ResQNet. Com, Inc. v. Lansa, Inc.

594 F.3d 860 (2010) RESQNET.COM, INC., Plaintiff-Appellant, and Kaplan & Gilman, LLP and Jeffrey I. Kaplan, Esq., Sanctioned Parties-Appellants, v. LANSA, INC., Defendant-Cross Appellant. Nos. 2008-1365, 2008-1366, 2009-1030. United States Court of Appeals, Federal Circuit. February 5, 2010. *862 Jeffrey I. Kaplan, Kaplan Gilman & Pergament, LLP, of Woodbridge, NJ, argued for plaintiff-appellant and sanctioned parties-appellants. With him on the brief was Michael R. Gilman. James H. Hulme, Arent Fox, LLP, of Washington, DC, argued for defendant-cross appellant. With him on the brief was Janine A. Carlan. Of counsel was Leo M. Loughlin. Before NEWMAN, LOURIE, and RADER, Circuit Judges. *863 PER CURIAM. The United States District Court for the Southern District of New York ruled that U.S. Patent No. 6,295,075 (the '075 patent), owned by ResQNet.com, Inc., is valid and is infringed by Lansa, Inc. The district court also ruled that ResQNet's U.S. Patent No. 5,831,608 (the '608 patent) is not infringed. The court awarded damages of $506,305 for past infringement based on a hypothetical royalty of 12.5%, plus prejudgment interest. The court denied ResQNet's motion for a permanent injunction, and imposed a license, at a royalty of 12.5%, for future activity covered by the '075 patent. The court assessed sanctions under Rule 11 against ResQNet and its counsel. We affirm the district court's rulings on the issues of validity and infringement, and reverse the imposition of sanctions. On Lansa's cross-appeal, we vacate the damages award and remand for redetermination of damages. BACKGROUND This litigation began in 2001. The district court issued a claim construction order, ResQNet.com, Inc. v. Lansa, Inc., No. 01 Civ. 3578(RWS), 2002 WL 31002811 (S.D.N.Y. Sept. 5, 2002). Thereafter, on November 4, 2002 a Consent Judgment was entered to enable appeal of the claim construction to the Federal Circuit, and also dismissing two of the five patents in suit. J.A. 265-68. On appeal, this court modified the district court's claim construction. ResQNet.com, Inc. v. Lansa, Inc., 346 F.3d 1374 (Fed.Cir.2003) (ResQNet I). On Lansa's motion filed in September 2004, Rule 11 sanctions were imposed. ResQNet.com, Inc. v. Lansa, Inc., 382 F.Supp.2d 424, 457 (S.D.N.Y. 2005) (ResQNet II). The appeal of the sanctions order was dismissed by this court, on the ground that the appeal was not ripe because the merits of the underlying litigation had not yet been decided. ResQNet.com, Inc. v. Lansa, Inc., 138 Fed. Appx. 312 (Fed.Cir.2005). A bench trial was conducted, with judgment reported at ResQNet.com, Inc. v. Lansa, Inc., 533 F.Supp.2d 397 (S.D.N.Y.2008) (ResQNet III), corrected on reconsideration, J.A. 56-61 (Mar. 17, 2008). The district court declined to withdraw the sanctions. ResQNet.com, Inc. v. Lansa, Inc., No. 01 Civ. 3578(RWS), 2008 WL 4376367 (Sept. 25, 2008) (ResQNet IV). This appeal and cross-appeal followed, with ResQNet's attorneys joining in the appeal of the sanctions order. The Technology The technology relates to screen recognition and terminal emulation processes that download a screen of information from a remote mainframe computer onto a local personal computer (PC). Before the use of PCs, each computer user would connect to the mainframe using a so-called "dumb terminal," which displayed information received from the mainframe and sent all data entries back to the mainframe for processing. Because a dumb terminal's monitor usually was a monochromatic green color, the display was called a "green screen." PCs came to replace dumb terminals, with the PC using software to facilitate communication to and from the mainframe, and processing the information into a graphical user interface (GUI) format. The GUI format displays and receives information to and from the user, and sends and receives information in the manner understood by the mainframe. The ResQNet patents facilitate recognition of the information that the mainframe sends to the PC. The technology is more fully described in ResQNet I, 346 F.3d at 1375-76. The accused product is a terminal emulator program called "NewLook," developed by Looksoftware Proprietary Limited *864 in Australia, and sold by Lansa in the United States. As described more fully in ResQNet III, 533 F.Supp.2d at 406-07, the NewLook product creates a GUI using a dynamic architecture whereby the software automatically converts green screens into GUI screens without using a table lookup or otherwise recognizing the actual screen being displayed. The NewLook program has two "editions." The standard edition of NewLook is designed for use on a personal computer, and is typically used by a user running a special application. The developer or professional edition of NewLook is used by a programmer or developer to create overrides to change how elements are displayed. ResQNet charged Lansa with infringement of five patents: the '608 patent, the '075 patent, U.S. Patent No. 5,812,127 (the '127 patent), U.S. Patent No. 5,792,659 (the '659 patent), and U.S. Patent No. 5,530,961 (the '961 patent). The '127 and '659 patents were removed from the complaint, with prejudice, by the Consent Judgment filed on November 4, 2002. The '127 and '608 patents are relevant to the Rule 11 sanctions. Substantive issues concerning the '608 and '075 patents are presented in this appeal. I The '608 Patent The '608 patent is entitled "User Interface for a Remote Terminal." Claim 1, the only claim, is as follows: 1. Apparatus for implementing a computer terminal to be connected to a remote computer, said apparatus comprising: means for identifying a particular user logged on to said remote computer through said computer terminal; means for identifying, based upon a position, length and type of each of a plurality of fields, a particular screen to be displayed to said user; and a plurality of special function keys, each key performing a specified function, the specified function performed by each key being determined by the particular user logged on and the particular screen identified to be displayed. The claim was construed in ResQNet I, 346 F.3d at 1382-83. On remand, the district court ruled that the '608 patent is not infringed, finding that the NewLook products do not meet the first and third of the three clauses of claim 1. With respect to the first clause, the "means for identifying a particular user," the district court found that the '608 patent is directed to the situation whereby each user can assign its own values to the function key depending on the user's needs, whereas NewLook allows only for key customization for an entire group of users. For example, NewLook allows a programmer to change the function of the F1 button for a specific screen, but once the function is changed, every user of that screen will have the same function assigned to its F1 button. With respect to the third clause, the "special function keys" set by each user for each screen, the district court found that NewLook does not meet this limitation, even if the activities of the users are considered to supplement the activities of the NewLook product under theories of induced or contributory or split infringement. ResQNet argues that NewLook uses the same means, for identifying users and determining key function, as are shown in the '608 patent. ResQNet argues that the district court's interpretation of the '608 patent claim to exclude the NewLook technology would also exclude the preferred embodiment set forth in the '608 specification. ResQNet states that NewLook allows a developer or an administrator *865 to re-assign functions of various keys, whereby NewLook allows for different key functions depending on the user. ResQNet states that even when the NewLook programmer does not perform the customization of special function keys, this is performed by the user. Thus ResQNet argues that the district court misconstrued and misapplied the limitations of claim 1. Lansa responds that in NewLook the function of each key is the same for all users and varies only depending on the screen, and thus that the third claim element is not performed by any user. Clear error has not been shown in the district court's finding that the NewLook system does not remap keys for specific screens. On this finding, the NewLook technology does not infringe claim 1 of the '608 patent. II The '075 Patent Claim 1 of the '075 patent is as follows: 1. The method of communicating between a host computer and a remote terminal over a data network comprising steps of: establishing a first communication session between said terminal and a communications server via a first communications channel; downloading, from said server to said terminal, communications software for communicating between said terminal and said host and a plurality of specific