George Johnson v. CRB (PUBLIC)

• Court: Court of Appeals for the D.C. Circuit
• Decided: August 11, 2020
• Docket: 19-1028
• Status: Published

Opinion

United States Court of Appeals FOR THE DISTRICT OF COLUMBIA CIRCUIT

Argued March 10, 2020 Decided August 7, 2020

No. 19-1028

GEORGE JOHNSON, APPELLANT

v.

COPYRIGHT ROYALTY BOARD AND LIBRARIAN OF CONGRESS, APPELLEES

NASHVILLE SONGWRITERS ASSOCIATION INTERNATIONAL, ET AL., INTERVENORS

Consolidated with 19-1058, 19-1059, 19-1060, 19-1061, 19-1062

On Appeals from a Final Determination of the Copyright Royalty Board

Andrew J. Pincus and Scott H. Angstreich argued the causes for appellants Amazon Digital Services LLC, Google LLC, Pandora Media, LLC, and Spotify USA Inc. With them on the briefs were Benjamin E. Marks, Gregory Silbert, Aaron 2

J. Curtis, Kenneth L. Steinthal, Jeffrey S. Bucholtz, Jason Blake Cunningham, Leslie V. Pope, and A. John P. Mancini.

Harold Feld was on the brief for amicus curiae Public Knowledge in support of appellants Amazon Digital Services LLC, Google LLC, Pandora Media, LLC, and Spotify USA Inc.

Kannon K. Shanmugam argued the cause for appellants National Music Publishers’ Association and Nashville Songwriters Association International. With him on the briefs were William T. Marks, Benjamin E. Moskowitz, Donald S. Zakarin, Frank P. Scibilia, Benjamin K. Semel, and Aaron J. Marks.

George Johnson, appellant appearing pro se, argued the cause and filed the briefs on his own behalf.

Jennifer L. Utrecht, Attorney, U.S. Department of Justice, argued the cause for appellees. With her on the brief was Joseph H. Hunt, Assistant Attorney General at the time the brief was filed, and Daniel Tenny, Attorney. Mark R. Freeman, Attorney, entered an appearance.

Jacqueline C. Charlesworth was on the brief for amici curiae Songwriters of North America, et al. in support of appellees and affirmance.

Kannon K. Shanmugam argued the cause for intervenors National Music Publishers’ Association and Nashville Songwriters Association International. With him on the briefs were William T. Marks, Benjamin E. Moskowitz, Donald S. Zakarin, Frank P. Scibilia, Benjamin K. Semel, and Aaron J. Marks. 3

Andrew J. Pincus argued the cause for intervenors Amazon Digital Services LLC, Google LLC, Pandora Media, LLC, and Spotify USA Inc. With him on the briefs were Benjamin E. Marks, Gregory Silbert, Aaron J. Curtis, Kenneth L. Steinthal, Jeffrey S. Bucholtz, Jason Blake Cunningham, Leslie V. Pope, and A. John P. Mancini.

Harold Feld was on the brief for amicus curiae Public Knowledge in support of intervenors Amazon Digital Services LLC, Google LLC, Pandora Media, LLC, and Spotify USA Inc.

Before: HENDERSON, GARLAND, and MILLETT, Circuit Judges.

Opinion for the Court filed by Circuit Judge MILLETT.

MILLETT, Circuit Judge: The Copyright Act requires the Copyright Royalty Board to undertake, every five years, the difficult task of setting the copyright royalty rates for the rights to reproduce and distribute musical works. See 17 U.S.C. § 115 (2012). These consolidated appeals deal with the royalty rates and terms established by the Board for the period January 1, 2018 through December 31, 2022. 84 Fed. Reg. 1918 (Feb. 5, 2019).

The appellants in this case are (i) four music streaming services—Amazon Digital Services LLC, Google LLC, Pandora Media, LLC, and Spotify USA Inc. (collectively, “Streaming Services”); (ii) the National Music Publishers’ Association and the Nashville Songwriters Association International (collectively, “Copyright Owners”); and (iii) George Johnson, a songwriter proceeding pro se. They disagree on multiple fronts with the Board and with each other. As a result, many issues devolved into Goldilocks’ arguments, 4

with the Streaming Services protesting that the rates are too high; the Copyright Owners objecting that they are too low; and the Copyright Royalty Board saying they are just right.

Having considered all of those arguments and the extensive administrative record, we affirm in part and vacate and remand to the Copyright Royalty Board in part because it failed to give adequate notice or to sufficiently explain critical aspects of its decisionmaking. Specifically, the Board failed to provide adequate notice of the rate structure it adopted, failed to explain its rejection of a past settlement agreement as a benchmark for rates going forward, and never identified the source of its asserted authority to substantively redefine a material term after publishing its Initial Determination.

I

A

The Copyright Act, 17 U.S.C. §§ 101 et seq., grants copyright owners certain legal rights in their copyrighted works. Those rights include the exclusive authority to reproduce, distribute, and perform the copyrighted work, and to allow others to do the same. Id. § 106.

This case deals with two specific types of copyrightable works: musical works and sound recordings. A “musical work” refers to the notes, lyrics, embedded performance directions, and related material composed by the creator of a song. See SoundExchange, Inc. v. Librarian of Congress, 571 F.3d 1220, 1222 (D.C. Cir. 2009). Think the Gershwin Brothers composing “Embraceable You.” 5

A “sound recording,” on the other hand, is a performing artist’s particular recording of a musical work. See Music Choice v. Copyright Royalty Bd., 774 F.3d 1000, 1004 (D.C. Cir. 2014); see also 17 U.S.C. § 101; SoundExchange, 571 F.3d at 1222. Think Billie Holiday’s stirring rendition of that jazz standard. BILLIE HOLIDAY, Embraceable You, on BODY AND SOUL (Verve Label Group 1957).

How those copyrighted works get from the songwriters into your ears is rather complicated. For starters, while “almost always intermingled in a single song, [the musical work and sound recording] copyrights are legally distinct and may be owned and licensed separately.” Recording Indus. Ass’n of America, Inc. v. Librarian of Congress, 608 F.3d 861, 863 (D.C. Cir. 2010).

So when you stream a particular recording of a song from your interactive music streaming service of choice, the service must have first obtained permission to disseminate both the underlying musical work and the specific sound recording. Specifically, such streaming services must acquire licenses to make and distribute copies of the sound recording and the musical work, 17 U.S.C. § 106(1), (3), as well as to publicly perform those copyrighted works, id. § 106(4), (6). In the context of interactive streaming services, the Copyright Royalty Board has the authority to set certain royalty rates for musical works, but not for sound recordings.1

1 An “interactive service” is defined in Section 114 of the Copyright Act 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[.]” 17 U.S.C. § 114(j)(7). Section 114 creates a compulsory licensing scheme for sound recordings, but that license does not extend to interactive services. Id. § 114(d). 6

As relevant here, Section 115 of the Copyright Act creates a compulsory license, which is a statutorily conferred authority to use certain copyrighted material in a specified manner as a matter of law, without the actual consent of the copyright holder. See Independent Producers Group v. Library of Congress, 759 F.3d 100, 101 (D.C. Cir. 2014). The Section 115 license allows any person who satisfies certain conditions, including the payment of a royalty, to reproduce and to distribute phonorecords of a copyrighted musical work. 17 U.S.C. § 115 (2012).2 This is commonly referred to as the “mechanical license.” 84 Fed. Reg. at 1918–1919. The Copyright Act charges the Copyright Royalty Board with setting the royalty rates and terms for the mechanical license. 17 U.S.C. §§ 115, 801(b)(1) (2012).3

Section 115’s compulsory license, however, does not include the right to publicly perform a musical work. See 17 U.S.C. § 115 (2012). Nor does Section 115 create a compulsory license for the sound recordings themselves. See id.; Recording Indus., 608 F.3d at 863. Therefore, interactive 2 The term “phonorecords” refers to “material objects,” such as vinyl records, CDs, or other digital storage devices, “in which sounds, other than those accompanying a motion picture * * *, are fixed * * *, and from which the sounds can be perceived, reproduced, or otherwise communicated, either directly or with the aid of a machine or device.” 17 U.S.C. § 101. 3 The Orrin G. Hatch-Bob Goodlatte Music Modernization Act, Pub. L. No. 115–264, 132 Stat. 3676 (2018), among other things, amended Sections 115 and 801 of the Copyright Act. While some of these changes went into effect immediately, changes in the statutory provisions at issue in this case only govern ratemaking proceedings commenced after the law’s enactment on October 11, 2018. So they do not apply to this decision of the Copyright Royalty Board. Pub. L. No. 115-264, § 102(c), 132 Stat. at 3722, 3725. 7

streaming services seeking the right to make, distribute, or publicly perform a sound recording, and those seeking the right to publicly perform a musical work, must negotiate with and obtain permission from the appropriate rightsholders. See 17 U.S.C. § 106(1), (3), (4), (6).

The Copyright Royalty Board is an “institutional entity in the Library of Congress” that “house[s] the Copyright Royalty Judges.” 37 C.F.R. § 301.1. The three Copyright Royalty Judges are responsible for presiding over royalty proceedings and for making “determinations and adjustments of reasonable terms and rates of royalty payments[.]” 17 U.S.C. § 801(a), (b)(1) (2012).

