578 F.3d 148 (2009)
Docket No. 07-2576-cv.
United States Court of Appeals, Second Circuit.
Argued: March 17, 2009.
Decided: August 21, 2009.
 Hadrian R. Katz (Sarah M. Brackney, on the brief), Arnold & Porter, LLP, Washington, DC (Robert A. Goodman, on the brief, New York, NY), for Plaintiffs-Appellants.
Michael S. Elkin (Thomas P. Lane, Shari Markowitz Savitt, on the brief), Winston & Strawn LLP, New York, NY, for Defendant-Appellee.
Before: CALABRESI, WESLEY, Circuit Judges, and DRONEY, District Judge.
We are the first federal appellate court called upon to determine whether a webcasting service that provides users with individualized internet radio stations—the content of which can be affected by users' ratings of songs, artists, and albums—is an interactive service within the meaning of  17 U.S.C. § 114(j)(7). If it is an interactive service, the webcasting service would be required to pay individual licensing fees to those copyright holders of the sound recordings of songs the webcasting service plays for its users. If it is not an interactive service, the webcasting service must only pay a statutory licensing fee set by the Copyright Royalty Board. A jury determined that the defendant does not provide an interactive service and therefore is not liable for paying the copyright holders, a group of recording companies, a licensing fee for each individual song. The recording companies appeal claiming that as a matter of law the webcasting service is an interactive service, and alternatively, that the district court's instruction to the jury, as well as its admission and exclusion of certain evidence and testimony, was harmful error. We affirm; the webcasting service is not an interactive service as a matter of law.
On May 24, 2001 Arista Records, LLC, Bad Boy Records, BMG Music, and Zomba Recording LLC (collectively, "BMG") brought suit against Launch Media, Inc. ("Launch") alleging that Launch violated provisions of the Digital Millennium Copyright Act of 1998, Pub.L. 105-304, 112 Stat. 2860 (1998) (the "DMCA"), codified in relevant part in 17 U.S.C. § 114, by willfully infringing sound recording copyrights of BMG from 1999 to 2001. The United States District Court for the Southern District of New York (Owen, J.) denied the parties' cross-motions for dismissal under Federal Rule of Civil Procedure 12(b)(6) and summary judgment. The case was tried before a jury, and after the district court denied BMG's motion for judgment as a matter of law, the jury returned a verdict for Launch.
Launch operates an internet radio website, or "webcasting" service, called LAUNCHcast, which enables a user to create "stations" that play songs that are within a particular genre or similar to a particular artist or song the user selects. BMG holds the copyrights in the sound recordings of some of the songs LAUNCHcast plays for users.
BMG, as a sound recording copyright holder, has no copyright in the general performance of a sound recording, see 17 U.S.C. §§ 106(4), 114(a), but BMG does have the exclusive right "to perform the copyrighted [sound recording] publicly by means of a digital audio transmission," 17 U.S.C. § 106(6). Launch does not dispute that LAUNCHcast provides a digital audio transmission within the definition of § 106(6). See 17 U.S.C. §§ 101, 114(j)(5). BMG has a right to demand that those who perform—i.e., play or broadcast—its copyrighted sound recording pay an individual licensing fee to BMG if the performance of the sound recording occurs through an "interactive service." See 17 U.S.C. § 114(d)(3)(C).
An interactive service is defined as a service "that enables a member of the public to receive a transmission of a program specially created for the recipient, or on request, a transmission of a particular sound recording ..., which is selected by or on behalf of the recipient." Id.  § 114(j)(7). If a digital audio transmission is not an interactive service and its "primary purpose ... is to provide to the public such audio or other entertainment programming," id. § 114(j)(6), the transmitter need only pay a compulsory or statutory licensing fee set by the Copyright Royalty Board made up of Copyright Royalty Judges appointed by the Library of Congress, see id. § 114(f).
At trial, BMG claimed that between November 1999 and May 2001 Launch— through LAUNCHcast—provided an interactive service and therefore was required to obtain individual licenses from BMG to play BMG's sound recordings. To demonstrate how LAUNCHcast functioned, BMG submitted an email from Jeff Boulter, Senior Director of Product Development at Launch, in which Boulter described how LAUNCHcast generated a list of songs using the user's preferences. In turn, Launch submitted a report from Margaret L. Johnson, a computer science professor at Stanford University, which also detailed how LAUNCHcast generated songs for a user. There is no material dispute between the parties with regard to how LAUNCHcast works. The jury returned a verdict in favor of Launch.
BMG appeals the district court's denial of BMG's motions for dismissal under 12(b)(6), summary judgment after discovery, and judgment as a matter of law before the jury verdict, arguing that LAUNCHcast is an interactive service as a matter of law because LAUNCHcast is "designed and operated to enable members of the public to receive transmissions of programs specially created for them." BMG claims that under the DMCA there is no tipping point for the level of influence a user must assert before the program becomes an interactive service—all that matters is that the alleged copyright infringer is "transmi[tting] ... a program specially created for" the user. Alternatively, BMG claims the district court's jury instruction was error in several respects.
The district court charged the jury with determining whether LAUNCHcast was an interactive service within the meaning of § 114(j)(7). The district court indicated that it was for the jury "to decide how much influence a consumer or a recipient can have on the programming offered by the transmitting entity—... the broadcaster—before that activity must be characterized as interactive, keeping in mind the purpose of the statute." On two separate occasions the district court noted that "there is no bright line marking the limits between an interactive service and a noninteractive service." These instructions may have misled the jury into believing that it was the jury's responsibility to determine the legal definition of interactive, a task far beyond the scope of its duty. See LNC Invs., Inc. v. First Fid. Bank, N.A. N.J., 173 F.3d 454, 467-68 (2d Cir. 1999).
