These appeals arise out of an action under the False Claims Act (FCA). 1 The relators prosecuted the action on their own after the government declined to intervene.

The relators’ theory of the case is as follows: General Dynamics (GD) contracted with the Air Force (AF) to produce, and provide engineering drawings for, an elaborate radar system called MUTES. MUTES is a mobile system used to train pilots by simulating the signals of various enemy radar. After a period of years, the AF solicited bids on a competitive basis from other companies who wanted to build MUTES. The AF awarded this follow-on contract to Aydin Corporation. The drawings GD had supplied to the AF proved inadequate for Aydin to produce MUTES according to its contract, and the AF was therefore forced to renegotiate the terms of its contract with Aydin at a cost of over $20 million.

The relators, who are insiders at Aydin, claim that GD knowingly provided inadequate drawings to the AF, and that this “fraud” actually damaged the United States in the amount the AF had to pay to alter the terms of the follow-on contract.

The district court held a bench trial to determine whether jurisdiction lay under the FCA. It found it had jurisdiction. The court then held a second bench trial on the merits of the FCA claim. After finding that GD had not knowingly submitted any false claims to the AF, the court entered judgment for GD. GD, as the prevailing party, then requested an award of costs. The court awarded it approximately twenty percent of its claimed costs. The relators appeal the judgment on the merits of their claim, and GD cross appeals the district court’s conclusion that it had jurisdiction under the FCA. The relators separately appeal the judgment for costs. 2

The district court had federal question jurisdiction under 28 U.S.C. § 1331 . This Court has jurisdiction over the appeal under 28 U.S.C. § 1291 , over the cross appeal under 28 U.S.C. § 1291 and Fed.R.App.P. 4(a)(3), and over the appeal from the costs judgment under 28 U.S.C. § 1291 .

I.

In the 1970s GD conceived of MUTES (Multiple Threat Emitter Systems). MUTES is a complex mobile system that simulates the threat of enemy forces. The AF uses it to train pilots. Based on GD’s idea, the AF conducted a competitive procurement for the award of a contract to build and produce engineering drawings for MUTES. GD won this contract, and it became effective in February 1978. This contract is generally referred to as “0103.”

The district court found that 0103 obligated GD to provide a technical data package including engineering drawings so that the AF could then use the drawings to procure spare parts. The contract further required that some of those drawings meet a set of requirements called “Level 3.” Level 3 requirements are described in a Department of Defense document entitled “MIL-D-1000A.” Although MIL-D-1000A does not specify precisely the kind of data to be included on Level 3 drawings, it does describe Level 3 drawings as those that would allow a competent manufacturer to reproduce the item based on the drawings.

In the following years, as a result of normal quality control audits, the parties evaluated the drawings and the MUTES system. As a result of these internal audits, GD made changes to its drawings. The AF, however, at all times considered GD’s drawings to be satisfactory under the 0103 contract.

In 1982, GD and the AF entered a second contract, referred to as “0366.” 0366 required GD to build more MUTES and to deliver new copies of the drawings that had been revised as a result of the internal audits. GD made approximately 6098 changes to its original drawings.

In late 1986 and early 1987, the AF initiated an effort to solicit proposals from other contractors to build more MUTES. Al *1406 though the AF intended to award the follow-on contract on a performance specification basis, 3 an AF contracting officer ultimately decided to solicit bids on a build-to-print basis. 4 As a part of this solicitation of bids, the AF warranted that certain of GD’s drawings — which would be provided to the winner of the follow-on contract — met Level 3 specifications. The AF did not warrant that the drawings were adequate for build-to-print contracts.

In 1987, the AF awarded the follow-on contract to Aydin on a build-to-print basis. Aydin had difficulty reproducing the MUTES systems, and was forced to seek permission from the AF to alter some of GD’s drawings. Moreover, many of the aperture cards the AF provided to Aydin, which contained engineering data, were illegible due to poor handling and copying by the AF. The AF and Aydin therefore contracted for a new set of these aperture cards (the Aperture Card Contract).