The Copyright Royalty Board initiates ratemaking proceedings every five years to set the royalty rates and terms associated with the compulsory mechanical license. 17 U.S.C. § 804(b)(4). After the commencement of those proceedings, the Copyright Act gives interested parties an opportunity first to try and settle on the royalty rate. Id. § 803(b)(3). If no settlement emerges, the Board presides over a contested royalty ratemaking proceeding. See id. § 803. Any person that the Board determines has a “significant interest in the proceeding” may participate. See id. § 803(b)(2)(C).

At the time the proceedings at issue here were initiated, the Copyright Act required that the Board prioritize four objectives when developing “reasonable [royalty] rates and terms” for the mechanical license. 17 U.S.C. § 115(c)(3)(C) (2012). Those objectives were:

(A) * * * maximiz[ing] the availability of creative works to the public; 8

(B) * * * afford[ing] the copyright owner a fair return for his or her creative work and the copyright user a fair income under existing economic conditions;

(C) * * * reflect[ing] the relative roles of the copyright owner and the copyright user in the product made available to the public with respect to relative creative contribution, technological contribution, capital investment, cost, risk, and contribution to the opening of new markets for creative expression and media for their communication;

(D) * * * minimiz[ing] any disruptive impact on the structure of the industries involved and on generally prevailing industry practices.

Id. § 801(b)(1).4

At the end of the ratemaking proceeding, the Board releases its initial determination setting the royalty rates and terms for the mechanical license for the licensing period at issue. See 17 U.S.C. § 803(c)(1); id. § 803(c)(2)(B), (E) (describing this determination as the Board’s “initial determination”). That determination must be agreed upon by at least a majority of the three Copyright Royalty Judges. See id. § 803(a)(3); 37 C.F.R. § 352.1.

4 The Orrin G. Hatch-Bob Goodlatte Music Modernization Act eliminated these four requirements for proceedings initiated after October 2018. For future ratemakings, the Board must instead “establish rates and terms that most clearly represent the rates and terms that would have been negotiated in the marketplace between a willing buyer and a willing seller.” See Pub. L. No. 115-264, § 102(a)(3), 132 Stat. at 3680. 9

Only in “exceptional cases” may the Board grant a participant’s request for rehearing after issuance of its initial determination. 17 U.S.C. § 803(c)(2)(A). The Board also maintains jurisdiction to “issue an amendment to a written determination to correct any technical or clerical errors in the determination or to modify the terms, but not the rates, of royalty payments in response to unforeseen circumstances that would frustrate the proper implementation of such determination.” Id. § 803(c)(4).

Once the Copyright Royalty Board issues its final determination, the Register of Copyrights may review the Board’s decision for legal error. 17 U.S.C. § 802(f)(1)(D). Within sixty days of the Board issuing its final determination, the Librarian of Congress must publish that determination in the Federal Register, along with any corrections identified by the Register of Copyrights. See id. §§ 802(f)(1)(D), 803(c)(6).

B

The Board’s two prior mechanical license ratemaking proceedings provide an important backdrop for understanding the disputes at hand.

In 2006, the Board commenced its first mechanical license ratemaking proceeding, which is commonly referred to as Phonorecords I. See 74 Fed. Reg. 4510, 4513–4514 (Jan. 26, 2009). The parties in Phonorecords I reached, and the Board approved, a settlement setting the royalty rates and terms for limited downloads, interactive streaming, and incidental phonorecord deliveries. Id. at 4514–4515. But the Board still had to adjudicate the rates and terms for the mechanical license as applied to permanent downloads and physical phonorecords. Id. at 4510. 10

Five years later, the Board initiated Phonorecords II, its second mechanical license ratemaking proceeding. 78 Fed. Reg. 67,938, 67,939 (Nov. 13, 2013). There, the Board approved a settlement that carried forward the rates from Phonorecords I and set rates for certain new services introduced into the market, such as mixed service bundles and purchased content locker services. Id. at 67,939, 67,942; 84 Fed. Reg. at 1919; see also 37 C.F.R. § 385.20–.26 (2014).5

Under both Phonorecords I and II, the applicable royalty rates depended on the type of service provided. The Phonorecords II settlement involved ten categories of streaming-related offerings, such as standalone portable and non-portable subscription services, bundled subscription services, free non-subscription/advertisement-supported services, and varied locker services. 37 C.F.R. § 385.13(a)(1)– (5), 385.23(a)(1)–(5) (2014).

The precise formula for calculating royalties for each category differed somewhat, but generally included the same four elements:

First, for each category, the service providers calculated the greater of (i) the “revenue prong,” which was a percentage of the service provider’s revenue associated with the particular offering, and (ii) the “total content cost prong,” which was a percentage of the royalties paid by the service provider to

5 Mixed service bundles involve packaging a streaming music service with other products or services, like Amazon Prime, and offering them as a single product. See 37 C.F.R. § 385.2 (2019); 37 C.F.R. § 385.21 (2014). Purchased content locker services allow a customer to stream a previously purchased song from a digital storage locker. See 37 C.F.R. § 385.2 (2019); 37 C.F.R. § 385.21 (2014). 11

sound recording copyright holders. 37 C.F.R. § 385.12(b)(1), 385.13(a)–(c), 385.22(b), 385.23(a)–(b) (2014). The percentages used to calculate the total content cost prong varied depending on the type of product offered. See id. § 385.13(b)– (c), 385.23(a). And for certain categories (such as the standalone non-portable subscription categories and the standalone portable subscription category), the total content cost prong was capped, while for others (such as the bundled subscription and the free non-subscription/advertisement- supported categories) it was not. Compare id. § 385.13(a)(1)– (3), with id. § 385.13(a)(4)–(5).

Second, for each type of offering, the service provider would subtract from the greater of the revenue and total content cost prongs the royalties it had already paid for the right to publicly perform musical works through that offering. See 37 C.F.R. § 385.12(b)(2), 385.22(b)(2) (2014).

Third, the service provider would calculate the minimum mechanical license payment, also known as the mechanical floor, for each category of offering. See 37 C.F.R. § 385.13, 385.23 (2014). The mechanical floor was usually determined by multiplying the number of subscribers-per-month by a specific monetary value. A few categories (such as the free non-subscription/advertisement-supported services category) did not have a mechanical floor at all. Compare id. § 385.13(a)(1)–(4), with id. § 385.13(a)(5).

Fourth, the service provider was required to pay, for each type of service offered, a royalty that was the greater of the amount calculated in step two or the mechanical floor calculated in step three. 37 C.F.R. § 385.12(b)(3) (2014). 12

C

This brings us to the case at hand. On January 5, 2016, the Board initiated proceedings to determine the appropriate mechanical license royalty rates and terms for the January 1, 2018 to December 31, 2022 period. 81 Fed. Reg. 255 (Jan. 5, 2016). The parties reached a settlement of the mechanical license royalty rates and terms for physical phonorecords, permanent digital downloads, and ringtones, which we will refer to as the Subpart A settlement. 84 Fed. Reg. at 1920. On March 28, 2017, the Copyright Royalty Board adopted this partial settlement, over the objections of only George Johnson. Id.

The parties were unable to agree, though, on the mechanical license rates and terms for the remaining interactive streaming offerings. As a result, the Board was tasked with adjudicating those rates and terms through adversarial proceedings.

The parties offered an array of competing proposals leading up to the five-week evidentiary hearing. See 84 Fed. Reg. at 1920, 1923–1925.

Prior to the hearing, each of the Streaming Services advanced a somewhat different plan. See 84 Fed. Reg. at 1923. All four broadly sought to maintain the Phonorecords II rate structure, but proposed to either lower or eliminate the mechanical floor. Id.

The Copyright Owners, for their part, proposed a unitary rate structure for all interactive streaming and limited 13

downloads. 84 Fed. Reg. at 1924, 1930–1931.6 Under their proposed royalty scheme, streaming services would pay the greater of (i) a per-play fee or (ii) a per-subscriber fee. Id. They also argued that the Board should continue applying the mechanical floor, but should modify the rate structure so that mechanical license royalties are no longer offset by the payment of performance royalties. Id.

Finally, songwriter George Johnson proposed that all interactive streaming services be required to include a “Buy Button” that allowed customers listening to a song to voluntarily buy or purchase a song as a permanent paid digital download. 84 Fed. Reg. at 1924. He further proposed that between 80% and 84% of the proceeds from these purchases be divided evenly between the owners of the musical work and the owners of the sound recording. Id.

After the evidentiary hearing had concluded and the evidentiary record closed, Google proffered an amended proposal that urged the Board to uncap the total content cost for all categories of offerings in conjunction with lowered royalty rates. 84 Fed. Reg. at 1924, 1930.