But we need not determine the propriety of the instruction because we agree with BMG that the central issue of this case—interactivity—presents an issue of  law. The parties do not materially disagree on how LAUNCHcast works; their point of conflict centers on whether the program is "interactive" as defined by the statute—clearly an issue of law and therefore strictly under the purview of the courts. See United States v. Nolan, 136 F.3d 265, 271 (2d Cir.1998).
An "interactive service" according to the statute "is one that enables a member of the public to receive a transmission of a program specially created for the recipient, or on request, a transmission of a particular sound recording, whether or not as part of a program, which is selected by or on behalf of the recipient." 17 U.S.C. § 114(j)(7). The statute provides little guidance as to the meaning of its operative term "specially created." According to Merriam-Webster's Collegiate Dictionary, "specially" means: (1) "in a special manner"; (2) "for a special purpose"; (3) "in particular" or "specifically"; or (4) "especially." Create, the root of "created," means: (1) "to bring into existence"; (2) "to produce"; (3) to "cause" or "occasion"; or (4) to "design."
These definitions indicate that a "specially created" program is a program produced or designed specifically for the user—not much help defining the terms of the statute in this case. BMG sees the issue as a simple one. BMG argues that any service that reflects user input is specially created for and by the user and therefore qualifies as an interactive service. But we should not read the statute so broadly. Justice Oliver Wendell Holmes once wrote that "[a] word is not a crystal, transparent and unchanged, it is the skin of a living thought and may vary greatly in color and content according to the circumstances and the time in which it is used." Towne v. Eisner, 245 U.S. 418, 425, 38 S.Ct. 158, 62 L.Ed. 372 (1918) (internal citation omitted). Holmes's observation seems pertinent here. The meaning of the phrase in question must significantly depend on the context in which Congress chose to employ it.
Congress extended the first copyright protection for sound recordings in 1971 by creating a right "[t]o reproduce and distribute" "tangible" copies of sound recordings. Sound Recording Act of 1971 (the "SRA"), Pub.L. 92-140, 85 Stat. 391; see also Bonneville Int'l Corp. v. Peters, 347 F.3d 485, 487-89 (3d Cir.2003) (detailing the history and evolution of the sound recording copyright). Congress drafted the SRA to address its concern about preventing "phonorecord piracy due to advances in duplicating technology." H.R.Rep. No. 104-274, at 11 (1995) (summarizing the history of sound recording copyright). Notably, unlike the copyright of musical works, the sound recording copyright created by the SRA did not include a right of performance. See Digital Performance Right in Sound Recordings Act of 1995, Pub.L. No. 104-39, 109 Stat. 336, codified as 17 U.S.C. § 106. Therefore, holders of sound recording copyrights—principally recording companies such as BMG—had no right to extract licensing fees from radio stations and other broadcasters of recorded music. The reason for this lack of copyright protection in sound recordings, as the Third Circuit has put it, was that the "recording industry and [radio] broadcasters existed in a sort of symbiotic relationship wherein the recording industry recognized that radio airplay was free advertising that lured consumers to retail stores where they would purchase recordings." Bonneville  Int'l Corp., 347 F.3d at 487. As the Bonneville court also noted, however, the relationship has been, and continues to be, "more nuanced" and occasionally antagonistic. Id. at 488 n. 3. In response to continued lobbying by the recording industry, Congress and the Copyright Office (the "Office") studied the need for stronger copyright protection for sound recordings for two decades after passage of the SRA. See 141 CONG. REC. S11,945-04, 11,949 (daily ed. Aug. 8, 1995) (statement of Sen. Hatch).
With the inception and public use of the internet in the early 1990s, the recording industry became concerned that existing copyright law was insufficient to protect the industry from music piracy. At the time, the United States Register of Copyrights referred to the internet as "the world's biggest copying machine." Stephen Summer, Music on the Internet: Can the Present Laws and Treaties Protect Music Copyright in Cyberspace?, 8 CURRENTS: INT' L. TRADE L.J. 31, 32 (1999). What made copying music transmitted over the internet more dangerous to recording companies than traditional analog copying with a tape recorder was the fact that there is far less degradation of sound quality in a digital recording than an analog recording. See id. Although data transmission over the internet was slow— in 1994 it took on average twenty minutes to download one song—the recording industry foresaw the internet as a threat to the industry's business model. See Judy Holland, Music Industry is Encouraged, STATES NEWS SERV., Jul. 24, 1994. If an internet user could listen to music broadcast over, or downloaded from, the internet for free, the recording industry worried that the user would stop purchasing music. Jason Berman, president of the Recording Industry Association of America (the "RIAA"), the lobbying arm of the recording industry, stated in 1994 that without a copyright in a right of performance via internet technology, the industry would be "unable to compete in this emerging digital era." Jube Shiver, Jr., Digital Double Trouble, L.A. TIMES (Apr. 11, 1994). Berman warned that "digital delivery would siphon off and eventually eliminate the major source of revenue for investing in future recordings" and that "[o]ver time, this [would] lead to a vast reduction in the production of recorded music." Digital Performance Right in Sound Recording: Hearing on H.R. 1506 Before the H. Comm. on the Judiciary, Subcomm. on Courts & Intellectual Prop., 104th Cong. (1995) (statement of Jason Berman, President, RIAA).