Ay din’s continuing difficulties led in 1989 to a renegotiation of the follow-on contract in which the AF agreed to pay an additional $22 million in exchange for Aydin’s agreement to continue to produce MUTES on a performance specification, rather than a build-to-print, basis. The relators filed them sealed complaint soon afterward in October, 1989.

The district court held trial in two phases. Phase I was a bench trial to determine whether the court had jurisdiction under the FCA. The court ruled that although the claim would have been barred under the version of the FCA in force prior to 1986 because the government had the knowledge forming the basis of the claim, the claim was saved by a 1986 amendment, which the court applied retrospectively. The court ruled that the new provision bars claims only when they are based upon certain public disclosures, and that these claims were not based upon any public disclosures within the meaning of the FCA. The court therefore asserted jurisdiction.

Phase II was a bench trial on the merits of the relators’ claims. The court found that the terms of both contracts, including the incorporation of the Level 3 specification, to be ambiguous, and therefore accepted extrinsic evidence regarding the understanding of the parties as to whether the drawings were expected to be error free or suitable for “build-to-print” follow-on contracts. It further found that the parties did not expect GD’s drawings to be error free or suitable for “build-to-print” follow-on contracts and, therefore, that none of the DD Form 250’s 5 or claims for payment constituted false or fraudulent claims. It also concluded that, even had GD submitted false claims, it did not have the scienter required in order for liability to attach.

After the entry of judgment, GD filed a bill of costs, which was eventually referred to a magistrate judge. The magistrate judge issued findings and recommendations, which the district court reviewed de novo and adopted. The district court then awarded GD $50,271.27 in costs.

II.

On cross-appeal, GD argues that the district court erred in applying a jurisdictional provision of the 1986 amendments, rather than applying the analogous provision from the preamendment act. It argues that under the preamendment act, the district court lacked jurisdiction over the action. Whether a provision of the 1986 amendments applies retrospectively to conduct that oc *1407 eurred prior to its enactment is a question of law reviewed de novo. United States ex rel. Anderson v. Northern Telecom, Inc., 52 F.3d 810, 814 (9th Cir.1995). Although this Court reviews the existence of jurisdiction de novo, it reviews the factual findings on jurisdictional issues only for clear error. United States v. Northrop Corp., 5 F.3d 407 , 409 n. 5 (9th Cir.1993), cert. denied, — U.S. -, 114 S.Ct. 1543 , 128 L.Ed.2d 195 (1994).

Prior to the 1986 amendments, the FCA barred actions brought by relators when the action was “based on evidence or information the Government had when the action was brought.” 31 U.S.C. § 3730 (b)(4) (1982); see also Anderson, 52 F.3d at 813 . The district court here found that the government knew of the evidence forming the basis of the action prior to 1986. Thus, the qui tam action would have been barred under the pre-1986 version of the FCA.

In 1986, however, Congress amended the FCA. Among the changes was a change to the jurisdictional provision. “The old language barring recovery if the government had ‘information’ before suit was filed was changed to an entirely new scheme, in which recovery was barred if the claim was based on ‘public disclosure,’ unless the relator was ‘an original source of the information.’ ” Anderson, 52 F.3d at 813 . This new provision states:

31 U.S.C. § 3730 (e)(4)(A) (1988). The district court ruled that this new provision applied to this case. Tr. of Oct. 1, 1992, at 19-20.

After the district court made its decision, however, the Supreme Court decided Landgraf v. USI Film Products, — U.S. -, 114 S.Ct. 1483 , 128 L.Ed.2d 229 (1994). This opinion signalled a significant shift in the manner in which federal courts should analyze questions involving the application of new civil statutes to conduct that has already occurred. Under Landgraf , a court must ask whether Congress has “expressly prescribed the statute’s proper reach.” Id. at-, 114 S.Ct. at 1505 . When Congress has not been so explicit, however,

Id.

This Court has clarified the terminology Landgraf employs by using the term “retrospective” to describe application of a new statute to events that occurred before its enactment, and reserving the term “retroactive” to describe a statute that, if applied, would attach new legal consequences to conduct or transactions already completed. See United States v. $814,254 76, 51 F.3d 207 , 210 n. 3 (9th Cir.1995). Thus, Landgraf teaches that courts should not apply “retroactive” statutes “retrospectively” absent clear congressional intent. Landgraf , — U.S. at -, 114 S.Ct. at 1505 .