screen identifying information; utilizing said communications software to implement a second communications session between said terminal and said host via a second communications channel independent of said server; receiving a screen from said host to said terminal; if said received screen matches one of the plurality of specific screen identifying information, displaying a customized GUI screen; and if said received screen does not match one of the plurality of specific screen identifying information, displaying a default GUI screen. In the '075 method, specific screen identifying information is downloaded from a communications server to a remote terminal, along with communications software, and is used to identify information from the communications server. The district court held the '075 patent to be valid, and infringed by the NewLook products. Lansa appeals both rulings. As to validity, Lansa argues that the claimed subject matter would have been obvious based on prior art that the district court excluded, and also that the claimed invention is barred or rendered obvious by Lansa's advertisement for sale and use of an early version of the accused products. The Flashpoint Manuals The only references on which Lansa relies for its validity challenge are two user manuals for a software product called "Flashpoint." The issue at trial was whether either or both of these manuals is a "printed publication" in terms of § 102(b), and thus available as evidence of anticipation or obviousness. One of the Flashpoint manuals is marked "an unpublished work and is considered a trade secret belonging to the copyright holder." The second manual is not marked with any indicium of either publication or secrecy. There was no evidence as to the source, publication, or public accessibility of either manual. The district court found that "no witness testified, nor was any evidence presented, that either of these documents was ever published or disseminated to the public." ResQNet III, 533 F.Supp.2d at 414. *866 Public accessibility is the "touchstone in determining whether a reference constitutes a `printed publication' bar under 35 U.S.C. § 102(b)." In re Hall, 781 F.2d 897, 899 (Fed.Cir.1986). The only argument presented by Lansa was ResQNet's subsequent inclusion of these manuals in an Information Disclosure Statement (IDS) that ResQNet submitted to the Patent Office during a reexamination proceeding for a different patent. Lansa states that ResQNet amended the claims in reexamination, in response to the examiner's rejection based on one of the IDS manuals, and that ResQNet thereby admitted that the manuals were printed publications. ResQNet responds that it learned of these manuals only when Lansa produced them in this litigation, and deemed it prudent to submit them in the unrelated reexamination proceeding, rather than risk the charge of concealing them. ResQNet states that its submission of the manuals was not an admission that they were publicly available publications. In Abbott Laboratories v. Baxter Pharmaceutical Products, Inc., 334 F.3d 1274, 1279 (Fed.Cir.2003), the court explained that "mere submission of an IDS to the USPTO does not constitute the patent applicant's admission that any reference in the IDS is material prior art." We agree that ResQNet did not convert these manuals into printed publication prior art by including them with the IDS submitted to the PTO. No other evidence of publication or public availability was provided. The only references identified by Lansa are the Flashpoint manuals. No error has been shown in the district court's ruling that these documents are not printed publications under § 102(b) and thus are not prior art. We affirm that invalidity on the ground of obviousness has not been shown. Lansa's Advertisement for Sale Lansa states that its initial product, NewLook version 1.0, was advertised for sale as early as March 1996, over a year before the '075 patent application was filed. Lansa states that if its NewLook product is indeed deemed to be infringing, then this advertisement constituted an offer for sale of the patented system, thereby invalidating the '075 patent. An offer for sale, sale, or public use, if more than one year before the patent application was filed, will bar patenting of the product, even if the sale was not authorized by the patentee. See In re Caveney, 761 F.2d 671, 675 (Fed.Cir.1985) ("[S]ales or offers by one person of a claimed invention will bar another party from obtaining a patent if the sale or offer to sell is made over a year before the latter's filing date. An exception to this general rule exists where a patented method is kept secret and remains secret after a sale of the unpatented product of the method. Such a sale prior to the critical date is a bar if engaged in by the patentee or patent applicant, but not if engaged in by another.") (citations omitted). As the court explained in Caveney, "The `on sale' provision of 35 U.S.C. § 102(b) is directed at precluding an inventor from commercializing his invention for over a year before he files his application. Sales or offers made by others and disclosing the claimed invention implicate the `public use' provision of 35 U.S.C. § 102(b)." Id. at 675 n. 5. The district court found that NewLook version 1.0, as offered for sale by Lansa, lacked an essential limitation of claim 1, which requires "downloading, from said server to said terminal, communications software for communicating between said terminal and said host and a plurality of specific screen identifying information"; that is, there must be communications software installed on the server that the dumb terminal can import after it connects to *867 that server. The district court found that NewLook version 1.0 lacked a built-in terminal emulator, and instead relied on third-party "emulation software" for communication with the host, after which NewLook version 1.0 would interface with the emulator to produce GUIs. The district court found, resolving conflicting testimony, that whatever downloading happened, it happened because of other software, not that of NewLook's product. Therefore, the district court found that NewLook version 1.0 did not embody all of the elements of claim 1 of the '075 patent; and Lansa has not argued that its advertisement provided sufficient detail to constitute disclosure of the system embodied in this version of NewLook. Clear error has not been shown in the district court's finding that the offer for sale of this earlier version did not constitute prior art. Lansa argues that even if a third party emulator were needed to fulfill the claim, the user of the system would perform the missing step and thus the entire claimed method would be practiced. However, Lansa did not establish that the entirety of the invention claimed in the '075 patent was known to others before the critical date. We affirm the court's ruling that invalidity was not established on this ground. Infringement Infringement is a question of fact, and the district court's finding of infringement is reviewed for clear error. Alza Corp. v. Mylan Laboratories, Inc., 464 F.3d 1286, 1289 (Fed.Cir.2006). The district court found that all of the limitations of claim 1 were embodied in the NewLook software product. The principal elements of the NewLook system that were at issue with respect to infringement are a "Screen ID" or text designated by the developer that tells the program to apply an override; an "Identify" function; and the role of "filters." The Screen ID does not necessarily identify a particular screen and can be present on multiple screens or none at all. The "Identify" function identifies specific elements on a screen, whereby the developer, but not the user, can use the Identify function to apply overrides and to make changes to a single screen. The NewLook system also uses "filters" to make global changes, and when a filter is defined, it applies across all screens. It was not disputed that NewLook has an "Identify" tool that allows the program developer to select the "Screen ID" that the computer program reads and to which it responds. Lansa argues that the '075 patent requires that each screen is identified by a Screen ID that is unique for each given screen, and that the district court incorrectly broadened claim 1 to require merely an algorithm that recognizes the screen based on the information downloaded from the mainframe. Lansa's argument is directed to a construction that is narrower than this court's claim construction ruling, which held that the '075 claim "require[s] an algorithm that recognizes the screen based on the information downloaded from the mainframe to the PC." ResQNet I, 346 F.3d at 1376. Lansa states that it cannot infringe claim 1, for ResQNet stipulated that NewLook "does not use a screen ID generator