The Commerce Department, which oversees the Copyright Office, recognized that failing to provide more protections to the recording industry would drive the music industry away from using the internet as a medium for legitimate music distribution. As then-Assistant Secretary of Commerce Bruce Lehman said, "If we don't make it clear you can't transmit a work without the permission of the owner, copyright owners aren't going to use the information highway.... There won't be any cars on the highway if we don't make certain the cars don't get hijacked at the entrance ramp." Teresa Riordan, Writing Copyright Law For an Information Age, N.Y. TIMES (Jul. 7, 1994) (internal quotation marks omitted). In a white paper to President Clinton, an information infrastructure task force chaired by Lehman recommended that Congress give sound recording copyright holders a right of performance. Bruce A. Lehman, Intellectual Property and the National Information Infrastructure: The Report of the Working Group on Intellectual Property Rights, INFORMATION INFRASTRUCTURE TASK FORCE 221-25 (1995).
In light of these concerns, and recognizing that "digital transmission of sound recordings  [were] likely to become a very important outlet for the performance of recorded music," Congress enacted the Digital Performance Right in Sound Recordings Act of 1995 (the "DPSR"), giving sound recording copyright holders an exclusive but "narrow" right to perform— play or broadcast—sound recordings via a digital audio transmission. H.R.Rep. No. 104-274, at 12, 13-14. The right was limited to exclusive performance of digital audio transmissions through paid subscriptions services and "interactive services." See 17 U.S.C. § 114(d) (1995). While non-interactive subscription services qualified for statutory licensing, interactive services were required to obtain individual licenses for each sound recording those interactive services played via a digital transmission. See Lydia Pallas Loren, Untangling the Web of Music Copyrights, 53 CASE W. RES. L.REV. 673, 692 (2003). Under the DPSR, interactive service was defined as
one that enables a member of the public to receive, on request, a transmission of a particular sound recording chosen by or on behalf of the recipient. The ability of individuals to request that particular sound recordings be performed for reception by the public at large does not make a service interactive. If an entity offers both interactive and non-interactive services (either concurrently or at different times), the non-interactive component shall not be treated as part of an interactive service.
17 U.S.C. § 114(j)(4) (1995).
The House report noted that the DPSR was enacted to address two related concerns. First, without "appropriate copyright protection in the digital environment, the creation of new sound recordings and musical works could be discouraged, ultimately denying the public some of the potential benefits of the new digital transmission technologies." H.R.Rep. No. 104-274, at 13. Second, "certain types of subscription and interaction audio services might adversely affect sales of sound recordings and erode copyright owners' ability to control and be paid for use of their work." Id. With regard to the latter concern, the House noted that "interactive services are most likely to have a significant impact on traditional record sales, and therefore pose the greatest threat to the livelihoods of those whose income depends upon revenues derived from traditional record sales." Id. at 14. As one article has noted, these interactive services were likely to have an impact on record sales "because the more advance information the user has about the digital transmission, the more the transmission facilitate[d] a user's private copying (in perfect digital copies) of the recorded performance, or, at least, enable[d] the user to substitute listening to the targeted performance for purchasing a copy of it." Jane C. Ginsburg, Copyright Legislation for the "Digital Millennium," 23 COLUM.-VLA J.L. & ARTS 137, 167 (1999) (emphasis added).
The House also noted the importance of striking a balance between, on one hand, protecting sound recording copyright holders to promote sales, distribution, and development of new music, and, on the other hand, making development of new media and forms of distribution "economically feasible." H.R.Rep. No. 104-274, at 14. Congress attempted to address this concern by making the right "narrow"—limiting it to performance of digital audio transmissions and exempting nonsubscriber services. See id. The Senate Report noted that it had similar concerns. See S. Rep. 104-128, at 13-17 (1995), reprinted in 1995 U.S.C.C.A.N. 356, 360-64.
Fairly soon after Congress enacted the DPSR, critics began to call for further legislation, charging that the DPSR was  too narrowly drawn and did not sufficiently protect sound recording copyright holders from further internet piracy. See Summer, 8 CURRENTS: INT'L TRADE L.J. at 36; see also Public Performance of Sound Recordings: Definition of Service, 65 Fed. Reg. 77330, 77,331 (letter of the Copyright Office noting that the digital transmission of sound recording "license was amended in 1998 in response to the rapid growth of digital communications networks, e.g., the Internet, and the confusion surrounding the question of how the DP[S]R applied to certain nonsubscription digital audio services"). For instance, webcasting services, which provide free—i.e., nonsubscription—services that do not provide particular sound recording on request and are therefore not interactive within the meaning of term under the DPSR, at that time fell outside the sound recording copyright holder's right of control. See Ginsburg, 23 COLUM.-VLA J.L. & ARTS at 167. Recording companies became concerned that these webcasting services were allowing users to copy music transmitted to their computer via webcast for free, see Note, June Chung, The Digital Performance Right in Sound Recording Act and Its Failure to Address the Issue of Digital Music's New Form of Distribution, 39 ARIZ. L.REV. 1361, 1367 (1997) (explaining how a webcasting user could record and copy webcasted music onto the user's computer), or to listen to these webcasting services in lieu of purchasing music, see Comment, Kimberly L. Craft, The Webcasting Music Revolution Is Ready to Begin, as Soon as We Figure Out the Copyright Law: The Story of the Music Industry at War with Itself, 24 HASTINGS COMM. & ENT. L.J. 1, 12-13 (2001). Record companies were concerned that these webcasting services were causing a diminution in record sales, which the companies feared would cut into profits and stunt development of the recording industry. See Craft, 24 HASTINGS COMM. & ENT. L.J. at 12-13. According to Cary Sherman, Senior Executive Vice President and General Counsel of the RIAA, by 1997, the record industry was losing $1 million a day due to music piracy. Copyright Piracy in the Internet: Hearing on H.R. 2265 Before H. Comm. on the Judiciary, Subcomm. on Courts & Intellectual Prop., 105th Cong. (1997) (statement of Cary H. Sherman, Senior Executive Vice President and General Counsel of the RIAA).