On January 27, 2018, the Copyright Royalty Board issued its Initial Determination regarding the appropriate royalties and terms for the 2018-2022 mechanical license. 84 Fed. Reg. at 1918. Two of the three Copyright Royalty Judges, Chief Judge

6 A “limited download” is a download that is only available for a limited period of time or that can only be played a limited number of times. See 37 C.F.R. § 385.11 (2014); 37 C.F.R. § 385.11 (2013); see also 37 C.F.R. § 385.2 (2019). 14

Barnett and Judge Feder, approved the determination. Judge Strickler dissented.

The Initial Determination retained the “All-In” feature that allows interactive streaming services to deduct performance royalties and retained the mechanical floor. J.A. 758–760. The Board explained that it retained the mechanical floor because it “appropriately balances the [streaming service providers’] need for the predictability of an All-In rate with publishers’ and songwriters’ need for a failsafe to ensure that mechanical royalties will not vanish[.]” J.A. 760.

The Board, however, abandoned its prior use of different formulas and percentages to calculate the total content cost prong for different categories of offerings. Instead, it adopted a single, uncapped total content cost rate that applied to all categories of offerings.

By pegging the mechanical license royalties to an uncapped total content cost prong, the Board sought to ensure that owners of musical works copyrights were neither undercompensated relative to sound recording rightsholders, nor harmed by the interactive streaming services’ revenue deferral strategies (such as student and family discount programs). Recall that the total content cost prong is calculated by taking a percentage of sound recording royalties. Therefore, as sound recording royalties increase, the mechanical license royalties generally will also increase, even if the interactive streaming services’ revenue is low as a result of their revenue deferral strategies (such as discounted student plans).

The Board acknowledged that the sound recordings market is a complementary oligopoly and that the sound recording copyright holders can wield their considerable market power to extract excessive royalties. It also recognized that its new 15

structure could increase the mechanical license royalties paid by streaming service providers without necessarily altering the sound recording copyright owners’ ability to extract excessive profits.

But the Board predicted that the sound recording copyright owners’ royalty rates would naturally decline in the course of their negotiations with interactive streaming services. This is so, the Board surmised, because the sound recording copyright owners would likely accept lower rates to “ensure[] the continued survival and growth of the music streaming industry.” J.A. 797. In other words, the only backstop identified by the Board majority was the prospect that sound recording copyright owners would want the existing interactive streaming services to survive rather than (for example) preferring to replace them with their own in-house streaming services. J.A. 798.

Having adopted an overarching rate formula for calculating royalty payments, the Board next considered the specific rates to apply within that structure. In setting the rates, the Board relied primarily on what are known as “Shapley Analyses” provided by the parties’ experts. J.A. 797–799. The Shapley methodology is a game theory model that seeks to assign to each market player the average marginal value that the player contributes to the market. J.A. 786. This methodology first determines the costs that each player should recover, then divides the “surplus” among the players in proportion to the value of their contributions to the worth of the hypothetical bargain that would be struck. J.A. 786.

Drawing from the parties’ competing Shapley Analyses, the Board decided to increase the mechanical license royalty rates paid by interactive streaming services. However, rather than adopt any one expert’s analysis wholesale, the Board drew 16

from multiple studies to construct a range of reasonableness for the royalty rates.

That process yielded a zone of reasonableness between 19.3% and 33.6% for the total content cost prong, and between 11.8% and 18.3% for the revenue prong. J.A. 799. Based on “the totality of the evidence presented[,]” the Board settled on 26.2% as the total content cost rate and 15.1% as the revenue rate. J.A. 799. Those rates were to be phased in gradually over five years. J.A. 812.

In so ruling, the Board rejected the alternative proposals advanced by George Johnson. The Board concluded that it did not have the authority (in setting rates and terms) to require the streaming services to include a “buy button” alongside their offerings, and that there was no evidence in the record to support allocating 80 to 84 percent of the revenue created from such a button to the various copyright owners. J.A. 738–739.

The Board also found that the mechanical license royalty rates should be set at zero in certain circumstances where user streams of copyrighted works produced no revenue for the streaming services. Namely, the rate was zeroed out for situations where the interactive streaming services provide (i) limited free trials to users (hoping to entice them into springing for a paid account); (ii) promotional plays of songs distributed freely by record companies for periods of time; and (iii) purchased content locker services where an individual who has already purchased a song may stream it from a digital locker at no additional revenue for the Streaming Services. In those limited circumstances, where the interactive streaming services draw no revenue from the play of copyrighted material, the Board found that they need “not pay mechanical musical works royalties.” J.A. 801. At the same time, the Board disallowed any “deduct[ion] [of] the costs of those 17

service offerings from service revenue, for purposes of calculating royalties payable on a percent of service revenue.” J.A. 801.

Finally, the Board defined several terms that would govern licenses during the mechanical license royalty rate period, two of which are relevant here.

First, the Board defined how “Service Revenue” would be calculated when interactive streaming services bundle their service with other subscription offerings. The Board defined “Service Revenue” in that context to be (i) the price paid by the consumer for the entire bundle, minus (ii) “the standalone published price for [consumers] for each of the [non-music streaming] component(s) of the Bundle,” (iii) provided that, if there is no published price for the standalone components, then the service providers “shall use the average standalone published price for [consumers] for the most closely comparable product or service in the U.S. or, if more than one comparable exists, the average of standalone prices for comparables.” J.A. 826–827.

Second, the Board concluded that, in setting the mechanical floor, student and family plans should be counted differently for purposes of computing the number of subscribers to a streaming service. Reflecting how the interactive streaming services generally priced those plans relative to their standard subscription offerings, the Board deemed “Family Plans” to be the equivalent of 1.5 subscribers and “Student Accounts” to be the equivalent of 0.5 subscribers. See J.A. 817, 831.7

7 The Board states that it is adopting Spotify’s proposal, which would treat family plans as 1.5 subscribers. The Board then adds that 18

Copyright Royalty Judge Strickler dissented from the Initial Determination. Among other things, he objected to the Board’s adoption of a rate structure that “was only proposed after the hearing, when the record had already been closed.” J.A. 834–835. Judge Strickler further reasoned that uncapping the total content cost prong could imperil the existence of the interactive streaming services. Specifically, the sound recording copyright owners “may decide to keep their rates high despite the increase in mechanical rates,” or they may simply create their own “in-house” streaming services and refuse to contract with the existing interactive streaming services at all. J.A. 837.

The Streaming Services moved for rehearing of the Initial Determination by the Board. See 17 U.S.C. § 803(c)(2)(A) (authorizing the Board, “in exceptional cases, upon motion of a participant in a proceeding[,] * * * [to] order a rehearing, after” issuing an Initial Determination, “on such matters as the [Board] determine[s] to be appropriate”). The Streaming Services’ motion was limited to fixing clerical errors and clarifying existing ambiguities in the proposed regulatory terms appended to the Initial Determination.

The Copyright Owners, for their part, disclaimed any intent to seek rehearing, but moved for “clarification or correction” of certain regulatory terms “to conform them to what appears to be the intent of the [Initial] Determination.” J.A. 103 (formatting modified). They purported to bring their motion under the Board’s general regulations governing

these plans will be counted as “one subscriber[.]” J.A. 817. That appears to be a typographical error. See J.A. 817, 831. 19

motions. See 37 C.F.R. § 303.3–.4 (formerly codified at 37 C.F.R. § 350.3–.4).

The Copyright Owners’ clarification motion argued, among other things, that the definition of Service Revenue as applied to bundled offering should be reworked. They argued that defining the revenue as the total price of the bundle, minus the standalone published prices for the non-streaming offerings in the bundle, undervalued the revenue created by the streaming offerings. To illustrate, the Copyright Owners (quoting a different Board ruling) reasoned that, “[i]f a vendor offered an ice cream cone * * * for $1.00, but offered two ice cream cones for $1.06, it would be absurd to conclude that the true market price of an ice cream cone is the incremental six cents.” J.A. 114 (quoting 81 Fed. Reg. 26,316, 26,382 (May 2, 2016)).

So instead, the Copyright Owners proposed that “Service Revenue” from bundled offerings be defined as “the standalone price of the [streaming] offering (or comparables).” They added that, in their view, that new definition would be “more consistent with the [Board’s] reasoning” in the Initial Determination. J.A. 115.

The Streaming Services objected to the Copyright Owners’ styling of their motion as something other than a motion for rehearing, describing it as an attempt “to skirt the standard governing motions for rehearing—a standard that the Copyright Owners have not even attempted to meet.” J.A. 1251 (internal quotation marks omitted). The Streaming Services also objected that the Copyright Owners had not previously proposed a definition of “Service Revenue” from bundled offerings, and that their “late-proposed” definition was unsupported by the record. J.A. 1266 (internal quotation marks omitted). 20

In October 2018, the Board issued an order “granting in part and denying in part motions for rehearing.” J.A. 1250– 1271 (formatting modified). The order concluded that neither party had met the “exceptional standard for granting rehearing motions.” J.A. 1251 (stating that the moving parties had failed to present “even a prima facie case for rehearing under the applicable standard”). The Board explained that it nevertheless found it “appropriate * * * to resolve the issues that the parties ha[d] raised[.]” J.A. 1251. The Board added that, to the extent such resolution “could be considered a rehearing under 17 U.S.C. § 803(c)(2),” it “resolve[d] [the] motions on the papers without oral argument.” J.A. 1251.