algorithm." ResQNet II, 382 F.Supp.2d at 444 n. 17. Lansa also states that, even under the claim construction adopted by the district court, the NewLook system does not infringe because the system converts into GUI format individual elements of green screens without ever recognizing the screen being displayed. ResQNet disputes this view of the evidence and of the NewLook system. At the trial there was testimony that the algorithm employed by NewLook generates a "unique Screen ID." Also, the evidence *868 at trial was that NewLook's Screen IDs do not necessarily identify a given screen, as would a unique alphanumeric code for an individual screen. However, there was also evidence that this does not mean that the NewLook Screen IDs cannot or do not uniquely identify screens. There was evidence that a program developer can select Screen IDs that uniquely identify a screen so that GUI overrides will apply to one screen only. The district court apparently placed weight on the NewLook user manual, which includes the statements that: "Identify is used to specify the green screen image overrides that are then stored in the SID database. Typically, you will select a unique screen element... to ensure you[r] overrides apply to a specific green screen only"; "The screen IDs for each language need to be unique to ensure a ... screen will be correctly recognized"; and "This process continues until all screens are known to be unique." The manual explains that NewLook displays a "conflict" message if a Screen ID appears on more than one screen, thus prompting the developer to correct a likely mistake, whereas when the developer selects a unique Screen ID, there is no conflict message. These aspects were fully explored in the district court, leading to the court's application of the claim to the NewLook system. We do not discern error in the district court's determination that the '075 claim does not require using an algorithm to generate a unique screen ID, and instead requires only the recognition of a screen via some algorithm, such as one that determines if there is a match between the screen identifying information and the received screen. See ResQNet I, 346 F.3d at 1376 (claim 1 of the '075 patent "require[s] an algorithm that recognizes the screen based on the information downloaded from the mainframe to the PC"). Lansa did not dispute that, in NewLook, a received green screen is read merely to see if there is a match. This tracks the claim requirement that "said received screen matches one of the plurality of specific screen identifying information." The district court's findings that NewLook uniquely identifies screens by generating screen IDs have not been shown to be clearly erroneous. On the record and argument, clear error has not been shown in the district court's finding of infringement of the '075 patent. The finding is affirmed. III Damages Lansa's cross-appeal challenges the district court's award of $506,305 in damages for infringement of ResQNet's '075 patent. This amount reflects the district court's acceptance of a 12.5% reasonable royalty rate applied to Lansa's revenues from the sale of infringing NewLook software. Lansa challenges the methodology used by ResQNet's damages expert, Dr. Jesse David, in determining this reasonable royalty. Because the district court's award relied on speculative and unreliable evidence divorced from proof of economic harm linked to the claimed invention and is inconsistent with sound damages jurisprudence, this court vacates the damages award and remands. A Upon a showing of infringement, a patentee is entitled to "damages adequate to compensate for the infringement, but in no event less than a reasonable royalty for the use made of the invention by the infringer." 35 U.S.C. § 284. A "reasonable royalty" derives from a hypothetical negotiation between the patentee and the infringer when the infringement began. See, e.g., Unisplay, S.A. v. American Elec. *869 Sign Co., 69 F.3d 512, 517 (Fed.Cir.1995). A comprehensive (but unprioritized and often overlapping) list of relevant factors for a reasonable royalty calculation appears in Georgia-Pacific Corp. v. United States Plywood Corp., 318 F.Supp. 1116, 1120 (S.D.N.Y.1970). "Determining a fair and reasonable royalty is often ... a difficult judicial chore, seeming often to involve more the talents of a conjurer than those of a judge." Fromson v. Western Litho Plate & Supply Co., 853 F.2d 1568, 1574 (Fed. Cir.1988). Still, a reasonable royalty analysis requires a court to hypothesize, not to speculate. Id. at 1575. At all times, the damages inquiry must concentrate on compensation for the economic harm caused by infringement of the claimed invention. See, e.g., Aro Mfg. Co. v. Convertible Top Replacement Co., 377 U.S. 476, 507, 84 S.Ct. 1526, 12 L.Ed.2d 457 (1964) ("[T]he present statutory rule is that only `damages' may be recovered."). Thus, the trial court must carefully tie proof of damages to the claimed invention's footprint in the market place. See, e.g., Grain Processing Corp. v. American Maize-Prods. Co., 185 F.3d 1341, 1350 (Fed.Cir.1999) ("To prevent the hypothetical from lapsing into pure speculation, this court requires sound economic proof of the nature of the market and likely outcomes with infringement factored out of the economic picture."); Riles v. Shell Exploration & Prod. Co., 298 F.3d 1302, 1312 (Fed.Cir.2002) ("[T]he market would pay [the patentee] only for his product.... [The patentee's damages] model [does not support the award because it] does not associate [the] proposed royalty with the value of the patented method at all, but with the unrelated cost of the entire Spirit platform."). Any evidence unrelated to the claimed invention does not support compensation for infringement but punishes beyond the reach of the statute. With these principles in mind, this court just recently rejected a patentee's reliance on licenses because "some of the license agreements [were] radically different from the hypothetical agreement under consideration" and the court was "unable to ascertain from the evidence presented the subject matter of the agreements." Lucent Techs., Inc. v. Gateway, 580 F.3d 1301, 1327-28 (Fed.Cir.2009). The majority of the licenses on which ResQNet relied in this case are problematic for the same reasons that doomed the damage award in Lucent. B ResQNet's expert Dr. David determined the "starting point" for a hypothetical negotiation based on the first factor of the Georgia-Pacific framework — royalties received by the patentee from existing licenses. The first Georgia-Pacific factor requires considering past and present royalties received by the patentee "for the licensing of the patent in suit, proving or tending to prove an established royalty." 318 F.Supp. at 1120 (emphasis added). By its terms, this factor considers only past and present licenses to the actual patent and the actual claims in litigation. This court has long required district courts performing reasonable royalty calculations to exercise vigilance when considering past licenses to technologies other than the patent in suit. See Lucent, 580 F.3d at 1329 ("[A] lump-sum damages award [based on a reasonable royalty] cannot stand solely on evidence which amounts to little more than a recitation of royalty numbers, one of which is arguably in the ballpark of the jury's award, particularly when it is doubtful that the technology of those license agreements is in any way similar to the technology being litigated here."). *870 Yet Dr. David used licenses with no relationship to the claimed invention to drive the royalty rate up to unjustified double-digit levels. Dr. David based his damages on seven ResQNet licenses, five of which had no relation to the claimed invention. These five re-branding or re-bundling licenses (hereinafter, the "re-bundling licenses") furnished finished software products and source code, as well as services such as training, maintenance, marketing, and upgrades, to other software companies in exchange for ongoing revenue-based royalties. These companies obtained the right to re-brand ResQNet's products before resale or bundle these products into broader software suites. While the specific numbers involved in these licenses are under a protective order, this court observes that two of them mentioned a top rate of 25%, two more a top rate of 30%, and still another a top rate of 40%. Notably, none of these licenses even mentioned the patents in suit or showed any other discernible link to the claimed technology. Dr. David tabulated an average of the royalty ranges specified in these agreements, a number substantially higher than 12.5%. The rates in the re-bundling licenses are not consistent at all with the other two licenses in the record. Those two "straight" licenses arose out of litigation over