In light of these concerns, Congress enacted the current version of § 114 under the DMCA in 1998. The term "interactive service" was expanded to include "those that are specially created for a particular individual." H.R.Rep. No. 105-796, at 87 (1998) (Conf.Rep.). As enacted, the definition of "interactive service" was now a service "that enables a member of the public to receive a transmission of a program specially created for the recipient, or on request, a transmission of a particular sound recording, whether or not as part of a program, which is selected by or on behalf of the recipient." 17 U.S.C. § 114(j)(7).
According to the House conference report,
The conferees intend that the phrase "program specially created for the recipient" be interpreted reasonably in light of the remainder of the definition of "interactive service." For example, a service would be interactive if it allowed a small number of individuals to request that sound recordings be performed in a program specially created for that group and not available to any individuals outside of that group. In contrast, a service would not be interactive if it merely transmitted to a large number of recipients of the service's transmissions a program consisting of sound recordings requested  by a small number of those listeners.
H.R.Rep. No. 105-796, at 87-88 (Conf. Rep.).
The House report continued that a transmission is considered interactive "if a transmission recipient is permitted to select particular sound recordings in a prerecorded or predetermined program." Id. at 88. "For example, if a transmission recipient has the ability to move forward and backward between songs in a program, the transmission is interactive. It is not necessary that the transmission recipient be able to select the actual songs that comprise the program." Id.
On April 17, 2000, the Digital Media Association ("DiMA"), the lobbying arm of transmitters of digital media such as Launch, filed a petition with the Copyright Office requesting that the Office amend the definition of "service" "to state that a service is not interactive simply because it offers the consumer some degree of influence over the programming offered by the webcaster." DiMA asked the Copyright Office to adopt the following as a rule:
A Service making transmissions that otherwise meet the requirements for the section 114(f) statutory license is not rendered "interactive," and thus ineligible for the statutory license, simply because the consumer may express preferences to such Service as to the musical genres, artists and sound recordings that may be incorporated into the Service's music programming to the public.
The DiMA did not ask the Copyright Office to determine whether any particular services was non-interactive.
On November 21, 2000, the Copyright Office issued a letter stating that because "of the rapidly changing business models emerging in today's digital marketplace, no rule can accurately draw the line demarcating the limits between an interactive service and a noninteractive service. Nor can one readily classify an entity which makes transmissions as exclusively interactive or noninteractive." The Copyright Office stated that determinations of interactivity "must be made on a case-by-case basis after the development of a full evidentiary record."
Despite refusing to make a broad rule governing interactivity, the Copyright Office did opine that in Page 20 of 42 enacting the § 114(j)(7), "Congress sought to identify a service as interactive according to the amount of influence a member of the public would have on the selection and performance of a particular sound recording." However, the Office stated that "the fact that some degree of consumer influence on a service's programming is permissible does not mean that a regulation to clarify that fact is necessary or even desirable." The Copyright Office also noted that "because the law and the accompanying legislative history make it clear that consumers can have some influence on the offerings made by a service without making the service interactive, there is no need to amend the regulations to make this point."
In a footnote, the Copyright Office noted that the RIAA and DiMA had discussed LAUNCHcast "to illustrate the type of offerings that are in dispute." The Copyright Office concluded that "[f]rom these descriptions, there is considerable doubt whether [LAUNCHcast] would qualify as an `interactive service.'" On December 8, 2000, the Copyright Office posted a message on its website that there had been a typographical error in its letter. It then reissued its letter with identical substantive footnote to read: "From these des considerable doubt whether [LAUNCHcast] would qualify as a noninteractive service." In the published version of the letter, however, the Copyright Office again  altered the footnote to read as it had read in the November 21, 2000 letter—noting that there was considerable doubt that LAUNCHcast was interactive. Public Performance of Sound Recordings, 65 Fed. Reg. at 77332.
In sum, from the SRA to the DMCA, Congress enacted copyright legislation directed at preventing the diminution in record sales through outright piracy of music or new digital media that offered listeners the ability to select music in such a way that they would forego purchasing records.
Armed with the statute's text and context, we must examine the complex nature of the service LAUNCHcast provided. Throughout the life of this case, the parties have expressed little disagreement as to how LAUNCHcast operates, however, the conclusions they reach as to whether the statute applies to LAUNCHcast are, not surprisingly, radically different.
After creating a username and password, and entering basic information and preferences unrelated to the music LAUNCHcast provides, a LAUNCHcast user is able to create and modify personalized radio stations. First, the user is prompted to select artists whose music the user prefers. The user is then asked which music genres the user enjoys and asked to rate the genres on a scale. The user is also asked the percentage of new music—songs the user has not previously rated—the user would like to incorporate into the user's station (the "unrated quota") and whether the user permits playing songs with profane lyrics. The minimum unrated quota is 20%, meaning no less than 20% of the songs played can be unrated.
Once LAUNCHcast begins playing music based on the user's preferred artists and genres, the user rates the songs, artists, or albums LAUNCHcast plays between zero and 100, with 100 being the best rating. Below the rating field are hyperlinks termed "history," "share," and "buy." The history hyperlink allows the user to see a list of the songs previously played, and the buy hyperlink facilitates the user's purchase of the songs. The share hyperlink allows the user to share the station with other users. This feature facilitates the "subscription" of one user to another user's station. When user A subscribes to the station of user B, user B becomes a "DJ" for user A. The DJ feature  of LAUNCHcast does not allow a user to play particular songs for other users, but instead gives users access to each other's stations, which they can modify by rating songs, artists, or albums. These modifications only affect what songs the user hears on the DJ's station and do not affect the songs the DJ hears on the DJ's station, nor do the modifications affect the DJ's personal modification of the DJ's station. While a song is playing, the user has the ability to pause the song, skip the song, or delete the song from the station by rating the song zero. Notably, the user may not go back to restart the song that is playing, or repeat any of the previously played songs in the playlist.