Regarding the definition of “Service Revenue” for bundled offerings, the Board noted that “[n]either party presented evidence adequate to support the approach it advocates” in its post-determination filing. J.A. 1266. (A curious statement since the Streaming Services were simply defending the Board’s chosen method.) Noting that the Streaming Services were “the party in possession of the relevant information,” the Board concluded that they “bore the burden of providing evidence that might mitigate the * * * problem inherent in bundling.” J.A. 1266–1267 (internal quotation marks omitted). Because the Streaming Services had failed that task, the Board, “by default,” ruled that it “must adopt an approach to valuing bundled revenue that is in line with what the Copyright Owners have proposed.” J.A. 1267 (formatting modified). As a result, the Board discarded the formula in the Initial Determination and ruled, instead, that streaming service providers “will use their own standalone price (or comparable) for the music component (not to exceed the value of the entire bundle) when allocating bundled revenue.” J.A. 1267. 21

On November 5, 2018, the Board issued its Final Determination. 84 Fed. Reg. at 1963. Three months later, after review by the Register of Copyrights, a partially redacted version of the Final Determination was published in the Federal Register. Id. at 1918–2036.

The Board’s Final Determination closely tracked the Initial Determination. It adopted the same rate structure and rate percentages set forth in the earlier rulemaking. Specifically, the Board retained the mechanical floor, but uncapped the total content cost prong for all categories of offerings. See 84 Fed. Reg. at 1934–1935. And it stood by its decision to phase in over five years, for all categories, a 15.1% revenue rate and a 26.2% total content cost rate. Id. at 1960. The Board also maintained the counting of family plans as the equivalent of 1.5 subscribers and student accounts as the equivalent of 0.5 subscribers when calculating the mechanical floor. Id. at 1962.

As relevant here, the Final Determination deviated from the Initial Determination in one key respect. Consistent with the Board’s order at the rehearing stage, the Final Determination redefined “Service Revenue” from bundled offerings as the lesser of (i) the revenue of the bundle, and (ii) the aggregate of standalone prices for the licensed music products included in the bundle. 84 Fed. Reg. at 2034.

The Final Determination made the new rates and terms effective from January 1, 2018 through December 31, 2022. 84 Fed. Reg. at 1918. That was the same effective period proposed by the Board in its notice of the ratemaking proceeding in January 2016 and adopted by each participant in its Proposed Findings of Fact and Proposed Conclusions of 22

Law. Id. (noting that 17 U.S.C. § 115(c)(3) (2012) permits the parties to agree to an effective rate period); see 17 U.S.C. § 803(d)(2)(B) (permitting the same).

The Streaming Services, the Copyright Owners, and George Johnson timely appealed the Board’s Final Determination. This court consolidated those appeals.

II

We have jurisdiction to review the Board’s Final Determination under 17 U.S.C. § 803(d)(1). In conducting our review, we apply the same standards set forth in the Administrative Procedure Act. Id. § 803(d)(3) (cross- referencing 5 U.S.C. § 706). That means that we will set aside the Final Determination “only if it is arbitrary, capricious, an abuse of discretion, or otherwise not in accordance with law, or if the facts relied upon by the agency have no basis in the record.” Independent Producers Group v. Librarian of Congress, 792 F.3d 132, 136 (D.C. Cir. 2015) (internal quotation marks omitted).

III

The Streaming Services, Copyright Owners, and George Johnson challenge numerous aspects of the Copyright Royalty Board’s Final Determination. First, the Streaming Services argue that the Board’s decision impermissibly applies retroactively. Second, the Streaming Services challenge the Board’s rate structure and the specific rates applicable under that structure. Third, the Streaming Services and the Copyright Owners each object to the Board’s definition of certain terms. Finally, George Johnson challenges the Board’s acceptance of the Subpart A settlement, as well as its adoption of the final rate structure. 23

We reject the Streaming Services’ retroactivity objection and the challenges brought by the Copyright Owners and George Johnson. But we agree with the Streaming Services that the Board failed to provide adequate notice of the final rate structure, failed to reasonably explain its rejection of the Phonorecords II settlement as a benchmark, and failed to identify under what authority it substantively redefined a term after publishing its Initial Determination. For those reasons, we vacate and remand the Final Determination for further proceedings consistent with this opinion.

The Streaming Services argue that the Board overshot its regulatory authority by giving its royalty rates and terms retroactive effect. Streaming Services Br. 63–69.8 Because there is nothing retroactive about the Board’s rate determination, that challenge fails.

Section 115(c)(3)(C) of Title 17 provides that the Board’s proceedings:

[S]hall determine reasonable rates and terms of royalty payments * * * during the period beginning with the effective date of such rates and terms, but not earlier than January 1 of the second year following the year in which the petition requesting the proceeding is filed, and ending on the effective date of successor rates and terms, or such other period as the parties may agree.

8 Spotify has not joined this argument. Streaming Services Reply Br. 31 n.13. So for purposes of the retroactivity portion of this opinion only, the “Streaming Services” appellation excludes Spotify. 24

17 U.S.C. § 115(c)(3)(C) (2012). Section 803(d)(2)(B) provides consistent (and more detailed) guidance regarding the effective dates of rates and terms. It provides in relevant part:

In cases where rates and terms have not, prior to the inception of an activity, been established for that particular activity under the relevant license, such rates and terms shall be retroactive to the inception of activity under the relevant license covered by such rates and terms. In other cases where rates and terms do not expire on a specified date, successor rates and terms shall take effect on the first day of the second month that begins after the publication of the determination of the [Board] in the Federal Register, except as otherwise provided * * * by the [Board], or as agreed by the participants in [the] proceeding that would be bound by the rates and terms.

17 U.S.C. § 803(d)(2)(B).

That is all a long way of saying that when, as here, prior royalty rates do not have a predetermined end date, Section 803(d)(2)(B) establishes a default effective date for new rates of “the first day of the second month that begins after the publication of the [Final Determination] of the [Board] in the Federal Register[.]” 17 U.S.C. § 803(d)(2)(B).

But both Sections 803 and 115 expressly authorize the participants in a ratemaking proceeding to agree to a different effective date for new rates and terms and allow the Board to adopt the participants agreed-upon effective dates. See 17 U.S.C. § 115(c)(3)(C) (2012) (effective date can be “such other period as the parties may agree”); 17 U.S.C. § 803(d)(2)(B) (effective date can be the date “as agreed by the participants in 25

[the] proceeding that would be bound by the rates and terms”). That is exactly what happened here.

In its Final Determination, the Board made its new rates “effective during the rate period January 1, 2018, through December 31, 2022[.]” 84 Fed. Reg. at 1918. That effective date was not the default date of April 1, 2019 (that is, the first day of the second month that began after publication of the Final Determination in the Federal Register on February 5, 2019). See 17 U.S.C. § 803(d)(2)(B); 84 Fed. Reg. at 1918.

Instead, the Board selected that effective date because it was the parties’ long-agreed-upon start date. 84 Fed. Reg. at 1918; see id. (noting that 17 U.S.C. § 115(c)(3) (2012) permits the parties to agree to an effective rate period); see also 17 U.S.C. § 803(d)(2)(B) (permitting the same). As evidence, the Board pointed out that: (i) it had proposed that effective rate period in its January 2016 notice of the ratemaking proceeding; (ii) the docket for the ratemaking proceeding had consistently included that same effective rate period; and (iii) “each party included in its Proposed Findings of Fact and Proposed Conclusions of Law a designation of” that same rate period, including specifically the January 1, 2018 effective date. 84 Fed. Reg. at 1918 (emphasis added).9

9 While Section 803(d)(2)(B) also permits the Board itself to “otherwise provide[]” for an effective rate period, the Board never mentioned that authority in its Final Determination as a basis for its actions. So while the Board now attempts to claim that authority, we may not sustain its action on that late-breaking, extra-record ground. See, e.g., Department of Homeland Security v. Regents of the Univ. of Cal., 140 S. Ct. 1891, 1907 (2020) (“It is a foundational principle of administrative law that judicial review of agency action is limited 26

The Streaming Services argue that the Board’s adoption of the rate period starting on January 1, 2018 was unlawful for two sets of reasons. Neither is persuasive.

To start, the Streaming Services argue that they never actually agreed to an effective rate period of January 1, 2018 through December 31, 2022. The record says otherwise.

The Streaming Services do not dispute the Board’s statement in the Final Determination that “each party included in its Proposed Findings of Fact and Proposed Conclusions of Law a designation of the rate period as January 1, 2018, through December 31, 2022.” 84 Fed. Reg. at 1918. Nor could they. The Streaming Services’ filings throughout the course of the ratemaking proceeding, not just their own proposed rates and terms, consistently used that January 1, 2018 effective date. See, e.g., J.A. 50 (“Amazon * * * proposes the following rates and terms for making and distributing phonorecords under the statutory license provided by 17 U.S.C. § 115 during the period January 1, 2018 through December 31, 2022); see also J.A. 42, 44, 46, 48, 51, 55, 651–652, 670, 682, 710.