This Court recently addressed whether to apply section 3730(e)(4)(A) retrospectively. Anderson, 52 F.3d at 814 , held that the provision does apply to a claim filed after the amendments by a relator who had revealed information to the government after the 1986 amendments became effective but before filing suit. Thus, because the government had the information when the relator filed suit, his claim would have been barred under the preamendment version of the FCA but not under the postamendment version.

The Anderson panel, however, expressly declined to decide whether the section was “retroactive.” Id. at 814 . Instead, it focused on the effect of the new statute on the relator, whose relevant conduct — disclosure of information to the government — occurred after the amendments. It reasoned that because the relator “acted after the amend *1408 ment, and the amendment changed the consequences only of [the relator’s] conduct,” id. at 814 , the case did not require the Court to decide whether to apply the amendment retrospectively: “[The relator’s] conduct took place after it, so benefitted from the new law.” Id.

The Anderson panel thereby avoided the key inquiry under Landgraf because application of the new law to the ease at hand did not constitute application to conduct that had occurred prior to 1986. In this case, by contrast, the government did have the information forming the basis of the claim prior to the enactment of the amendments, and there is no relevant postamendment conduct that might allow us to say our application of the new law is prospective only. We therefore cannot avoid deciding whether the new provision is “retroactive” simply by focusing on an event such as disclosure that occurred after the amendments. Instead, we directly face the question whether section 3730(e)(4)(A) should apply retrospectively. For the following reasons, we hold that the new jurisdictional provision does not have retroactive effect, and therefore that the district court properly applied it retrospectively.

The new jurisdictional provision does not impair any rights GD had when it acted, nor did the provision at issue increase GD’s liability for past conduct, or impose new duties. 6 This amendment relates to whether a qui tam relator can bring the action, not to whether the GD is liable for the underlying fraud. See Anderson, at 814 (“[The defendant] submitted the allegedly false claims before the 1986 amendment became effective. But the 1986 amendment did not change the legal consequences of [the defendant’s] conduct. If [the defendant] really did submit a fraudulent claim, it became liable and remained liable to the government....”). Thus, the amendment did not alter GD’s underlying liability; it only altered the conditions under which a qui tam relator can bring an action to enforce that liability.

GD argues that the amendment has retroactive effect because it altered GD’s rights by eliminating an “absolute defense.” We reject this argument. Proof that the government had the information when suit was brought was not an “absolute defense” under the preamendment FCA. Rather, it was simply a jurisdictional defense to an action brought by a qui tam relator, as opposed to an action brought directly by the government.

Additionally, the Landgraf Court, in dicta, noted that it has “regularly applied intervening statutes conferring or ousting jurisdiction, whether or not jurisdiction lay when the underlying conduct occurred or when the suit was filed.” — U.S. at-, 114 S.Ct. at 1501 . “Present law normally governs in such situations because jurisdictional statutes ‘speak to the power of the court rather than to the rights or obligations of the parties.’ ” Id. at-, 114 S.Ct. at 1502 (quoting Republic Nat’l Bank v. United States, — U.S. -, -, 113 S.Ct. 554, 565 , 121 L.Ed.2d 474 (1992) (Thomas, J., concurring)). The provision at issue here fits this paradigm precisely. Prior to the amendment, a court had no authority to hear an FCA qui tam claim if the government already had the information forming the basis of the claim; after the amendment, it did have authority to hear such a claim, provided that the claim was not based on a public disclosure. The amendment simply did not alter GD’s underlying obligation not to commit fraud upon the government. We therefore affirm the district court’s retrospective application of section 3730(e)(4)(A) to this case.

III.

GD argues that even if the 1986 jurisdictional provision applies to this case, the action is barred under this new provision because it is based on “information that was publicly disclosed in a series of government administrative audits and in an administra *1409 tive investigation of the condition of GD’s drawing package.”