Whenever the user logs into LAUNCHcast and selects a station, LAUNCHcast generates a playlist of fifty songs based on several variables. LAUNCHcast does not provide a list of the pool of songs or of the songs in the generated playlist, and therefore, the user does not know what songs might be played. LAUNCHcast selects the songs by first looking to the unrated quota and whether to exclude songs with profane lyrics or songs that cannot be transmitted over the user's bandwidth. Next LAUNCHcast creates a list of all the potential songs that can be put in the playlist (called a "hashtable"). LAUNCHcast then generates a list of all songs played for the user within the last thirty days, a list of all DJs, genres, and radio stations to which the user subscribes, and a list of all the ratings of all the songs, artists, and albums rated by either the user or any DJ to which the user subscribes. Songs that the user has rated are "explicitly rated" songs. LAUNCHcast "implicitly rates" songs that appear in an album that the user or a subscribed-to DJ has rated and songs that appear in the same album as another song the user has already rated. All of these songs are initially added to the hashtable. LAUNCHcast then excludes: (1) all songs that the user, or a DJ to which the user subscribes, requests be skipped permanently (rated as zero) and (2) songs played within the last three hours for the user on any LAUNCHcast station. This yields approximately 4,000 songs.
LAUNCHcast then adds to the hashtable the 1,000 most popular songs—songs most highly rated by all LAUNCHcast users—in the bandwidth specified by the user, provided those songs are not already on the hashtable. LAUNCHcast then adds another 5,000 songs. To generate this group of songs, LAUNCHcast first counts the total number of songs contained in each of all the genres the user has selected and divides that number by the total number of songs in LAUNCHcast's database. If the resulting quotient is less than 5% of the number of songs in LAUNCHcast's database then LAUNCHcast picks only songs listed as within the  genres the user has selected. This calculation is performed in order to ensure that of the 5,000 "random" songs added to the hashtable, a sufficiently large number are of genres eligible to be selected for inclusion on the final playlist. That means that when a user has selected a very small number of genres, the selection of songs of those genres from the hashtable would return only a few songs allowable for play.
If the quotient of total songs in the user's selected genres is greater than 5%, however, a sufficiently large number of the 5,000 randomly chosen songs added to the hashtable will be eligible for inclusion on the final playlist such that limiting the selection of playlist songs by genre will not perceptively decrease the randomness of the songs actually played; therefore, LAUNCHcast picks the 5,000 songs randomly from its entire database, rather than solely from the user's selected genres.
At this point, the hashtable contains approximately 10,000 songs. All of the songs in the hashtable are then sorted according to rating: (1) explicitly rated; (2) implicitly rated; or (3) unrated. Based on these categories, LAUNCHcast determines which songs will be played from each category based on several criteria. First, if the user's list of explicitly and implicitly rated songs is smaller than 100, 90% of the songs LAUNCHcast selects for the playlist are unrated. If the user's list of explicitly and implicitly rated songs is greater than 100, LAUNCHcast uses the unrated quota provided by the user—the minimum unrated quota the user can choose being 20%. Second, no more than 20% of all explicitly rated songs are selected from the hashtable for the playlist. In other words, if a user has only rated ten songs, no more than two of those songs can be selected for the playlist. Third, LAUNCHcast selects no more than three times the quotient of the total number of explicitly rated songs divided by the sum of implicitly and explicitly rated songs. In other words, if a user rates five songs and ten others are implicitly rated— making a total of fifteen songs—then no more than one explicitly rated song can be selected (3 × (5/[5 + 10]) = 3 × 1/3 = 1). Fourth, if the user is listening to the station of a DJ and the total number of explicitly and implicitly rated songs is greater than 200, then the total number of explicit songs LAUNCHcast selects will be the lesser of 20% of all explicit songs or half of the songs to be selected in the playlist. In other words, if the user, who has explicitly rated 150 songs and as a result has fifty-one implicitly rated songs (totaling 201), selects the station of a DJ, twenty-five explicitly rated songs will be selected for the playlist from the DJ's station because half of the songs in the  playlist of fifty (twenty-five) is less than 20% of the number of explicitly rated song (20% of 150 = 30).
Next, after determining what songs can be selected from the hashtable, LAUNCHcast picks songs from the hashtable to add to the playlist. To do this, LAUNCHcast first selects a song at random from one of the three categories, explicitly rated, implicitly rated, or unrated, but with some restrictions. First, songs are excluded from the playlist if including them would violate the criteria listed above. Second, LAUNCHcast does not play the same song twice in a playlist. Third, LAUNCHcast excludes a song from a playlist if three other songs by that artist have already been selected for the playlist. In other words, if the Beatles' "Here Comes the Sun," "A Day in the Life," and "Eleanor Rigby" have already been selected for the playlist, no other Beatles' song could be added to the playlist. Fourth, LAUNCHcast excludes a song from a playlist if two other songs from the same album have already been selected for the playlist. Fifth, LAUNCHcast has an additional layer to prevent songs by the same artist or from the same album from playing on the playlist. If a song being reviewed for selection for the playlist is by an artist already included in the playlist or from the same album as the previous song selected for the playlist, that song is excluded from the playlist unless—at the end of the process—a user has a playlist with less than 50 songs. Although selection is random, according to Boulter the algorithm used to select songs for the playlist "[is] biased towards the top" of the list— i.e., choosing more highly rated songs— "but would pick randomly from there. So maybe the first time it picked No. 2 and then it picked 37. Then maybe it picked one and then it picked 300." In her report, Johnson stated that "a random song from that particular list of songs is chosen... such that there is a high probability that the song is picked from the higher scored songs in the list."