In the face of those express endorsements of the January 1, 2018 start date, all the Streaming Services can point to is a single sentence in their nearly six-hundred-page, post-trial reply brief. Streaming Services Br. 68–69 (citing J.A. 711).

That was too little, too late. For starters, this court “generally will not consider an argument that was not raised before the agency at the time appropriate under its practice.”

to the grounds that the agency invoked when it took the action.”) (internal quotation marks omitted). 27

BNSF Ry. Co. v. Surface Transp. Bd., 453 F.3d 473, 479 (D.C. Cir. 2006) (internal quotation marks omitted). It is doubtful that a reply brief is the appropriate time or place to disavow an agreement reaffirmed repeatedly during the course of ratemaking proceedings. See 37 C.F.R. § 351.14(b) (“A party waives any objection to a provision in the determination unless the provision conflicts with a proposed finding of fact or conclusion of law filed by the party.”).

On top of that, the meaning of the cited sentence in the Streaming Services’ reply brief is not even clear. The reply brief, in fact, states that “January 1, 2018 would be a proper effective date for rates to be determined in this proceeding[.]” J.A. 711.

The sentence then goes on to “note” the Streaming Services’ assumption that operation of that agreed-upon date presupposes a November 2017 publication of the Board’s final determination because that would make the effective date coincide with the statutory default rule. J.A. 711 (“[T]he Services note that while January 1, 2018 would be a proper effective date for rates to be determined in this proceeding, it will actually be the effective date only if the Judges publish their determination in the Federal Register in November of 2017. See 17 U.S.C. § 803(d)(2)(B) (noting that ‘successor rates and terms shall take effect on the first day of the second month that begins after the publication of the determination of the Copyright Royalty Judges in the Federal Register’ and that ‘the rates and terms, to the extent applicable, shall remain in effect until such successor rates and terms become effective.’).”).

Nothing in that passing comment argues or analyzes the parties’ independent authority to agree to an effective period different from the default start date. Nor does it indicate that 28

the Streaming Services’ longstanding agreement with the prospectively announced rate period is now suddenly contingent on publication by a certain date—a publication date that would render the agreed-upon date redundant of the statute’s default effective date.

Critically, months later when the Initial Determination included the same effective dates, J.A. 725, which was published after those effective dates began, the Streaming Services made no objection to those dates or otherwise communicated that they were withdrawing their prior agreement with the January 1, 2018 start date.

The Board, in short, was well within its rights to take the Streaming Services (and the other parties) at their word as to the long-recognized and oft-repeated agreement to a January 1, 2018 effective date.

The Streaming Services argue, in the alternative, that Section 803(d)(2)(B)’s language allowing an agreed-upon start date does not apply if enforcing that agreement would give the rates and terms retroactive effect. They point out that the Copyright Act mandates the retroactive effect of new rates only in limited circumstances. For example, the first sentence of Section 803(d)(2)(B) requires that new rates be retroactive “where rates and terms have not, prior to the inception of any activity, been established for that particular activity under the relevant license[.]” 17 U.S.C. § 803(d)(2)(B). Section 803(d)(2)(A) similarly requires retroactive rates if the prior rates have a fixed expiration date that predates the Board’s determination of new rates. 17 U.S.C. § 803(d)(2)(A). Neither provision applies here because (i) prior rates existed for the relevant activities by virtue of the Phonorecords II 29

settlement, and (ii) those rates did not have a fixed expiration date. Streaming Services Reply Br. 31.

As the Streaming Services’ argument goes, because the Copyright Act mandates retroactive effect of rates only in specified circumstances, the statute’s general grant of authority to the Board (in other cases) to adopt an effective date to which the parties agreed does not include the ability to adopt agreed- upon rates that would have retroactive effect. Streaming Services Reply Br. 31. Especially because “a statutory grant of legislative rulemaking authority will not * * * be understood to encompass the power to promulgate retroactive rules unless that power is conveyed by Congress in express terms.” Streaming Services Br. 63 (quoting Bowen v. Georgetown Univ. Hosp., 488 U.S. 204, 208 (1988)).

The Streaming Services also point to the Register of Copyrights’ prior statement that “[n]either the [Board] nor the participants [in a rate setting proceeding] have the power to engage in retroactive rate setting” for successor rates and terms. 74 Fed. Reg. 4537, 4542 (Jan. 26, 2009) (setting aside Board’s adoption of retroactive rate setting). The Board, they stress, is bound to follow “prior determinations and interpretations of * * * the Register of Copyrights[.]” 17 U.S.C. § 803(a)(1); Independent Producers Group, 792 F.3d at 137 & n.3.

Those are thoughtful arguments. But they do nothing to advance the ball for the Streaming Services. Even assuming without deciding that Section 803(d)(2)(B) does not authorize agreed-upon “retroactive” rate setting, that is not at all what the Board and the participants did here.

Recall that the Board announced in January 2016 that this rate setting proceeding was “to determine reasonable rates and terms for making and distributing phonorecords for the period 30

beginning January 1, 2018, and ending December 31, 2022.” 81 Fed. Reg. at 255–256. All participants and the Board then consistently repeated that prospective effective date as they wound their way through the ratemaking proceeding. See, e.g., J.A. 42, 44, 46, 48, 50 51, 55, 651–652, 670, 682, 710. The start date, in other words, was set prospectively, not retroactively. The Streaming Services signed off. It is as simple as that.

Of course, the Board’s Final Determination was not published in the Federal Register until February 5, 2019, which was more than a year after the effective rate period actually began on January 1, 2018. But given the Board’s prospective announcement of the effective rate period in 2016 and the parties’ continuous agreement over the ensuing years to those dates, what was done here is a far cry from retroactive rate setting. It bears no resemblance to cases like Bowen, where the Secretary of Health and Human Services published in February 1984 a proposal to reissue an invalidated wage-index rule and, without advance notice to the affected parties, made it “retroactive to July 1, 1981.” 488 U.S. at 207; see also Landgraf v. USI Film Products, 511 U.S. 244, 272 (1994) (describing Bowen as “a paradigmatic case of retroactivity in which a federal agency sought to recoup, under cost limit regulations issued in 1984, funds that had been paid to hospitals for services rendered earlier”).

Relying on our decision in Treasure State Resources Industry Association v. EPA, 805 F.3d 300, 305 (D.C. Cir. 2015), the Streaming Services argue that the Board’s Final Determination applied the new rates and terms retroactively because it attached “new legal consequences to [transactions] completed before its enactment.” Streaming Services Reply Br. 32 (quoting Treasure State, 805 F.3d at 305). 31

While it is undoubtedly quite relevant in deciding whether provisions are retroactive to determine whether they attach new legal consequences to events completed before their enactment, that “is far from the end of the story.” Treasure State, 805 F.3d at 305. A provision “does not operate retrospectively merely because it is applied in a case arising from conduct antedating [its] enactment.” Id. (formatting modified; quoting Landgraf, 511 U.S. at 269–270). Instead, considerations of “fair notice, reasonable reliance, and settled expectations” of the parties also carry weight. Id. (internal quotation marks omitted).

None of those considerations put the Board’s determination in the retroactive camp. When the Board announced in 2016 that it would be conducting a rate-setting proceeding to set different rates and terms for January 1, 2018 through December 31, 2022, it gave the Streaming Services “ample public notice of the impending change in” rates, and— at minimum—signaled “the prospect of” different rates starting on January 1, 2018, Treasure State, 805 F.3d at 306 (emphasis added); see Columbia Gas Transmission Corp. v. FERC, 895 F.2d 791, 796–797 (D.C. Cir. 1990) (Agency rates are not “retroactive” where the agency “itself places parties on notice * * * that the rates they will be paying are subject to retroactive adjustment at a later date.”).

The Board’s issuance of its Initial Determination in January 2018—within weeks of the agreed-upon January 1, 2018 effective date—provided still more and quite precise notice of the rates that were intended to govern. Add in the parties’ repeated confirmation of that effective rate period throughout the entire proceedings, long before the rate period began, and the record leaves no ground for labeling that long- forewarned effective rate period an exercise in retroactive ratemaking. Nor have the Streaming Services explained how, given that enduring agreement, they had any reasonable 32

reliance interest in continuing the old rates into the period that everyone had stipulated all along would be governed by the new rates.

For the same reasons, the Streaming Services’ citation to a prior determination of the Librarian of Congress fails. See 74 Fed. Reg. at 4542. The referenced rate-setting proceeding did not involve, as here, a prospective announcement of the effective rate period. See 71 Fed. Reg. 1453, 1453–1454 (Jan. 9, 2006) (Board’s notice announcing ratemaking proceeding that does not include a proposed effective date).

In sum, Section 803(d)(2)(B) authorized the Board to do what it did here: Prospectively announce and stick with an effective rate period to which the parties had repeatedly and expressly agreed in writing throughout the proceedings. That the final rule was not published in the Federal Register until a year later does not amount to retroactive rate setting that either surprised the Streaming Services (who had long agreed to the pre-established start date) or disrupted any reasonable reliance interests.