The relevant amendment, again, reads as follows:

31 U.S.C. § 3730 (e)(4)(A) (1988). 7

GD argued to the district court that seven different documents constituted public disclosures that triggered the jurisdictional bar. The relators counter that none of these documents resulted from an audit conducted by an administrative agency, and therefore that they cannot qualify as public disclosures in an “administrative audit.” We hold that none of the alleged disclosures GD presented to the district court triggers the jurisdictional bar.

Although no published opinion of this Court expressly so holds, a fair reading of section 3730(e)(4)(A) indicates that a district court lacks jurisdiction when:

A survey of cases also suggests that each of these elements must be satisfied in order to trigger the bar. 8

Because each of the documents GD claimed trigger the bar fails at least one of the four requirements outlined above — -in addition to potentially failing part (3) because the “audits” were not conducted by an administrative agency — we need not decide whether, as the relators urge, an administrative audit can only be conducted by an administrative agency. Instead, we hold that each of the documents fails some other part of the test. 9

Exhibit 20:

This is an excerpt from the offering documents the AF provided to bidders for the “follow-on” contract for the production of *1410 MUTES. It states that where the drawings the AF provides are inadequate or obsolete, the contractor will need to update the design and specifications. This caveat does not constitute a disclosure of any “allegations or transactions” regarding any fraud on the part of GD. This exhibit therefore fails to meet requirement (2).

Exhibit D-189:

This is a letter from relator Lindenthal in his capacity as Vice President for Contracts for Aydin Corporation to an AF contracting officer. It discusses proposed price increases due to defective specifications. It makes no mention of GD specifically, nor does it allege that the inadequacies stem from any fraud. Moreover, this letter does not constitute a “public” disclosure. It is stamped:

“PROPRIETARY TO AYDIN CORP.”

Exhibit W-190:

These are the minutes of a quarterly report prepared by Aydin for the AF. They “reflect the agreements reached between US Government and contractor personnel” regarding difficulties Aydin was having producing MUTES. These minutes were not “publicly disclosed.” Moreover, disclosure of problems with the drawings is not tantamount to a disclosure that the producer of those drawings committed fraud in submitting them to the government. Cf. Rabushka, 40 F.3d 1509, 1512 (8th Cir.1994) (“the essential elements exposing the transaction as fraudulent must be publicly disclosed”).

Exhibit J-125; Exhibit A-127:

These are claimed to be factual predicates for two other documents. They are internal government documents regarding the approval or disapproval of certain data. They are not public and do not allege or reveal the factual predicates of fraud. They merely reveal inaccuracies in certain data GD provided to the AF.

Exhibit D-166:

This is a letter from the AF to Aydin requesting that Aydin submit a proposal for changing the contract from a “build-to-print” contract to a “performance” contract. This letter does not mention any shortcomings of any drawings, does not mention GD by name, does not suggest that any shortcomings were intentional, and appears to be a private, rather than public, correspondence between the AF and Aydin.

Exhibit F-166:

This is a letter from the AF to Aydin requesting Aydin to submit a proposed price for modifying the contract due to the inadequacies of the drawings. It states: “The government recognizes that the technical data package provided to your firm for performance under the subject contract is inadequate and contains errors.” This admission by the AF, however, does not by itself give rise to an inference of fraud on the part of GD. In fact, it does not even constitute an admission that GD’s drawings were inadequate when submitted to the AF, only that they were inadequate for Aydin’s purposes. Moreover, the letter is addressed and presumably was sent to Aydin. Such a letter cannot constitute a “public disclosure.”

We therefore hold that none of the alleged public disclosures triggered the jurisdictional bar, and affirm the district court’s conclusion that section 3730(e)(4)(A) does not bar this action.

IV.

Turning to the merits, the relators’ theory at trial was that GD breached its contract with the AF and provided drawings that did not meet Level 3 specifications, as evidenced by the fact that they were inadequate for the “build-to-print” contract awarded to Aydin. Thus, GD’s submission of DD Form 250’s and claims for payment for work described in those forms — if that work did not in fact meet Level 3 specifications — would constitute “false” claims. 10 Central to the merits of this case, therefore, is the question whether GD’s drawings satisfied the requirement of contracts “0103” and “0366” that the drawings meet the Level 3 specification.