Finally, once all fifty songs are selected for the playlist, LAUNCHcast orders the playlist. The ordering of the songs is random, provided LAUNCHcast does not play more than two songs in the same album or three songs by the same artist consecutively.
It is hard to think of a more complicated way to "select songs," but this is the nature of webcast music broadcasting in the digital age. Given LAUNCHcast's format, we turn to the question of whether LAUNCHcast is an interactive service as a matter of law. As we have already noted,  a webcasting service such as LAUNCHcast is interactive under the statute if a user can either (1) request— and have played—a particular sound recording, or (2) receive a transmission of a program "specially created" for the user. 42 U.S.C. § 114(j)(7). A LAUNCHCAST user cannot request and expect to hear a particular song on demand; therefore, LAUNCHcast does not meet the first definition of interactive. But LAUNCHcast may still be liable if it enables the user to receive a transmission of a program "specially created" for the user. It comes as no surprise to us that the district court, the parties, and others have struggled with what Congress meant by this term.
The language and development of the DPSR and DMCA make clear that Congress enacted both statutes to create a narrow copyright in the performance of digital audio transmissions to protect sound recording copyright holders—principally recording companies—from the diminution in record sales. Congress created this narrow right to ensure that "the creation of new sound recordings and musical works [would not] be discouraged," and to prevent the "threat to the livelihoods of those whose income depends upon revenues derived from traditional record sales." H.R.Rep. No. 104-274, at 13-14.
Contrary to BMG's contentions, Congress was clear that the statute sought to prevent further decreases in revenues for sound recording copyright holders due to significant reductions in record sales, perceived in turn to be a result of the proliferation of interactive listening services. If the user has sufficient control over the interactive service such that she can predict the songs she will hear, much as she would if she owned the music herself and could play each song at will, she would have no need to purchase the music she wishes to hear. Therefore, part and parcel of the concern about a diminution in record sales is the concern that an interactive service provides a degree of predictability—based on choices made by the user—that approximates the predictability the music listener seeks when purchasing music.
The current version § 114(j)(7) was enacted because Congress determined that the DPSR was not up to the task of protecting sound recording copyright holders from diminution in record sales, presumably because programs not covered by the DPSR's definition of interactive service provided a degree of control—predictability—to internet music listeners that dampened the music listeners' need to purchase music recordings. By giving sound recording copyright holders the right to require individual licenses for transmissions of programs specially created for users, Congress hoped to plug the loophole the  DPSR had left open for webcasting services.
To determine what additional service or program Congress intended to include as an interactive service under the DMCA, we look to what language was added to the definition of interactive service. One term notably absent from the DPSR definition, which was later included under the DMCA, is "transmission of a program." 17 U.S.C. § 114(j)(7). While not semantically identical, Title 17 defines "transmission program" as "a body of material that, as an aggregate, has been produced for the sole purpose of transmission to the public in sequence and as a unit." 17 U.S.C. § 101. This definition views a transmission program as a body of material presented as a single unit, as opposed to a selection of individual works. Therefore, in expanding the definition of interactive service to include the transmission of programs specially created for the user, Congress intended to include bodies of prepackaged material, such as groups of songs or playlists specially created for the user. Given this definition, we turn to the ultimate issue of whether the LAUNCHcast playlists, uniquely generated for the user each time the user selects a station, are specially created and therefore interactive.
Launch does not deny that each playlist generated when a LAUNCHcast user selects a radio station is unique to that user at that particular time. However, this does not necessarily make the LAUNCHcast playlist specially created for the user. Based on a review of how LAUNCHcast functions, it is clear that LAUNCHcast does not provide a specially created program within the meaning of § 114(j)(7) because the webcasting service does not provide sufficient control to users such that playlists are so predictable that users will choose to listen to the webcast in lieu of purchasing music, thereby—in the aggregate—diminishing record sales.
First, the rules governing what songs are pooled in the hashtable ensure that the user has almost no ability to choose, let alone predict, which specific songs will be pooled in anticipation for selection to the playlist. At least 60% of the songs in the hashtable are generated by factors almost entirely beyond the user's control. The playlist—a total of fifty songs—is created from a pool of approximately 10,000 songs, at least 6,000 of which (1,000 of the most highly rated LAUNCHcast songs among all users and 5,000 randomly selected songs) are selected without any consideration for the user's song, artist, or album preferences. The user has control over  the genre of songs to be played for 5,000 songs, but this degree of control is no different from a traditional radio listener expressing a preference for a country music station over a classic rock station. LAUNCHcast generates this list with safeguards to prevent the user from limiting the number of songs in the list eligible for play by selecting a narrow genre. Also, no more than 20% of the songs the user rates—marked by LAUNCHcast as explicitly rated—can be pooled in the hashtable, and no more than three times the number of explicitly rated songs divided by the total number of rated songs can be in the hashtable. This ensures that a limited number of explicitly rated songs will eventually be selected for the playlist. Ironically, this effectively means that the more songs the user explicitly rates, the less the user can predict which explicitly rated songs will be pooled in the hashtable and played on the playlist.