The Copyright Royalty Board’s Final Determination adopted a rate structure for computing the mechanical license that uncapped the total content cost prong for every category of streaming service offered, while simultaneously increasing both the total content cost and revenue rates. With no cap in place, the Board’s decision removed the only structural limitation on how high the total content cost (which is pegged to unregulated sound recording royalties) can climb.

The Streaming Services argue that, in adopting that rate structure, (i) the Board violated their procedural right to fair 33

notice by choosing a structure that was not advanced by any party; and (ii) the Board’s decision both to uncap the total content cost prong and to increase the percentages used to calculate the revenue prong and total content cost prongs were arbitrary and capricious.

The Streaming Services are correct that the Board failed to provide adequate notice of the drastically modified rate structure it ultimately adopted. We also hold that the Board did not provide a reasoned explanation for its refusal to treat the Phonorecords II settlement as a benchmark when setting the total content cost and revenue rates. For those reasons, we vacate and remand the Board’s adopted rate structure and percentages for further proceedings consistent with this opinion.

An agency “setting a matter for hearing [must] provide parties with adequate notice of the issues that [will] be considered, and ultimately resolved, at that hearing.” Wallaesa v. Federal Aviation Admin., 824 F.3d 1071, 1083 (D.C. Cir. 2016) (quoting Public Serv. Comm’n of Ky. v. FERC, 397 F.3d 1004, 1012 (D.C. Cir. 2005)). This ensures that agencies provide a fair process in which each party is able “to present [its] case or defense * * *, to submit rebuttal evidence, and to conduct such cross-examination as may be required for a full and true disclosure of the facts” that bear on the agency’s decision and choices. See 5 U.S.C. § 556(d).

While the Streaming Services knew, at the 50,000-foot level, that the Board would be deciding the royalty rates and terms to govern the mechanical license, they had no fair notice that the Copyright Royalty Board would take the dramatic step of uncapping the total content cost prong for every category of 34

service offering, let alone pair that with significant increases in the total content cost and revenue prongs. As a result, the Streaming Services argue, they “had no notice or opportunity to present evidence” about that rate structure because “no party advocated [it] during or before the hearing.” Streaming Services Br. 26–27.

We agree. There is no dispute that before and throughout the evidentiary hearing, no party had proposed or even hinted at the structure the Board ultimately adopted—an uncapped total cost content prong combined with significantly increased rates. See Streaming Services Br. 17; Board Br. 43–44; Copyright Owners Intervenor Br. 10–11; see also Oral Arg. Tr. 55:7–60:19, 68:21–69:5, 75:4–13. And the Board, itself, offered no hint of such a dramatic change in course from prior decisions. So the Streaming Services had no notice that they needed to defend against and create a record addressing such a significant, and significantly adverse, overhaul of the mechanical license royalty scheme.

This is no mere formality. Interested parties’ ability to provide evidence and argument bearing on the essential components and contours of the Board’s ultimate decision not only protects the parties’ interests, it also helps ensure that the Board’s ultimate decision is well-reasoned and grounded in substantial evidence. See Settling Devotional Claimants v. Copyright Royalty Bd., 797 F.3d 1106, 1121 (D.C. Cir. 2015) (The lack of record support for the Copyright Royalty Board’s approach was “ma[de] * * * worse” by the fact that the approach was “first presented in the * * * determination and not advanced by any participant.”) (quoting Intercollegiate Broad. Sys., Inc. v. Copyright Royalty Bd., 571 F.3d 69, 87 (D.C. Cir. 2009)). That vetting by the parties did not occur here because the Streaming Services were procedurally blindsided. 35

That is not to say that the Board is strictly limited to choosing from among the proposals set forth by the parties. Agencies have the authority to modify proposals set forth by the parties, or to suggest models of their own. See SoundExchange, Inc. v. Copyright Royalty Bd., 904 F.3d 41, 50–51, 57 (D.C. Cir. 2018) (upholding Copyright Royalty Board’s decision to modify a party’s proposed rates in light of its interpretation of Section 114 of the Copyright Act); Association of American Publishers, Inc. v. Governors of USPS, 485 F.2d 768, 773 (D.C. Cir. 1973). Some degree of deviation and combination is permissible.

But the ultimate proposal adopted by the Board has to be within a reasonable range of contemplated outcomes. Here it was not.

The Board’s decision went far beyond modifying or piecing together a rate structure, the economic and policy consequences of which had already been explored and developed by the parties in the record. Instead, the Board’s decision deviated substantially and unforeseeably from the parties’ pre-hearing proposals, the arguments made at the evidentiary hearing, and the preexisting rate structures. The end result was a rate structure that could significantly increase costs for the Streaming Services, and that eliminated the prior structural protection that braced the streaming services against unregulated increases in sound recording royalties. If the Board wanted to implement such an extreme change in the rate structure, it was duty bound to give a heads up to the parties.

The Copyright Owners argue that the Streaming Services nevertheless had adequate notice because they received “notice of every component of the Board’s final structure[.]” Copyright Owners Intervenor Br. 10. In particular, the final rate structure adopted by the Board for all categories was 36

virtually identical to the rate structure for the single “bundled subscriptions” category prior to Phonorecords III, see 37 C.F.R. § 385.13(a)(4) (2013), which Amazon’s filings below proposed to maintain to that limited extent, J.A. 66–67.

But the fact that some of the Streaming Services’ proposals contemplated continued use of an uncapped total content cost prong for a small number of preexisting categories does not mean they anticipated that the Board would uncap the total content cost prong across the board. That is quite different. Prior to issuance of the Final Determination, the vast majority of the categories of offerings were capped. See Oral Arg. Tr. 57:7–23 (Board explaining that, previously, only two out of ten categories were uncapped).

Uncapping the total content cost prong across all categories leaves the Streaming Services exposed to potentially large hikes in the mechanical license royalties they must pay. That is because the total content cost prong is calculated by taking a percentage of the royalties paid to sound recording rightsholders. See 84 Fed. Reg. at 1963 n.165 (Strickler Dissent). As the Board acknowledges, sound recording rightsholders have considerable market power vis-à-vis interactive streaming service providers, and they have leveraged that power to extract excessive royalties. See Id. at 1934 n.75 (“[T]he [interactive streaming] services are * * *exposed to the labels’ market power. Record companies could, if they so chose, put th[ose] [s]ervices out of business entirely.”); id. at 1952–1953 (explaining that, by virtue of their oligopoly power, the sound recording copyright holders have extracted “inflated” royalties relative to what the Shapley Analyses would predict). By eliminating any cap on the total content cost prongs, the Final Determination yokes the mechanical license royalties to the sound recording rightsholders’ unchecked market power. 37

Worse still, the Board not only stripped away the total content cost caps, but also significantly hiked both the revenue rate and the total content cost rates the streaming services would have to pay. The Final Determination increased the revenue rate by 44% relative to the preexisting rates—that is, from 10.5% (for most but not all categories) to 15.1%. See 84 Fed. Reg. at 1960.10 The Board also raised the total content cost rate to 26.2%. Id. at 1954, 1960. That rate previously fell between approximately 17% and 22%. See 37 C.F.R. § 385.13(b)–(c), 385.23(a) (2014). Therefore, the Streaming Services were not only deprived of the opportunity to voice their objections to a completely uncapped total content cost prong, they were also given no opportunity to address the interplay between that rate structure and the increased revenue and total content cost rates.

In defense of the Board’s decision, the Copyright Owners point to Google’s post-hearing proposal that advocated for a rate structure that included a completely uncapped total content cost prong. J.A. 678–679, 699 (Google’s Amended Proposed Finding of Fact). There are two problems with that argument.

First, Google conditioned its proposed adoption of an uncapped total content cost prong on the simultaneous adoption of much lower rates than those adopted by the Board. J.A. 483 (Google confirming that its amended proposed rate structure “works but only with” the specific rates it proposed); see also 84 Fed. Reg. at 1981 (Google’s Amended Proposal advanced a “10.5% of net service revenue [rate] and an uncapped 15[%]

10 Some of the categories had slightly higher revenue rates. See 37 C.F.R. § 385.23(a)(1), (2), (4), (5) (2014). The Board, however, generally treats 10.5% as the “prevailing headline rate[.]” 84 Fed. Reg. at 1952; see id. at 1960 (describing the new revenue rate as a 44% increase over the “current headline rate”). 38

[total content cost] component.”). Google specifically warned that pairing the uncapped structure with higher rates would be intolerable. J.A. 483.

Second, Google’s package was proposed for the very first time after the evidentiary record was closed. 84 Fed. Reg. at 1935 n.79. So the other parties never had a chance to submit evidence regarding the problems with Google’s proposal, let alone the viability of the Board’s pairing of uncapped total content costs with significantly increased total content cost and revenue rates.

In sum, because the Copyright Royalty Board failed to provide fair notice of the rate structure it adopted, that aspect of its decision must be vacated and remanded for further proceedings. If the Board wishes to pursue its novel rate structure, it will need to reopen the evidentiary record.

The Streaming Services separately challenge the uncapped rate structure as arbitrary and capricious. In particular, they argue that the rate structure formulated by the Copyright Royalty Board failed to account for the sound recordings rightsholders’ market power. Streaming Services Br. 28–32, 34. They also object that the Board failed to provide a “satisfactory explanation,” or root in substantial evidence, its conclusion that an increase in mechanical license royalties would lead to a decrease in sound recording royalties. Streaming Services Br. 33–36.

