trial. If First Wisconsin selected and communicated an item-by-item posting method, that should be used. Yamaguchi's maximum liability then would be the amount shown on First Wisconsin's ledger as unsatisfied advances before April 2, 1981. If First Wisconsin did not select or communicate an item-by-item method, the district judge must supply an appropriate method--either the lowest intermediate balance method or the item-by-item method, whichever more accurately reflects the risk to which First Wisconsin was exposed on April 2, 1981, that remains outstanding.

VACATED AND REMANDED.