Second, the selection of songs from the hashtable to be included in the playlist is governed by rules preventing the user's explicitly rated songs from being anywhere near a majority of the songs on the playlist. At minimum, 20% of the songs played on the station are unrated—meaning the user has never expressed a preference for those songs. If the user attempts to increase her chances of hearing a particular song by rating only a small number of songs—making the user's list of explicitly and implicitly rated songs smaller than 100-90% of the songs LAUNCHcast selects for the playlist will be unrated, flooding the playlist with songs for which the user has never expressed a preference.
Even the ways in which songs are rated include variables beyond the user's control. For instance, the ratings by all of the user's subscribed-to DJs are included in the playlist selection process. When the user rates a particular song, LAUNCHcast then implicitly rates all other songs by that artist, subjecting the user to many songs the user may have never heard or does not even like. There are restrictions placed on the number of times songs by a particular artist or from a particular album can be played, and there are restrictions on consecutive play of the same artist or album. Finally, because each playlist is unique to each user each time the user logs in, a user cannot listen to the playlist of another user and anticipate the songs to be played from that playlist, even if the user has selected the same preferences and rated all songs, artists, and albums identically as the other user. Relatedly, a user who hears a song she likes and wants to hear again cannot do so by logging off and back on to reset her station to disable the restriction against playing the same song twice on a playlist. Even if a user logs off LAUNCHcast then logs back on and selects the same station, the user will still hear the remainder of the playlist to which she had previously been listening with its restrictions still in operation, provided there were at least eight songs left to be played on the playlist—or, in other words, until the user listens to at least forty-two of the playlist's songs.
 Finally, after navigating these criteria to pool a hashtable and generate a playlist, LAUNCHcast randomly orders the playlist. This randomization is limited by restrictions on the consecutive play of artists or albums, which further restricts the user's ability to choose the artists or albums they wish to hear. LAUNCHcast also does not enable the user to view the unplayed songs in the playlist, ensuring that a user cannot sift through a playlist to choose the songs the user wishes to hear.
It appears the only thing a user can predict with certainty—the only thing the user can control—is that by rating a song at zero the user will not hear that song on that station again. But the ability not to listen to a particular song is certainly not a violation of a copyright holder's right to be compensated when the sound recording is played.
In short, to the degree that LAUNCHcast's playlists are uniquely created for each user, that feature does not ensure predictability. Indeed, the unique nature of the playlist helps Launch ensure that it does not provide a service so specially created for the user that the user ceases to purchase music. LAUNCHcast listeners do not even enjoy the limited predictability that once graced the AM airwaves on weekends in America when "special requests" represented love-struck adolescents' attempts to communicate their feelings to "that special friend." Therefore, we cannot say LAUNCHcast falls within the scope of the DMCA's definition of an interactive service created for individual users.
When Congress created the sound recording copyright, it explicitly characterized it as "narrow." There is no general right of performance in the sound recording copyright. There is only a limited right to performance of digital audio transmission with several exceptions to the copyright, including the one at issue in this case. We find that LAUNCHcast is not an interactive service within the meaning of 17 U.S.C. § 114(j)(7). Because we so conclude, we need not reach the other issues raised by BMG on appeal.
The district court's judgment of May 16, 2007 in favor of Appellee is hereby AFFIRMED with costs.
 The Honorable Christopher F. Droney, Judge United States District Court for the District of Connecticut, sitting by designation.
 The recording companies also ask us to reach an issue of damages calculation and to reassign the case to another district court judge. Because we affirm on the issue of law we need not reach these issues.
 Plaintiffs-Appellants appeal the judgment of the United States District Court for the Southern District of New York (Owen, J.), entered on May 16, 2007, and from interlocutory orders merged into the judgment, finding in favor of Defendant-Appellee Launch Media, Inc., now owned by Yahoo!, Inc. ("Yahoo").
 Prior to 2004, parties were required to submit their claims for statutory licensing fees to Copyright Arbitration Royalty Panels. This system was phased out by the Copyright Royalty and Distribution Reform Act of 2004, Pub.L. 108-419, 118 Stat. 2341 (2004), codified in relevant part at 17 U.S.C. § 114(f).
 The substance of Boulter and Johnson's accounts of LAUNCHcast operation are provided in greater detail infra.
 BMG has argued as much by moving for dismissal under 12(b)(6), summary judgment, and for judgment as a matter of law on the issue of interactivity and appealing the denial of these motions to this Court.
 In a memorandum from Kenneth L. Steinthal, attorney for Launch, admitted into evidence at trial, Steinthal stated that he spoke with the individual in the Copyright Office who drafted the footnote. Steinthal stated that according to that individual, someone from the RIAA had called the Copyright Office and as a result, the substance of the footnote was changed. According to Steinthal, the individual told him that the footnote "had been intended to send a message ... that ... it would be a `long shot' for a service such as Launchcast ... to be considered non-interactive." Whatever the etiology of the Copyright Office's inability to make up its mind, we find the Copyright Office efforts here of little help.
 For the following analysis of LAUNCHcast's service we rely on Plaintiffs' Exhibits 260 and 262, in which Boulter describes how LAUNCHcast operated, as well as the report submitted by Johnson offering essentially the same analysis.
 Federal judges are appointed for life. U.S. CONST. art. III, § 1. Our familiarity with the ever-changing terms and technology of the digital age is, to say the least, varied. We have attempted to portray the processes and procedures of LAUNCHcast in lay terms, understandable to ourselves and the public.
 LAUNCHcast disregards the unrated quota selected by the user if the number of rated songs LAUNCHcast uses to generate a playlist for the user is less than 100. This is detailed further infra.