Because we have vacated the rate structure devised by the Board for lack of notice, we need not address these arguments. Should the Board on remand provide notice that it is again 39

contemplating such a scheme, the Streaming Services can present their concerns to the Board in the first instance.

Apart from their challenges to the uncapped rate structure, the Streaming Services separately leveled objections to the particular percentages adopted by the Copyright Royalty Board to calculate the revenue and total content cost prongs. Specifically, the Streaming Services object to the Board’s reliance on conclusions drawn from multiple different expert analyses and its rejection of the Phonorecords II and Subpart A settlements as rate benchmarks. They further argue that the Board’s conclusions with respect to the four statutory objectives were unreasoned and unsupported by substantial evidence.

Our “[r]eview of administratively determined rates is ‘particularly deferential’ because of their ‘highly technical’ nature.” Intercollegiate Broad. Sys. v. Copyright Royalty Bd., 574 F.3d 748, 755 (D.C. Cir. 2009) (quoting East Ky. Power Coop. v. FERC, 489 F.3d 1299, 1306 (D.C. Cir. 2007)). Taken in that light, the Streaming Services’ first argument fails, but the second succeeds. The third argument is largely bound up in the remand and therefore, with one exception, will not be resolved here.

The Streaming Services’ first contention is that it was arbitrary and capricious for the Board to rely on information drawn from different expert analyses in calculating the mechanical royalty rates. Streaming Services Br. 38–39. In their view, the Board’s approach was “virtually identical” to the arbitrary approach struck down in Settling Devotional 40

Claimants because it involved “taking some numbers from discredited methodologies and others from methodologies” that the Board “otherwise refused to consider.” Streaming Services Br. 43 (formatting modified).

This case looks nothing like Settling Devotional Claimants. In that case, two parties each claimed a right to the royalties associated with devotional programming. Settling Devotional Claimants, 797 F.3d at 1111. But neither party presented competent evidence of how the devotional programming royalties should be divided between them. So the Board rejected both parties’ methodologies in total. Id. at 1113. That left the Board with no reliable evidence at all of how to allocate the devotional programming royalties. So the Board chose to split the difference. For two of the years at issue, the Board picked the one royalty allocation number where the parties’ (already rejected) proposed ranges happened to coincide. Id. at 1114. For the other two years, the Board simply averaged the values selected for the first two years. Id.

This court reversed, explaining that “simply picking a number out of [a] flawed and otherwise-rejected proposal just because it happened to roughly coincide with the lowest bound proposed by [an opposing party] falls beyond the bounds of reasoned decisionmaking.” Settling Devotional Claimants, 797 F.3d at 1120. Even worse, for two of the years, the Board simply “split the difference of the allocations from the two other years” without any evidentiary support or rationale for that decision. Id. at 1121.

Unlike in Settling Devotional Claimants, the Board here did not split the difference between analyses that it had already rejected as fatally flawed. Rather, the Board found the analyses upon which it relied informative and accurate in the relied- upon parts, even if imperfect in other respects. In particular, 41

based on analyses submitted by Spotify’s expert, Leslie Marx, and the Copyright Owners’ experts, Richard Watt and Joshua Gans, the Board found that the Phonorecords II mechanical license royalties were too low and that the sound recording rightsholders were extracting more than their fair share of royalties. 84 Fed. Reg. at 1951–1954. The Board then carefully analyzed the competing testimony and drew from it rates that were grounded in the record and supported by reasoned analysis. See Association of American Publishers, 485 F.2d at 773 (“[T]he rough splitting of a difference between two fairly but not wholly satisfactory rate calculations is a familiar permissible technique,” since a ratemaking entity “may fashion its own adjustments within reasonable limits.”).

All three expert analyses “were broadly consistent insofar as they all found that the ratio of sound recording to musical works royalty rates should decline,” 84 Fed. Reg. at 1951– 1952, because sound recording royalties were too high relative to the musical works royalties. Watt explained that this phenomenon was at least partially explained by the fact that mechanical license royalties were “significantly below the predicted fair rate.” Id. at 1952. That gap had empowered sound recording rightsholders to extract that surplus amount in their negotiations with the interactive streaming services. Id.

To select the specific revenue rate that would respond to that problem, the Board began by determining the total percent of the interactive streaming services’ revenue that should be paid out in royalties to the sound recording rightsholders and the musical works rightsholders collectively. To that end, the Board treated Marx’s upper estimate of the percent of total revenue that interactive streaming services should pay in royalties as the lower bound for the Board’s range. 84 Fed. Reg. at 1954. The Board took this approach because it found, based on Watt’s testimony, that Marx’s analysis understated 42

the fair allocation of surplus to the copyright owners by overstating the streaming services’ costs and understating their revenue. Id. at 1953–1954. Substantial evidence supports that judgment.

The Board then took the lower bound of Watt’s estimate of the percentage of revenue that should be paid out in royalties and treated that value as the upper bound for the Board’s purposes. 84 Fed. Reg. at 1954. The Board explained that it did so because Watt’s royalty figures “were presented as rebuttal testimony” to Marx’s testimony, so Marx had no opportunity to respond to them. Id. Therefore, the Board took the conservative approach of “viewing [Watt’s] lowest figure * * * as an upper bound[.]” Id. That type of weighing of evidence and decision to proceed cautiously is well within the Board’s discretion.

The Board proceeded to calculate the zone of reasonableness for the revenue rate by applying the ratios of sound recording royalties to musical works royalties from Gans’ and Marx’s analyses to the Board’s upper and lower bounds for the percent of revenue that should be paid out in total royalties. 84 Fed. Reg. at 1954.

The Board ultimately settled on the revenue rate of 15.1% “based on the highest value of overall royalties predicted by Professor Marx’s model and the ratio of sound recording to musical work royalties determined by * * * Gans’s analysis.” 84 Fed. Reg. at 1959–1960. Gans’ analysis sought to estimate an appropriate per-subscriber or per-play mechanical license royalty rate. Id. at 1951. This involved estimating the ratio of sound recording royalties to musical works royalties in an unconstrained market. Id. at 1950. 43

The Streaming Services argue that the Board acted arbitrarily by relying on the ratio Gans derived even though it “explicitly rejected [Gans’ model] as unreliable[.]” Streaming Services Br. 43 (emphasis in original). That is not what happened.

The Board rejected Gans’ per-subscriber and per-play royalty proposals, but not because of any infirmity in his analysis of the ratio of sound recording royalties to musical works. The Board just found that Gans’ reliance on another expert’s unsound per-play sound recording rate fatal to his proposed per-subscriber and per-play mechanical royalty rates. 84 Fed. Reg. at 1951. When it came to the expert evidence on which the Board chose to rely—the “ratio of sound recording to musical work royalties that * * * Gans derived from his analysis”—the Board specifically found that aspect of Gans’ analysis to be reasonable and “informative.” Id. (formatting modified); id. (Board finds reasonable Gans’ equal value assumption and his reliance on Goldman Sachs’ profit projections). That type of line-drawing and reasoned weighing of the evidence falls squarely within the Board’s wheelhouse as an expert administrative agency.11

11 The Streaming Services also argue that the Board arbitrarily, and without explanation, selected the midpoint of the zone of reasonableness as the revenue rate. Streaming Services Br. 43. That misunderstands the Board’s decision. The Board explained that the revenue rate is “based on the highest value of overall royalties predicted by Professor Marx’s model and the ratio of sound recording to musical work royalties determined by * * * Gans’s analysis.” 84 Fed. Reg. at 1959. That value happens to coincide with the midpoint of the revenue rate’s zone of reasonableness. See id. at 1954. That is not the same as arbitrarily choosing to split the baby between two equally invalid numbers, as occurred in Settling Devotional Claimants, 797 F.3d at 1120. 44

The Streaming Services argue, secondly, that the Board arbitrarily rejected two potential rate benchmarks—the Subpart A settlement and the Phonorecords II settlement—without adequate explanation.

The Subpart A settlement governs the royalty rates and terms for physical phonorecords, permanent digital downloads, and ringtones. 82 Fed. Reg. 15,297, 15,297–15,299 (March 28, 2017); see also 84 Fed. Reg. at 1920. The Board’s decision selected higher revenue rates for the disputed streaming service categories than the Subpart A settlement imposed on the three uncontested categories. Compare J.A. 1407–1410, with 84 Fed. Reg. at 1960. The Streaming Services contend that the Board failed to explain that differential.

Not so. The Board addressed that difference quite directly, explaining that there is “less access value in the sale of a download or a CD, compared to the access value of a subscription to a streaming service[.]” 84 Fed. Reg. at 1946. For that reason, the Board reasonably treated the Part A settlement rates as, “at best,” a floor below which the disputed categories rates should not fall. Id.

In their reply brief, the Streaming Services argue that increased access value from streaming services is exclusively attributable to the streaming service providers’ efforts and so does not justify the payment of higher royalties to copyright holders. Services Reply 23–24. But “an argument first made in a reply brief is forfeited.” Bartko v. SEC, 845 F.3d 1217, 1224 n.7 (D.C. Cir. 2017).