*1411 A. The district court did not err in admitting extrinsic evidence to determine the parties’ expectations.

Contracts 0103 and 0366 both required GD to submit “Level 3” engineering drawings. Level 3 is defined by a Department of Defense document entitled “Military Specification: Drawings, Engineering and Associated Lists” and labelled “MIL-D-1000A.” Both contracts incorporate MIL-D-1000A by reference.

MIL-D-1000A states that “[engineering drawings and associated lists prepared to this level [Level 3] shall provide engineering definition sufficiently complete to enable a competent manufacturer to produce and maintain quality control of item(s) ... without additional product design effort, additional design data, or recourse to the original design activity.” It further establishes that the intended use for Level 3 drawings is “[t]o provide engineering data for support of quantity production to permit competitive procurement for items substantially identical to original drawings.”

Nowhere does MIL-D-1000A state that Level 3 drawings must be error free or adequate to support a “build-to-print” contract. Thus, GD posits, it is impossible to determine, simply by reference to the contracts and to MIL-D-1000A, whether the fact that GD’s drawings were inadequate for the follow-on contract means that they did not meet the requirements of contracts 0103 and 0366. The district court therefore found that the “contract terms regarding what information had to be contained in the drawings are ambiguous, and the intent of the parties must accordingly be determined with reference to extrinsic evidence.” 11

We agree that extrinsic evidence was properly admitted to flesh out the meaning of “Level 3.” “Where ... contractual language ... is unclear and suggests several speculative interpretations, the scope of the language must be read in accordance with the parties’ contemporaneous construction, and extrinsic evidence is admissible to show what the parties intended it to mean.” Lockheed Aircraft Corp. v. United States, 553 F.2d 69, 89 (Ct.Cl.1977) (citations omitted). Here, the relators claim that language in MIL-D-1000A stating that drawings meeting Level 3 specifications should allow a competent manufacturer to build the product without resort to additional engineering unambiguously means the same thing as drawings that are sufficient to support a “build-to-print” contract. 12

MIL-D-1000A, however, is not so precise. It also states, for instance, that the “numbers and types of engineering drawings selected is predicated on sound engineering judgment.” As an example, the specification states that a Level 3 drawing for a simple device might only require “a few detailed drawings” whereas Level 3 drawings for a sophisticated system could require “hundreds of individual drawings.” Id. Thus, we agree that MILD-1000A does not unambiguously define the parties’ expectations regarding the degree of expected accuracy or the suitability for future uses. It certainly does not unambiguously state that Level 3 drawings must be suitable for build-to-print follow-on contracts. We reject the plaintiffs’ argument that the district court erred in admitting extrinsic evidence regarding the parties’ expectations.

B. The district court did not err in concluding that GD’s drawings met the parties’ expectations and were therefore not the basis of any “false claims. ”

Based on testimony and documentary evidence, the district court found that the AF did not intend its contracts to require GD to provide drawings that could be used for follow-on contracts on a build-to-print basis. Rather, the AF “intended to procure from General Dynamics Level 3 drawings only for general Dynamics-built equipment.” The court also concluded that “[drawings adequate to satisfy the Level 3 requirements of General Dynamics’ MUTES *1412 contracts are not necessarily adequate to support a build-to-print follow-on contract with a different contractor,” and that “General Dynamics’ drawings met the Level 3 requirements of its MUTES contracts.” We review the district court’s findings based on the extrinsic evidence for clear error, and the principles of law applied to those facts de novo. Stephens v. City of Vista, 994 F.2d 650, 655 (9th Cir.1993).

None of these findings is clearly erroneous. One AF officer, for instance, testified in deposition that the government’s interest in purchasing the drawings was “to support repair and was to support the acquisition of spare parts,” and that “[i]t was generally recognized ... that Level 3 data rarely supported build-to-print.” Another officer, when asked whether data for competitive procurement meant data for a build-to-print contract responded: “No. Data for competitive procurement means enough data to sustain a true level of competition, which can mean a variety of levels of data.”