the patents in suit. One of them was a lump-sum payment of stock which Dr. David was unable to analogize to a running royalty rate. The other was an ongoing rate averaging substantially less than 12.5% of revenues. In his own words, Dr. David recommended a rate for his hypothetical negotiation "somewhere in the middle" of the re-bundling licenses and the straight rate-based license on the claimed technology. Trial Tr. 34:7, May 21, 2007. He considered a few of the other Georgia-Pacific factors, but dismissed them because "[f]or the most part, the other factors have no real impact here." Id. at 36:13-14. Thus, Dr. David calculated that a mid-range of 12.5% was the appropriate royalty rate. The inescapable conclusion is that Dr. David used unrelated licenses on marketing and other services — licenses that had a rate nearly eight times greater than the straight license on the claimed technology in some cases — to push the royalty up into double figures. The district court adopted Dr. David's 12.5% royalty rate and set damages at $506,305. This court finds two parts of this analysis particularly troubling: first, the extremely high rates in the re-bundling licenses compared with the license on the claimed technology, and second, the unconvincing reasons that Dr. David gave for considering these re-bundling licenses at all. On this second point, the trial transcript indicates several instances where Dr. David misunderstood (or worse, misrepresented) the re-bundling licenses as somehow amounting to "patent plus software" licenses when, in fact, the record shows no use in these licenses of ResQNet's claimed invention: Those [re-bundling] licenses do, of course — ResQNet does, of course, provide code; they don't just provide the patent or rights to use the patent. .... * * * * * I'd just like to add one more point about the straight patent license versus a code plus a patent license. ... [If] you don't want to do exactly what that code is written to do, with a straight patent license you can customize the product, whereas if you get code you're stuck with it.... * * * * * For the most part, the other [Georgia-Pacific] factors have no real impact here. The reason is we are comparing *871 one license, this hypothetical negotiation license, to a bunch of other licenses. In both cases ResQNet is licensing its patents and maybe some code as well, but in all the cases ResQNet is the licensor and the product is basically analogous. Id. at 33:23-36:18 (emphases added). In sum, Dr. David offers little or no evidence of a link between the re-bundling licenses and the claimed invention. Yet he relies on these licenses to inflate his royalty recommendation.[1] Thus, the district court in this case made the same legal error that this court corrected in Lucent. This trial court, like the one in Lucent, made no effort to link certain licenses to the infringed patent. For his part, Dr. David did not provide any link between the re-bundling licenses and the first factor of the Georgia-Pacific analysis. Without that link, as this court explained in Lucent: "We ... cannot understand how the [fact finder] could have adequately evaluated the probative value of [the] agreements." 580 F.3d at 1328. In addition to Lucent, this court's decision in Trell v. Marlee Electronics Corp., 912 F.2d 1443 (Fed.Cir.1990), contains this same instructive rule. In Trell, the patentee owned rights to a claim covering a "common combination lock device" that would allow a visitor to open a locked door by entering a special code. Id. at 1444. The trial court awarded a 6% royalty based on Trell's prior license with a European company, Bewator. This court vacated even this 6% rate (far less than in this case), finding that the Bewator license was not commensurate to the patent in suit. As relevant here, the Bewator agreement "conveyed rights more broad in scope than those covered by Trell's patent." Id. at 1446. Accordingly, this court found that "[t]he district court's apparent failure to consider the fact that the Bewator license was exclusive and that it encompassed the right to other inventions compels reversal." Id. at 1447. This case presents a situation far more egregious than found in Trell. In Trell, the parties did not even dispute that the European license related directly to the claim in question — the only quarrel was whether the Bewator license encompassed more than the infringed claim. But here, as Lansa protested to the district court, ResQNet's five re-bundling licenses are absolutely silent on any relation to the patents in suit. Dr. David did not even attempt to show that these agreements embody or use the claimed technology or otherwise show demand for the infringed technology. In simple terms, the '075 patent deals with a method of communicating between host computers and remote terminals — not training, marketing, and customer support services. The re-bundling licenses simply have no place in this case. Dr. David's decision to adjust his proposed rate downward to arrive at a (still unsubstantiated) starting point for the hypothetical negotiation resulted in a rate that was still more than twice the rate on the straight rate-based license that covered the claimed invention. Actually, Dr. David's downward shift from the re-bundling royalties is an admission that his calculations are speculative without any relation to actual market rates at all. The *872 first Georgia-Pacific factor, which Dr. David found to be controlling and which the district court in turn adopted, must consider licenses that are commensurate with what the defendant has appropriated. If not, a prevailing plaintiff would be free to inflate the reasonable royalty analysis with conveniently selected licenses without an economic or other link to the technology in question. The district court seems to have been heavily influenced by Lansa's decision to offer no expert testimony to counter Dr. David's opinion. But it was ResQNet's burden, not Lansa's, to persuade the court with legally sufficient evidence regarding an appropriate reasonable royalty. See Lucent, 580 F.3d at 1329 ("Lucent had the burden to prove that the licenses were sufficiently comparable to support the lump-sum damages award."). As a matter of simple procedure, Lansa had no obligation to rebut until ResQNet met its burden with reliable and sufficient evidence. This court should not sustain a royalty award based on inapposite licenses simply because Lansa did not proffer an expert to rebut Dr. David. See SmithKline Diagnostics, Inc. v. Helena Labs. Corp., 926 F.2d 1161, 1168 (Fed.Cir.1991) ("A court is not restricted in finding a reasonable royalty to a specific figure put forth by one of the parties."). Moreover the record already contained evidence of licenses on the claimed technology. Lansa was entitled to rely on that record evidence to show a royalty rate reasonably related to the technology in this litigation. This court observes as well that the most reliable license in this record arose out of litigation. On other occasions, this court has acknowledged that the hypothetical reasonable royalty calculation occurs before litigation and that litigation itself can skew the results of the hypothetical negotiation. See Hanson v. Alpine Valley Ski Area, Inc., 718 F.2d 1075, 1078-79 (Fed.Cir.1983) ("[S]ince the offers were made after the infringement had begun and litigation was threatened or probable, their terms should not be considered evidence of an `established royalty,' since license fees negotiated in the face of a threat of high litigation costs may be strongly influenced by a desire to avoid full litigation.") (quotations and alterations omitted). Similarly this court has long recognized that a reasonable royalty can be different than a given royalty when, for example, widespread infringement artificially depressed past licenses. See, e.g., Nickson Indus., Inc. v. Rol Mfg. Co., 847 F.2d 795, 798 (Fed.Cir.1988); Fromson, 853 F.2d at 1577 n. 15 ("[A] court should not select a diminished royalty rate a patentee may have been forced to accept by the disrepute of his patent and the open defiance of his rights.") (quotation marks and citation omitted). And a reasonable royalty may permissibly reflect "[t]he fact that an infringer had to be ordered by a court to pay damages, rather than agreeing to a reasonable royalty." Maxwell v. J. Baker, Inc., 86 F.3d 1098, 1109-10 (Fed. Cir.1996); see also TWM Mfg. Co. v. Dura Corp., 789 F.2d 895, 900 (Fed.Cir.1986) ("That [the patentee] might have agreed to a lesser royalty is of little relevance, for to look only at that question would be to pretend that the infringement never happened."). On remand, the district court will have the opportunity to reconsider the reasonable royalty calculation. At that time, the district court may also consider the panoply of "events and facts that occurred thereafter and that could not have been known to or predicted by the hypothesized negotiators." Fromson, 853 F.2d at 1575. During that remand, however, the trial court should not rely on unrelated licenses to increase the reasonable royalty rate *873 above rates more clearly linked to the economic demand for the claimed technology. In sum, the district court erred by considering ResQNet's re-bundling licenses to significantly adjust upward the reasonable royalty without any factual findings that accounted for the technological and economic differences between those licenses and the '075 patent. A reasonable royalty based on such speculative evidence violates the statutory requirement that damages under § 284 be "adequate to compensate for the infringement." Thus, this court vacates the damages award and remands to the district court for a recalculation of a reasonable royalty in accordance with this opinion. IV The Rule 11 Sanctions The district court assessed sanctions against ResQNet and its counsel, on the ground that they should have withdrawn both the '608 patent and the '127 patent at an early stage of suit. The imposition of sanctions is reviewed on the standard of abuse of discretion. See Cooter & Gell v. Hartmarx Corp., 496 U.S. 384, 405, 110 S.Ct. 2447, 110 L.Ed.2d 359 (1990) ("A district court would necessarily abuse its discretion if it based its [Rule 11] ruling on an erroneous view of the law or on a clearly erroneous assessment of the evidence."). This court looks to regional circuit law for precedential guidance when reviewing a district court's imposition of sanctions under Rule 11, Power Mosfet Techs., L.L.C. v. Siemens AG, 378 F.3d 1396, 1406-07 (Fed.Cir.2004), for local practices and standards control matters of attorney misconduct. In arguing that the sanction was improper, ResQNet states first that Lansa's Rule 11 motion was excessively untimely, could not have been satisfied, and should have been denied on that ground alone. ResQNet also argues that the district court seriously misconstrued the counsel's letter on which the sanction was based, in that the letter did not concede that there was not infringement of the '127 and '608 patents, but stated only that if Lansa's representations were confirmed by discovery, of which none had yet occurred, the patents would be withdrawn. ResQNet also states that it was incorrect to consider this settlement correspondence as a basis for sanctions, and that Lansa breached the conditions under which the parties had been exchanging information in attempted settlement. The events are summarized as follows: ResQNet filed its infringement complaint in April 2001, stating that Lansa's NewLook system infringed four patents: the '608 patent, the '127 patent, the '659 patent, and the '961 patent. Discussions ensued between the parties, including correspondence between counsel where Lansa stated that its product did not infringe any of the asserted patents, and described some of the technology. In a reply dated September 24, 2001, ResQNet's attorney stated the following as to two of the four patents: 1) With respect to the '127 patent, it does not appear that the Lansa system would infringe any claim, either literally or under the doctrine of equivalents. Thus, presuming we discover no contrary evidence as the case moves forward, [ResQNet] is prepared to remove this patent from the litigation. 2) With regard to the '608 patent, your detailed letter and the materials we have appear to show that the Lansa system does not infringe the claim in the '608 patent. Accordingly, unless we discover evidence to the contrary, ResQNet is *874 also prepared at this point to remove the '608 patent from the litigation. ResQNet II, 382 F.Supp.2d at 455. This correspondence occurred before any discovery had taken place. On December 4, 2001, ResQNet filed an Amended Complaint which continued to allege infringement of the four patents and added the '075 patent, which had issued on September 25, 2001. Discovery began in early 2002, directed to all five patents. After some discovery, ResQNet advised Lansa of its withdrawal of the '127 and '659 patents. The record does not show the exact date, which according to various filings and rulings occurred before May 2002. On June 12, 2002, the district court held a Markman hearing as to the remaining three patents. On September 5, 2002, the district court issued its claim construction order. On November 4, 2002, the parties entered into a stipulated Consent Judgment of non-infringement of the '127 patent and the '659 patent. For the remaining three patents, the '608 patent, the '075 patent, and the '961 patent, the parties agreed to a Consent Judgment that in view of the claim construction rulings of the district court, NewLook does not infringe the claims at issue. The district court entered the Consent Judgment, leading to the first appeal to this court. On that appeal the Federal Circuit affirmed the claim construction as to the '961 patent, modified the construction as to the '608 and '075 patents, and remanded for further proceedings. ResQNet I, 346 F.3d at 1374. In view of the parties' agreement and the Consent Judgment, the affirmance of the claim construction as to the '961 patent removed that patent from the case. Id. at 1384. The parties then engaged in further discovery with respect to the '608 patent and the '075 patent, and various motions for summary judgment were filed. A motion filed by Lansa on August 23, 2004 requested summary judgment of noninfringement of the '608 patent. This motion was denied on January 13, 2005. ResQNet II, 382 F.Supp.2d at 444-48. Meanwhile, on September 3, 2004 Lansa served on ResQNet a motion for Rule 11 sanctions, charging, among other criticisms, that ResQNet and its counsel had continued to litigate the '127 and '608 patents notwithstanding their earlier statement, in the September 24, 2001 letter between counsel, that these patents appeared not to be infringed. In accordance with Rule 11, ResQNet had 21 days to withdraw the offending complaint, which was the Amended Complaint filed on December 4, 2001.[2] ResQNet took no action during the 21 day period following service of the motion, and Lansa filed the Rule 11 motion with the district court on September 29, 2004. On January 13, 2005 the district court granted the motion for sanctions, concurrently with denying Lansa's motion for summary judgment of noninfringement of the '608 patent. Id. at 452-57. In awarding the Rule 11 sanction, the court recited the correspondence between ResQNet and Lansa, specifically the letter quoted above, and held that in light of that letter ResQNet had no good faith basis on which to allege infringement of the '127 and '608 patents at the time it filed the Amended Complaint in December 2001 "after having expressly determined that *875 the prior belief of infringement of those patents had been incorrect and in the absence of any intervening developments from which a good faith basis to bring the claims might be inferred." Id. at 457. ResQNet points out that it had withdrawn the '127 patent after initial discovery in the spring of 2002 and had so informed Lansa, and that the '127 patent was formally dismissed in November 2002. ResQNet points out that when Lansa's Rule 11 motion was served on September 3, 2004, the alleged violation as to both the '127 and the '608 patents had occurred three years earlier, and the '127 patent had been dismissed with prejudice two years earlier and thus could not be "remedied" within 21 days. Courts that have discussed the matter have endorsed the application of time limits on Rule 11 motions. See, e.g., In re Pennie & Edmonds LLP, 323 F.3d 86, 89 (2d Cir.2003) ("the `safe harbor' provision functions as a practical time limit, and motions have been disallowed as untimely when filed after a point in the litigation when the lawyer sought to be sanctioned lacked an opportunity to correct or withdraw the challenged submission"). The Advisory Committee's Notes to the 1993 amendments to the rule state that any motion alleging violation of Rule 11 "should be served promptly after the inappropriate paper is filed, and, if delayed too long, may be viewed as untimely." The general practice is that the motion must be filed before the offending contention has been withdrawn or resolved. See Perpetual Sec., Inc. v. Tang, 290 F.3d 132, 142 (2d Cir.2002). Lansa's motion was untimely as to the '127 patent, which had been withdrawn two years before the motion for sanctions was filed. As to the '608 patent, no act of bad faith has been shown, or proposed, in ResQNet's not having withdrawn the '608 patent after its attorney letter of September 24, 2001, which explicitly stated that its position was based on Lansa's representations, which had not been the subject