 If there are more than eight songs left unheard on the playlist previously begun by the user during the last session, the playlist is less than a week old, and fewer than fifteen songs of the playlist that have been played have been rated by the user for the first time, then LAUNCHcast will begin the old playlist from the first unheard song.
 Ratings for individual songs are modified based on the ratings the user (or the DJ to which the user subscribes) has given the artist performing the song or the album on which the song appears. The scope of the modification is irrelevant for purposes of this opinion.
 In other words, if the user rates the band U2 or U2's album The Joshua Tree, then U2's song "With or Without You" on the album will be implicitly rated. Likewise, if the user rates Gordon Lightfoot's song "Rainy Day People" but does not rate Lightfoot's iconic song "Canadian Railroad Trilogy"—found on a different album—"Rainy Day People" will be explicitly rated while "Canadian Railroad Trilogy" will be implicitly rated.
 For example, if the user selects five genres and LAUNCHcast has 1,000 songs listed under each of those genres, with no songs being counted twice for being in multiple genres, and 150,000 songs in its database (the total number of songs in LAUNCHcast's database at that time according to Boulter), then the total number of songs in a selected genre is 5,000. Because 5,000 is 3.3% of 150,000, LAUNCHcast would choose the 5,000 songs from this pool.
 The record is unclear whether the unrated quota—if not selected by the user—was fixed at 50% or 20%. See Plaintiffs' Exhibits 260, 447. However, for our purposes the dispute is irrelevant because in determining whether LAUNCHcast is an interactive service we consider the particular aspect of LAUNCHcast that is the most interactive, or in other words, the aspect that provides the user with the greatest possible amount of influence on the outcome of the LAUNCHcast playlist. Therefore, we assume a user has a 20% unrated quota.
 For purposes of describing how LAUNCHcast functions, Boulter "[a]ssume[s] the size of a playlist is 50 songs." We also assume this.
 This could occur independent of the second rule if the album is a compilation of multiple artists.
 This brings LAUNCHcast into compliance with the "sound recording performance complement," which limits webcasters to playing no more than three selections from a given record in a three-hour period. 17 U.S.C. § 114(d)(2)(C)(i), (j)(13).
 While we have on other occasions noted that "administrative agencies have broad discretion to interpret the statutes they are charged with implementing," Khan v. U.S. Dep't of Justice, 494 F.3d 255, 258 (2d Cir. 2007), we will not give weight to the Copyright Office's footnote regarding LAUNCHcast in its December 2000 letter. Notwithstanding the fact that the footnote specifically discusses LAUNCHcast, the footnote is surplusage to the issue before the Office, and the history of its revisions—complete reversals of opinion by the Copyright Office over a matter of days—seriously call into question the thought process, if any, that went into drafting the footnote.
 While file-sharing services like Napster initially caused a decline in records sales, see Anita Hamilton, The Pirates of Prime Time, TIME, Feb. 25, 2002, at 54 (noting that in 2001 CD sales declined for the first time in a decade), recently webcasting services have been credited with "becom[ing] a massive driver in digital [music] sales" by exposing users to new music and providing an easy link to sites where users can purchase this music. Jeb Gottlieb, Pandora Lifts Lid on Personalizing Online Radio, BOSTON HERALD, Feb. 26, 2008, at 32; see also SoundExchange Open to Bill Targeting Small Webcasters, COMMC'NS DAILY, May 3, 2007 (noting that independent musician Mike Holden "enjoyed a `huge increase' in iTunes downloads ... when Pandora[, another webcasting service,] added his music"). The difference between the two types of services likely explains the different effect on record sales. File-sharing services allow users to copy music files to their computer, thereby enabling the user to listen to the music any time. Webcasting services, however, do not allow the user to download files of the music being webcast, and therefore, do not enable music piracy.
 While we decline to give weight to the Copyright Office's analysis in its November and December 2000 letters, it is worth noting that in these letters the Office found that the "RIAA readily acknowledges that consumers may express preferences for certain music genres, artists, or even sound recordings without the service necessarily becoming interactive."
 In its brief, BMG offers evidence of Launch describing LAUNCHcast as "interactive" in its marketing literature. But our task is to determine whether LAUNCHcast was an interactive service as that term is defined in the statute and not how it was marketed to the public.
 This assumes that the word "specially" implies that the program was made by taking some of the user's preferences into consideration. We cannot conclude that a specially created program could include a playlist of completely random songs selected without regard for the user's preferences. First, such a result would make no sense because a unique playlist of random songs would be little different from a radio station streamed over the internet—which at least one Circuit has held only requires that a broadcaster pay a statutory—and not an individual—license. See Bonneville Int'l Corp., 347 F.3d at 487. Second, such a random webcast should have no effect on record sales, and therefore imposing an individual licensing requirement for songs played on a random playlist would be inconsistent with the congressional purpose of combating diminutions in records sales. In its letters of November and December 2000, the Copyright Office noted that the "RIAA acknowledge[d] that all music programming services[, including webcasting services,] are likely to be influenced by their consumers' tastes."
 It would be wrong, for instance, to assume that because a user likes the Beatles' album A Hard Day's Night the user would also like The White Album.
 The hyperlink in LAUNCHcast giving users the option to buy the music being played also cuts against the contention that LAUNCHcast's service diminishes record sales, as does the option to view a list of previously played songs, which would give the user a reference for selecting songs to purchase. See Music and Radio: Hearing before S. Jud. Comm., 110th Cong. (2008) (statement of Joe Kennedy, President and Chief Executive Officer Pandora Media, Inc.) (claiming that "Pandora [is] ... among the top promotional partners of iTunes and Amazon.com" for music).