Similarly, evidence amply supports the conclusion that the drawings satisfied GD’s contractual obligations to the AF. For example, the AF Program Manager responsible for the MUTES program testified in deposition that “We’ve been using those drawings for ten or 12 years without any difficulty.... And I stand by the data. We’ve been using the drawings and continued to use those drawings. The United States government is very happy with the drawings, and we’re going to continue to be happy with them.”

We therefore affirm the district court’s findings that GD’s drawings met the AF’s expectations and satisfied GD’s contractual obligations. Logically, then, any claims for payment based on work that satisfied GD’s contractual obligations to the AF could not have been “false or fraudulent” within the meaning of the FCA. We also therefore affirm the district court’s conclusion that none of the DD 250 Forms or claims for payment under the MUTES contracts contained any false claim for payment or fraudulent statement.

V.

Because we conclude that GD did not submit any false claims, we need not reach the question whether it submitted false claims “knowingly” so as to allow liability to attach under the FCA. The FCA imposes liability on “[a]ny person who ... knowingly presents, or causes to be presented ... a false or fraudulent claim for payment.” 31 U.S.C. § 3729 (a)(1). Knowing submission of claims that, as here, are not false or fraudulent, obviously does not give rise to liability.

VI.

After entry of judgment, GD submitted a bill of costs requesting over $200,000 in costs. The clerk denied the bill of costs. GD moved the district court to review the denial of costs. The district court reviewed the denial, ordered costs to be awarded, and referred the matter to a magistrate judge. The magistrate judge disallowed a significant portion of GD’s bill of costs and recommended an award of $50,271.27. The district court reviewed the magistrate judge’s findings and recommendations, adopted them, and entered judgment awarding costs to GD.

The relators appeal from the district court’s order awarding costs. They argue that because the district court had previously declined to award GD attorney’s fees and expenses under authority of the FCA upon a finding that the relators’ claim was not “frivolous, clearly vexatious, or brought primarily for purposes of harassment,” it lacked authority to award costs under Rule 54(d). The relators also argue that, even if the district court did have authority to award costs, it abused its discretion in doing so. We address these arguments in turn. 13

A. The district court had authority to award costs under Rule 5Jf.

Rule 54 provides:

Fed.R.Civ.P. 54(d)(1). The relators contend that the court lacked authority under Rule 54(d) because 31 U.S.C. § 3730 constitutes an “express provision” regarding costs, which displaces the authority the court would otherwise have under Rule 54(d).

Although several subsections of section 3730 do expressly mention costs, the subsection governing the relators in this case does not. The relevant subsection states:

31 U.S.C. § 3730 (d)(4) (emphasis added). On its face, this subsection does not constitute an “express provision” regarding “costs”; the word “costs” is simply absent from this provision.

Relators, however, suggest that the term “expenses” includes “costs,” that this subsection thus is an express provision relating to costs, and that the FCA therefore only provides for costs to a defendant who prevails against a relator proceeding on his own behalf when the relator’s claim is clearly vexatious. Although the question is close, we reject this argument because under the FCA, fees, expenses, and costs are three distinct categories. “Costs,” under the FCA, is not a subset of “expenses.”

Section 3730(d)(2), for instance, establishes that when a relator proceeds on his own behalf and prevails, he shall receive “an amount for reasonable expenses which the court finds to have been necessarily incurred, plus reasonable attorneys’ fees and costs.” 31 U.S.C. § 3730 (d)(2) (emphasis added). Similarly, section 3730(g) provides that when the government intervenes but does not prevail, 28 U.S.C. § 2412 (d) shall apply. Section 2412(d), as does section 3730(d)(2), distinguishes between fees, expenses, and costs. See 28 U.S.C. § 2412 (d)(1)(A) (“a court shall award to a prevailing party other than the United States fees and other expenses, in addition to any costs”) (emphasis added). The language of these two sections strongly suggests that costs, under the FCA, constitute a category distinct from expenses: as a matter of logic, a court could not award costs in addition to expenses if costs were merely a subset of expenses.