of any discovery. It is significant that the district court declined to grant summary judgment of noninfringement, and the '608 patent proceeded to full trial. Although the district court stated, in its order refusing to vacate the sanction, that the sanction was based on ResQNet's continued assertion of the '127 and '608 patents after the September 24, 2001 letter, and not on later events, ResQNet IV, 2008 WL 4376367, at *6, the district court's denial of summary judgment of noninfringement reflects the belief that it was reasonable for ResQNet to have retained that patent for suit. The Advisory Committee's Notes to the 1993 amendments to Rule 11 state that "if a party has evidence with respect to a contention that would suffice to defeat a motion for summary judgment based thereon, it would have sufficient `evidentiary support' for purposes of Rule 11." ResQNet's counsel's letter of September 24, 2001 responded to Lansa's description of its technology, as provided during settlement discussions, and stated that if Lansa's position were verified upon discovery, the '127 and '608 patents would not be infringed. The '127 patent was withdrawn after discovery, but the '608 patent was not. We cannot share the district court's reading of this letter as requiring immediate withdrawal of the '127 patent and the '608 patent. For the several reasons we have discussed — the untimeliness of the motion, the prompt withdrawal of the '127 patent, and the recognition of litigation substance concerning the '608 patent[3] — *876 we conclude that the award of sanctions was an abuse of the court's discretion, and is reversed. Each party shall bear its costs. AFFIRMED IN PART, REVERSED IN PART, VACATED AND REMANDED IN PART NEWMAN, Circuit Judge, concurring in part, dissenting in part. I join the court's opinion with the exception of Part III, damages. On the question of damages, my colleagues on this panel have departed from the guidance and the requirements of precedent, distorting the principles of this court's decisions, including such recent rulings as Lucent Technologies, Inc. v. Gateway, Inc., 580 F.3d 1301 (Fed.Cir.2009), and i4i Limited Partnership v. Microsoft Corp., 589 F.3d 1246 (Fed.Cir.2009). My colleagues on this panel hold that it is improper to consider, for the purpose of understanding the value of the infringed patents, any licenses involving the technology of those patents bundled with additional technologies, such as software code. Thus the court holds that it was legal error to take cognizance of most of the existing licenses introduced at trial. The reasoned consideration by the district court is ignored, and the evidence is misconstrued. This is not a case of constructing, and applying, a royalty rate from totally unrelated content; it is simply a case of determining the evidentiary value of the infringed subject matter by looking at the various licenses involving that subject matter, and allocating their proportional value, with the assistance of undisputed expert testimony. From my colleagues' misperception of this process, I respectfully dissent from Part III of the court's opinion. I In the district court, ResQNet's damages expert Dr. David, a qualified economist with experience in the field, followed the traditional application of the Georgia-Pacific factors, see Georgia-Pacific Corp. v. United States Plywood Corp., 318 F.Supp. 1116 (S.D.N.Y.1970), analyzing the impact of all of these factors in an extensive Expert Report and in testimony at trial. He was subject to examination and cross-examination in the district court, and the district court provided a full and reasoned analysis of the evidence. No flaw in this reasoning has been assigned by my colleagues, who, instead, create a new rule whereby no licenses involving the patented technology can be considered, in determining the value of the infringement, if the patents themselves are not directly licensed or if the licenses include subject matter in addition to that which was infringed by the defendant here. In this case, the added subject matter was usually the software code that implements the patented method, as the district court recognized, and whose contribution to the value of the license was evaluated by the damages *877 expert and discussed by the court. My colleagues' ruling today that none of that information is relevant to the assessment of damages is unprecedented, and incorrect. The evidence at trial fully explored the licenses granted by ResQNet for its technology. Dr. David's Expert Report addressed all seven of these license agreements, explaining what they cover and how they relate to the infringement herein. Two of the licenses had been entered in the settlement of litigation, and the others included not only the patented technology but some additional subject matter, generally the software code for use with the patented system. These are the "bundled" licenses that my colleagues exclude in their entirety, apparently equating them with discredited verdicts in unrelated cases on different facts, that were based upon a royalty base that included the value of unpatented and unrelated subject matter. That is not here the case. To the contrary, the district court here recognized the differences between these licenses and the analysis at hand, and took those differences into account. The district court also took into account the licensing practices and royalty rates in this industry, in determining a fair royalty for the specific infringement by Lansa. Dr. David explained that the 12.5% royalty was in the conventional range of one fourth to one third of the licensee's profit from use of the patented technology. He explained that this rate was significantly lower than the royalties in the bundled licenses, and higher than the royalty in one of the litigation settlements. All of these agreements were analyzed at trial, and their application explained, as the district court applied the principles of the hypothetical negotiation. The district court, discussing the expert's Report and testimony, stated: None of the licenses considered by Dr. David is a perfect approximation of the hypothetical license between ResQNet and Lansa. The licensing agreements ResQNet reached with IBM, Hummingbird, Crystal Point, ICOM, and Ericom between 1998 and 2002 each involved licensing of ResQNet's software or code, and each involved rates higher (some substantially so) than 12.5%. ResQNet's only two straight patent licenses, one of which was lower than 12.5%, were granted in the shadow of litigation, and without the assured validity of the '075 Patent. Dr. David's conclusion that the reasonable rate lies between these two categories' averages (and closer to the lower one) is well-reasoned and supported in the record. In fact, by omitting the upfront payments present in the majority of ResQNet licenses, Dr. David's methodology is actually biased in favor of lowering the estimated reasonable royalty. ResQNet.com, Inc. v. Lansa, Inc., 533 F.Supp.2d 397, 417-18 (S.D.N.Y.2008) (citations omitted). My colleagues on this panel assign no error to this solid reasoning; they simply ignore it, stating that "[t]he re-bundling licenses simply have no place in this case." Maj. Op. at 871. Neither Lucent nor any other precedent dictates such a blanket exclusion of relevant evidence. The correct approach is that which was followed by the district court. In testimony, Dr. David explained the higher rates for the bundled licenses, and explained why these rates should not be applied to Lansa's infringement. However, he did not ignore these licenses, instead explaining why all of the licenses are properly considered: ResQNet does, of course, provide code; they don't just provide the patent or rights to use the patent. And so clearly *878 there should be a recognition that you're getting more for your money there in a rebranding or bundling arrangement than you are in a straight patent license. On the other hand, the two straight patent licenses that we do have were reached in the course of litigation when the patents were clearly being challenged. So the right number ought to be somewhere in the middle. JA2264-65. In his Expert Report, Dr. David tied the bundling licenses to the patented technology. He explained that "ResQNet's products are known as `terminal emulation' software and are based on the technology described in the patents in suit." The report also states that the bundling licenses are directed to ResQNet's software products, as well as code for the terminal emulation technology. This evidence was unrebutted. My colleagues now hold that no attention at all can be given to the bundled licenses, or the relation of their content to the patented technology, for my colleagues conclude without explanation that the licenses have "no relation to the claimed invention." Maj. Op. at 870. My colleagues hold that it is reversible error, as a matter of law, to have considered these licenses at all. In the heavily fact-driven obligation of the district court with respect to assessment of damages, this is clearly incorrect. Further, Lansa did not dispute the evidence and testimony provided by ResQNet.