In a nutshell, then, our analysis is as follows: two other subsections of section 3730 distinguish between expenses and costs and suggest that the two categories are discrete. Section 3730(d)(2) does so explicitly while section 3730(g) does so by reference to another statute. It would therefore be incongruous to conclude that section 3730(d)(4), which refers only to “fees” and “expenses,” constitutes an express provision regarding “costs,” despite that it is silent regarding “costs.”

The relators, however, invoke as support for their position several authorities that we find unpersuasive. First, they quote language from United States ex rel. Kelly v. Boeing Co., 9 F.3d 743 (9th Cir.1993), cert. denied, — U.S. -, 114 S.Ct. 1125 , 127 L.Ed.2d 433 (1994), which was a constitutional challenge to the qui tam provisions of the FCA. Some language from that opinion would support the position that “costs” are included in the term “expenses” and may only be awarded upon a finding of frivolousness or vexatiousness. See, e.g., 9 F.3d at 749 (“the qui tam plaintiff may be liable for costs if the suit is frivolous”); 9 F.3d at 752 (“a relator who pursues a frivolous action can be ordered to pay the defendant’s costs and attorneys’ fees”).

This language, however, is not a holding. The appellant in Kelly raised no issues regarding costs, expenses, or fees, and this Court addressed none. The relators’ argument relies merely on imprecise word choice in dicta. The language is simply not binding on this panel in this case.

*1414 Similarly, other courts addressing other questions have used language that might give rise to an inference that costs are a subset of expenses. See, e.g., Crawford, Fitting Co. v. J.T. Gibbons, Inc., 482 U.S. 437, 441-42 , 107 S.Ct. 2494, 2497 , 96 L.Ed.2d 385 (1986) (“Section 1920 enumerates expenses that a federal court may tax as a cost under the discretionary authority found in Rule 54(d).”); but see Hairline Creations, Inc. v. Kefalas, 664 F.2d 652, 655 (7th Cir.1981) (‘“costs’ are not equivalent to ‘expenses,’ but include only such items as court fees and witness fees”).

These cases are not directly relevant to our inquiry, however. The question in orn-ease is not whether, as a general matter, costs are included in expenses. Rather, the question is much more precise: under the FCA, does the category “expenses” include “costs.” Only if it does would section 3730(d)(4)’s provision for fees and expenses constitute an express provision for costs that would displace the discretion to award costs a court would otherwise have under Rule 54(d). Because we conclude that section 3730(d)(4) does not constitute an express provision regarding costs, we hold that the district court did not err in awarding costs under Rule 54.

B. The district court did not abuse its discretion in awarding costs.

The relators also argue that even if the district court had authority to award costs, it abused its discretion in doing so. This argument is without merit.

This court recently held:

National Info. Servs. v. TRW, Inc., 51 F.3d 1470 , 1472 (9th Cir.), amended, 52 F.3d 334 (1995). The relators list in their brief twelve reasons supporting their assertion that the district court abused its discretion. None of these reasons relates to any impropriety or misconduct by GD. Moreover, the district court, in a written order, considered many of the factors the relators raise on appeal, as well as caselaw guiding the decision whether to award costs, before ordering that GD should recover its costs. Finally, the magistrate judge, whose findings the district court adopted, carefully analyzed GD’s bill of costs and reduced the requested award by over seventy-five percent. We hold that the district court did not abuse its discretion in awarding costs under Rule 54(d).

VII.

We AFFIRM the district court’s assertion of jurisdiction (No. 93-16823) and AFFIRM the judgment entered in favor of General Dynamics (No. 93-16823). We also AFFIRM the award of costs (No. 94-16005). Each party shall bear its own costs on appeal.

. See 31 U.S.C. § 3729 et seq.

. These appeals have been consolidated.

. According to GD, a performance specification contract obligates the contractor to produce a system, for the contract price, that meets the performance requirements of the contract, regardless whether the drawings are perfectly accurate.

. A "build-to-print” contract requires the contractor to build the product precisely according to the technical data package supplied by the government — and requires the government to pay additional sums every time the contractor has to change a drawing because of an error or omission.

.When a contractor has finished a portion of work called for in the contract, it submits to the AF a "DD Form 250,” which describes the work completed and the amount the AF is to pay for that work. The AF approves the DD Form 250 and returns it to the contractor. The contractor then includes the cost reflected in the DD Form 250 in its next claim for payment.

. We emphasize that our analysis and holding applies only to § 3730(e)(4)(A). The analysis might be different for other provisions of the 1986 amendments, such as those that explicitly raise the penalties for submission of false claims. Compare 31 U.S.C. § 3729 (1982) (imposing penalty of $2000 plus two times the actual damages) with 31 U.S.C. § 3729 (a) (1988) (imposing penalty between $5000 and $10,000 plus three times the actual damages).

. Of course, even actions based on public disclosures within the meaning of this section are not barred if the relator is "an original source” of the information. See 31 U.S.C. § 3730 (e)(4)(A)-(B). In this appeal, however, the relators do not claim to be original sources, and the district court in fact found that they are not. Tr. of Oct. 1, 1992, at 19.

. See United States ex rel. Rabushka v. Crane Co., 40 F.3d 1509 (8th Cir.1994) (holding no bar because information disclosed did not amount to "allegation” of fraud or fraudulent "transaction"), cert. denied, -U.S. -, 115 S.Ct. 2579 , 132 L.Ed.2d 829 (1995); United States ex rel. Siller v. Becton Dickinson & Co., 21 F.3d 1339 (4th Cir.) (holding that "based upon” means "derived from”; discovery filed with a court constitutes “civil hearing"), cert. denied, - U.S. -, 115 S.Ct. 316 , 130 L.Ed.2d 278 (1994); Cooper v. Blue Cross & Blue Shield of Florida, Inc., 19 F.3d 562 (11th Cir.1994) (holding no bar because general rather than specific "allegations” were disclosed; "based on” means "based in any part on”); United States ex rel. Springfield Terminal Ry. Co. v. Quinn, 14 F.3d 645 (D.C.Cir.1994) (holding that only discovery materials that are actually, rather than just potentially, filed are "publicly" disclosed; "allegations or transaction” means an allegation of fraud or allegation of the critical elements of fraud); United States v. Northrop Corp., 5 F.3d 407 (9th Cir.1993) (holding that information revealed in a criminal indictment based on information provided by the qui tarn plaintiff is not “publicly disclosed”), cert. denied, - U.S. -, 114 S.Ct. 1543 , 128 L.Ed.2d 195 (1994); United States ex rel. Williams v. NEC Corp., 931 F.2d 1493 (11th Cir.1991) (holding that section 3730(e)(4)(A)’s list of methods of public disclosure is exclusive; administrative audit means issued by an administrative agency); United States ex rel. Hagood v. Sonoma County Water Agency, 929 F.2d 1416 (9th Cir.1991) (holding that participation in drafting a contract does not amount to an "administrative investigation”).

.The district court “found” that none of the documents met the requirements of § 3730(e)(4)(A). Although this Court generally reviews factual findings underlying jurisdictional questions for clear error, a decision regarding whether a particular document triggers the statutory bar seems best characterized as a mixed question of law and fact. We therefore review the district court’s conclusions de novo.

. The FCA provides that "[a]ny person who ... knowingly presents, or causes to be presented, to an officer or employee of the United States Government or a member of the Armed Forces of the United States a false or fraudulent claim for payment or approval ... is liable to the United States Government.” 31 U.S.C. § 3729 (a).

. The "extrinsic evidence” consisted of testimony from several witnesses from the AF and from GD and declarations and/or deposition testimony from over 30 declarants/witnesses.

. We note that the relators, perhaps unintentionally, resort to extrinsic evidence to support the contention that the terms are not ambiguous.

. We generally review a decision to award costs for an abuse of discretion. National Info. Servs. Inc. v. TRW, Inc., 51 F.3d 1470 , 1471 (9th Cir.1995), amended, 52 F.3d 334 (1995). However, whether a district court has the authority to award costs after denying fees and expenses under the FCA is a question of law we review de novo.