[1] The district court discussed all seven existing licenses. Two of these agreements were in settlement of litigation. One of these settlement licenses, between ResQNet and Seagull Software Systems, Inc., included a lump sum amount and equity participation. The panel majority holds that the rates in the bundling licenses are "not consistent at all" with the Seagull license. Maj. Op. at 870. This is speculation, for ResQNet's damages expert was not able to analogize the lump sum amount to a royalty rate due to the absence of additional financial information pertinent to the value of the settlement transaction. Indeed, this court observed in Lucent, 580 F.3d at 1330, that "fundamental differences exist between lump-sum agreements and running-royalty agreements" — not for the purpose of excluding such evidence, but to point out that such differences must be recognized. The Seagull license is relevant, for the lump sum amount therein is substantially greater than the amount that was here awarded to ResQNet. The other settlement license was in settlement of infringement litigation against Zephyr Development Corporation. The existence of this license and its lower rate[2] was a major factor in the district court's determination that a hypothetical negotiation would produce a royalty significantly lower than the 24% average of ResQNet's other licenses. The court also was influenced by Lansa's profit margin in its sale of the infringing products, which supported the traditional rate of one fourth to one third of profits. In considering the settlement licenses, the district court commented that the settlement of ongoing litigation can involve considerations quite different from the "hypothetical negotiation," which is conducted on the premise that the *879 patent is valid and would be infringed. Thus the district court selected a royalty higher than that in the litigation settlement, although much lower than for any of the licenses that included the software code. My colleagues, in setting strict barriers as to what evidence can be considered, leave the damages analysis without access to relevant information. However, it is not necessary that the identical situation existed in past transactions, for the trier of fact to determine the value of the injury. Story Parchment Co. v. Paterson Parchment Paper Co., 282 U.S. 555, 563, 51 S.Ct. 248, 75 L.Ed. 544 (1931) ("it will be enough if the evidence show the extent of the damages as a matter of just and reasonable inference, although the result be only approximate"); State Industries, Inc. v. Mor-Flo Industries, Inc., 883 F.2d 1573, 1576-77 (Fed.Cir.1989) ("Deciding how much to award as damages is not an exact science, and the methodology of assessing and computing damages is committed to the sound discretion of the district court."). Lansa offered no evidence, no testimony, no witness. The role of the trial is to provide evidence for the trier of fact to consider, weigh, and credit as appropriate. Paper Converting Mach. Co. v. Magna-Graphics Corp., 745 F.2d 11, 22 (Fed.Cir. 1984) ("Determining the weight and credibility of the evidence ... is the special province of the trier of fact.") (internal quotations omitted). The district court's endorsement of a royalty of 12.5% was explained as based on the adjudication of validity and infringement, and as a balance of the royalties actually paid for licenses to this technology with software code and the royalties in the litigation settlement, with due consideration to Lansa's profits on the infringing products. This damages assessment is not analogous to that criticized in Lucent, where the damages award was based on the entire market value of a system in which the infringing component was but a small part. Here, in contrast, the patented technology was a large part of the "bundled" licenses, and these licenses were fairly considered for their content and value. In addition, my colleagues diverge from the principles of the "hypothetical negotiation," for the theory of such "negotiation" as a tool in assessing patent damages is that the patent is valid and that a license is needed to avoid infringement. Such a hypothetical negotiation takes into account the profit of the licensee for use of the licensed patent. Lansa testified, through its chief financial officer, that the infringing technology was resold by Lansa at specified profit margins.[3] There was evidence that in this field of commerce the customary royalty is one fourth to one third of the licensee's profit on the licensed subject matter. See, e.g., i4i Limited Partnership, 589 F.3d at 1268 (referring to "the 25-percent rule ... which assumes the inventor will keep 25% of the profits from any infringing sales"). That evidence conforms with the rate of 12.5% awarded by the district court. The panel majority thus appears to exclude all evidence except for the royalty in the settlement agreement between ResQNet and Zephyr Development Corporation. The district court observed that licenses entered during litigation are not necessarily comparable to licenses negotiated between a willing licensor and licensee before infringement has begun. The court is not required to pretend that the litigation context was absent, with its burdens, costs, and uncertainties. The case that originated the "hypothetical negotiation" itself cautioned that when a "reasonable *880 royalty" is the basis for damages, it is not "the normal, routine royalty non-infringers might have paid." Panduit Corp. v. Stahlin Bros. Fibre Works, 575 F.2d 1152, 1158 (6th Cir.1978). The courts thus implemented the statutory prescription that patent damages shall be "in no event less than" a reasonable royalty. 35 U.S.C. § 284. It is the minimum, not the maximum. Bandag, Inc. v. Gerrard Tire Co., 704 F.2d 1578, 1583 (Fed.Cir.1983) (a reasonable royalty is "merely the floor below which damages shall not fall"). The district court observed that in litigation the patent is already at risk. The unpredictability of patent litigation remains notorious. In addition, particular litigation settlements may be based on unique considerations. Lansa itself argues that the royalties of litigation-induced licenses should not be considered, citing Hanson v. Alpine Valley Ski Area, Inc., 718 F.2d 1075, 1078-79 (Fed.Cir.1983) ("since the offers were made after the infringement had begun and litigation was threatened or probable, their terms should not be considered evidence of an established royalty, since license fees negotiated in the face of a threat of high litigation costs may be strongly influenced by a desire to avoid full litigation"). Lansa itself argues that "negotiations performed in the context of litigation are not reliable as a basis for determining a reasonable royalty," Corrected Brief in Opposition, at 42, contrary to the position of my colleagues herein. In contrast to precedent, the panel majority moves the Zephyr agreement to the forefront of the analysis, assuring the infringer, after losing in litigation, of no worse penalty than the lowest royalty previously accepted in settlement. As stated in TWM Manufacturing Co. v. Dura Corp., 789 F.2d 895, 900 (Fed.Cir.1986), such a rule would "make an election to infringe a handy means for competitors to impose a `compulsory license' policy upon every patent owner." It is also contrary to the protocol of the hypothetical negotiation, which is designed for use when there is no established royalty. See, e.g., Grain Processing Corp. v. American Maize-Products Co., 185 F.3d 1341, 1353 n. 5 (Fed.Cir.1999) ("The court candidly stated that the 3% rate is its `best estimate,' an honest observation that would apply to most reasonable royalty analyses, given the difficulty of determining a hypothetical agreement between parties which did not actually agree on anything at all."). The panel majority states that Lucent requires its ruling. Lucent held that the damages award should relate to the value of the patented technology, not to the entire market value of a system of which the patented technology is a demonstrably small part. In Lucent the damages award was "roughly three to four times the average amount in the lump-sum agreements in evidence," id. at 1332, grossly unlike the present situation, where the royalty rate is less than the average of all the agreements, related to the patented technology. This case is in marked contrast to the situation in Lucent, where the court stated: "This is not an instance in which the jury chose a damages award somewhere between maximum and minimum lump-sum amounts advocated by opposing parties." 580 F.3d at 1332. In contrast, the damages award herein was indeed somewhere between the highest and lowest rates of these licenses. This court in Lucent also made clear that "we do not conclude that the aforementioned license agreements (or other evidence) cannot, as a matter of law, support the damages award in this case." 580 F.3d at 1335. Today's revision of the principles of Lucent is not tenable. Indeed, the court in Lucent recognized that "any *881 reasonable royalty analysis `necessarily involves an element of approximation and uncertainty.'" Id. at 1325 (quoting Unisplay, S.A. v. American Elec. Sign Co., 69 F.3d 512, 517 (Fed.Cir.1995)). In the thorough analysis of valuation factors that was presented in this case, all of the Georgia-Pacific factors were explained at trial. My colleagues misdescribe this testimony by stating that Dr. David considered only "a few" of those factors, Maj. Op. at 870, for he discussed all fifteen factors. He explained that two of these factors have an "upward influence" on the royalty rate as applied to this particular case, one has a "downward influence," and the remaining factors are neutral. He explained that in this case the first factor was most useful, for it relates to other licenses granted by the patentee for the same technology. On my colleagues' conclusion that most of the licenses granted by the patentee should not be considered at all, this court then should at least consider the other factors, for all were explored at trial, and all were applied to the extent they were relevant. For example, the second factor states that consideration may be given to royalties paid by the licensee to others. Dr. David described Lansa's royalty payments to Looksoftware on revenues of the Newlook product, as well as Lansa's royalty payments to a company called Momentum on another software product. These rates were all significantly higher than the 12.5% that the district court assessed. My colleagues do not mention this evidence, although it is powerful support for the district court's ruling. The fourth Georgia-Pacific factor, concerning the licensor's policies and practices regarding the grant of licenses to its technology, was also discussed in the district court, for it also weighed on the side of a higher royalty rate. Dr. David explained in his Report that, "except to settle ongoing litigation, ResQNet has not provided straight licenses to its patents. Rather, the company has chosen to license the code for its software products or negotiate distributor agreements for the products themselves...." Dr. David also discussed the Georgia-Pacific factor that accounts for portions of the profit that should be credited to features separate from the patented invention; he stated that, "I assume that ResQNet's straight licenses for the patents in suit would have a greater influence on the negotiated royalty rate than the other agreements.... Consequently, I conclude that this factor would tend to decrease the negotiated royalty payment below its initial level." Also relevant is Dr. David's consideration of the twelfth Georgia-Pacific factor, directed to the customary profit for use of the invention or analogous inventions. Dr. David testified as to the evidence of Lansa's profits on the infringing Newlook software and discussed the oft-utilized "25% rule" for damages. He explained that the 12.5% royalty rate met this factor. The other Georgia-Pacific factors were determined by Dr. David to have a neutral effect. The majority ignores the thoroughness of this analysis, and assigns neither flaw nor error to its conclusions. In sum, the district court presented a thorough opinion in which the court explained its selection of the 12.5% rate, while acknowledging that "[t]he determination of the outcome of a hypothetical negotiation is by its very nature an imprecise art." ResQNet, 533 F.Supp.2d at 417. I repeat that Lansa presented no testimony and proffered no evidence. Although Lansa waived the position on which my colleagues rely, this court fills the gap. For example, my colleagues do not discuss the district court's reasoning, but state that they find "particularly troubling" that *882 some of the licenses in evidence had "extremely high rates [when] compared with the [litigation-induced] license on the claimed technology." Maj. Op. at 870. The district court fully considered this aspect, and factored it into a competent overall analysis in which no flaw has been shown. The court's conclusion warrants affirmance. II My colleagues do not discuss the separate award by the district court of an ongoing, "compulsory license" to Lansa, at the same rate of 12.5%. ResQNet did not appeal this aspect of the final judgment. Issues with respect to future infringing activity are quite different from those where the infringement occurred in the past. Compulsory future licenses are rare, particularly when the patentee is itself practicing the invention and would thus be judicially placed in market competition with a licensee it did not seek. My colleagues do not state whether the 12.5% rate may continue to apply to future activity or whether an injunction would now be available. In Paice LLC v. Toyota Motor Corp., 504 F.3d 1293, 1315 (Fed.Cir.2007), this court remanded "for the limited purpose of having the district court reevaluate the ongoing royalty rate." This court also stated: In most cases, where the district court determines that a permanent injunction is not warranted, the district court may wish to allow the parties to negotiate a license amongst themselves regarding future use of a patented invention before imposing an ongoing royalty. Should the parties fail to come to an agreement, the district court could step in to assess a reasonable royalty in light of the ongoing infringement. Id. at 1314-15. CONCLUSION This court today holds that only the royalty in the settlement agreement can be considered in the hypothetical license negotiation. This ruling, excluding all of the other considerations relevant to determining damages for a patent that has been held valid and infringed, is contrary to all precedent. I respectfully dissent. NOTES [1] Dr. David's conclusion that ResQNet's products are "`based on the technology described in the patents in suit,'" Dis. Op. at 878 (quoting Dr. David's expert report), is a far cry from a conclusion that ResQNet's products are coextensive with the claimed invention. Neither this court nor the district court had any way of knowing from Dr. David's report whether ResQNet's products practice, for example, the prior art described in the patents in suit or the actual claimed invention. ResQNet is only entitled to rely on licenses that cover the latter. [2] The purpose of the 21 days is to provide a "safe harbor" against Rule 11 motions, "in that a party will not be subject to sanctions on the basis of another party's motion unless, after receiving the motion, it refuses to withdraw that position or to acknowledge candidly that it does not currently have evidence to support a specified allegation." Fed.R.Civ.P. 11, advisory committee's notes (1993 Amendments). [3] ResQNet's September 24, 2001 letter was designated "FOR SETTLEMENT ONLY," J.A. 625, and Lansa's correspondence stated that the information provided could not be used "for any purpose other than reaching an amicable settlement," J.A. 614 (Aug. 21, 2001). The district court did not mention these conditions, but recognized that settlement letters are exempted from discovery by Rule 408, Fed.R.Evid. However, the court cited authority related to settlement correspondence that was considered for the purpose of impeaching an unrelated claim. Although not necessary to our conclusion that the sanction awarded here was improper, we doubt the relevance of this impeachment authority to this case, particularly in view of the strong policy favoring settlement. See, e.g., Manko v. United States, 87 F.3d 50, 54 (2d Cir. 1996) ("The primary purpose of Rule 408 is the `promotion of public policy favoring the compromise and settlement of disputes' that would otherwise be discouraged with the admission of such evidence.") (quoting Rule 408, advisory committee's notes). [1] Lansa apparently argued to the district court that an "IBM bundled software" agreement was relevant, but, as the district court found, Lansa "has not cited any part of the record to support its contention that that license is analogous, or even that [its licensing rate] was in fact one to three percent." ResQNet.com, 533 F.Supp.2d at 417. Lansa has not raised this agreement or any other evidence on this appeal. [2] The actual rate for this license is subject to a protective order. [3] Lansa's specific profit margins are subject to a protective order.