The Streaming Services next argue that the Board failed to keep its promise to “incorporate” the rates from the Subpart A 45

settlement “into the development of a zone of reasonableness of royalty rates within the rate structure adopted[.]” 84 Fed. Reg. at 1947. But the Board stayed true to its word by adopting a zone of reasonableness that was higher than the revenue rate set forth in that settlement. See id. at 1954 (establishing the zone of reasonableness for the revenue prong); see also id. at 1946.

The Board’s treatment of the Phonorecords II settlement, though, is muddled.

In rejecting that settlement as a possible benchmark, the Board faulted the Streaming Services for failing to explain why the parties to the Phonorecords II settlement agreed to the rates in that settlement. 84 Fed. Reg. at 1944 (Board noting the absence of evidence regarding the parties’ negotiations leading up to the adoption of the settlement). The Board also rejected the notion that the Streaming Services’ reliance on the continuation of the Phonorecords II rates alone justified the use of that settlement as a relevant benchmark. Id.

But nowhere does the Final Determination explain why evidence of the parties’ subjective intent in negotiating the Phonorecords II settlement is a prerequisite to its adoption as a benchmark.

On appeal, the Board changes tack and argues that its rejection of the Phonorecords II settlement was reasonable because “[t]he weight of the evidence * * * showed that the prior rates had been set far too low, thus negating the usefulness of the prior settlement as a benchmark.” Board Br. 64. The Board further suggests that it reasonably rejected this benchmark because it was “outdated[.]” Board Br. 65. 46

Those arguments by counsel are nowhere to be found in the Final Determination’s discussion of the appropriateness of the Phonorecords II settlement as a potential benchmark. So we cannot rely on them to sustain that decision. Council for Urological Interests v. Burwell, 790 F.3d 212, 222 (D.C. Cir. 2015) (Court “must judge the propriety of agency action solely by the grounds invoked by the agency.”) (formatting modified) (quoting SEC v. Chenery Corp., 332 U.S. 194, 196 (1947)).

The Copyright Owners, for their part, attempt to defend the Board’s rejection of the Phonorecords II settlement based on the lack of evidence of subjective intent. They argue that the subjective intent of the parties to the Phonorecords II settlement is relevant because it “would have revealed whether the agreed-upon rates were based on economic realities or instead were driven by other considerations.” Copyright Owners Intervenor Br. 28 (Copyright Owners arguing that “streaming was of no economic significance” when the Phonorecords II settlement was adopted, and that they agreed to the settlement to avoid “the distraction of litigating” the issue) (formatting modified). Perhaps. But the Copyright Owners’ post hoc explanation cannot make up for the Board’s failure to adequately explain itself in the Final Determination. See Chenery, 332 U.S. at 196.

Because we cannot discern the basis on which the Board rejected the Phonorecords II rates as a benchmark in its analysis, that issue is remanded to the Board for a reasoned analysis.

Finally, the Streaming Services argue that the Copyright Royalty Board’s determinations failed to adequately consider the four statutory objectives. 47

Recall that Section 801(b)(1) required the Board’s decision to advance four competing priorities: (A) maximizing the availability of creative works, (B) affording copyright owners a fair return and copyright users a fair income, (C) reflecting the relative roles of the copyright owners and users in making music available to the public, and (D) minimizing any disruption on the “structure of the industries involved and on generally prevailing industry practices.” 17 U.S.C. § 801(b)(1) (2012).

Beginning with factor A, the Streaming Services argue that no substantial evidence supports the Board’s conclusion that an increase in the royalty rates for mechanical licenses was necessary “to ensure the continued viability of songwriting as a profession.” Streaming Services Br. 51–52 (quoting 84 Fed. Reg. at 1958). That is incorrect.

The Board found that there was “ample, uncontroverted testimony that songwriters have seen a marked decline in mechanical royalty income over the past two decades,” making it “increasingly difficult for non-performing songwriters * * * to earn a living practicing their craft.” 84 Fed. Reg. at 1957. The Board found that “this decline has led to fewer songwriters” and that, “[i]f this trend continues, the availability of quality songs will inevitably decrease.” Id.

The Streaming Services respond that the question whether declining mechanical license royalties have made it harder for professional songwriters to make a living was, in fact, controverted. Streaming Services Br. 52. They point to testimony by Nashville Songwriters Association International Executive Director Bart Herbison. But that testimony suggests only that streaming may not be the sole cause of the reduction in songwriters’ mechanical license income, while acknowledging that streaming is at least a “complicating 48

factor.” J.A. 456–457 (“I’m not blaming the loss of songwriters on streaming. It is a complicating factor. What I’m saying * * * is streaming doesn’t pay enough as radio goes away, and there [are] no more record sales to allow songwriters to earn a living.”); see J.A. 454–455 (Herbison admitting that mechanical license income had decreased before the rise of music streaming, but stating that the problem worsened after streaming became popular).

The Streaming Services also contend that there was “no evidence in the record that songwriters as a group have diminished their supply of musical works to the public.” Streaming Services Br. 51–52 (quoting 84 Fed. Reg. at 1957). They point to record evidence indicating that the membership and musical works repertoires of two performance-rights organizations—the American Society of Composers, Authors and Publishers and Broadcast Music, Inc.—have grown in recent years. See J.A. 1426–1428.

But an increase in repertoires is not the same as the creation of new songs. In fact, the Streaming Services’ own expert, Mark Zmijewski, testified that he could not tell whether the increases were mostly the result of new songs being written or simply the result of those organizations acquiring the rights to preexisting songs. J.A. 1361–1362.

Nor do the organizations’ increasing memberships necessarily signify an increase in the number of songwriters in the industry. Those associations’ members are not limited to songwriters, and anybody who owns (alone or jointly) the rights to a song—including, for example, a songwriter’s multiple heirs—can register as a member of a performance- rights organization. J.A. 475. 49

To be supported by substantial evidence, the Board’s decision did not have to be irrefutable. See Biestek v. Berryhill, 139 S. Ct. 1148, 1154 (2019). It just had to reflect a reasonable reading of the record. Id. (defining substantial evidence as “such relevant evidence as a reasonable mind might accept as adequate to support a conclusion”) (internal quotation marks omitted). The Board met that test here.

Turning to statutory factors B and C, the Streaming Services argue that the Board failed to consider whether interactive streaming services would receive fair revenue under the rates and rate structure it adopted. Streaming Services Br. 50. They specifically contend that, under the Board’s approach, streaming services will keep less revenue than they should. Streaming Services Br. 50–51. That is because, as the Board recognized, 84 Fed. Reg. at 1952–1953, the sound recordings rightsholders are currently extracting more than their fair share of profits. Streaming Services Br. 50.

As for factor D, the Streaming Services argue that the Board failed to account for the possibility that the new rate structure and heightened rate would eventually result in the elimination of “all existing providers of interactive streaming services” and result in “their substitution with vertically integrated [music] providers[.]” Streaming Services Br. 49.

The question whether the Board adequately addressed factors B through D is bound up with the Board’s analysis of sound recording rightsholders’ likely responses to the new rate structure. See 84 Fed. Reg. at 1953 (Board stating that there is “no basis to assume that record companies will head for the exits” and create their own streaming services rather than lower their royalty rates in response to the new mechanical license rates and rate structure). This argument, in turn, is intertwined with the nature of the rate structure ultimately imposed by the 50

Board. Because we vacate and remand the Final Determination in part for lack of notice to the parties with respect to the final rate structure, we need not at this juncture address whether the Board adequately considered these remaining factors.

The Streaming Services and the Copyright Owners each found something to dislike in the Board’s definitions of certain important terms. The Streaming Services object to the Board’s late-in-the-game reformulation of how “Service Revenue” for bundled offerings was to be calculated. And the Copyright Owners object to the Board’s method for counting the number of subscribers attributable to student and family subscription plans for interactive streaming services. We agree with the Streaming Services but find no merit to the Copyright Owners’ protest.

The Streaming Services challenge both the legal authority and the substantive soundness of the Board’s decision, after it had already issued its Initial Determination, to reformulate the definition of “Service Revenue” for bundled offerings. Because the Board failed to explain the legal authority for its late-breaking rewrite, we vacate and remand that aspect of the decision.

In its Initial Determination, the Board directed that the revenue from streaming services that are included in bundled offerings would generally be measured by the value remaining after subtracting the prices attributable to the other products in the bundle. When the Copyright Owners objected to the substance of that definition in their motion for “clarification,” the Board adopted an entirely new definition of Service 51

Revenue for bundled offerings. J.A. 1267. This new definition generally measured the value of the streaming component of a bundle as the standalone price of the streaming component. J.A. 1267.

The problem is that the Board has completely failed to explain under what authority it was able to materially rework that definition so late in the game.

Section 803(c)(2) of Title 17 deals explicitly with “Rehearings” by the Board after it issues an Initial Determination. 17 U.S.C. § 803(c)(2). That Section provides that the Board “may, in exceptional cases, upon motion of a participant[,] * * * order a rehearing, after the [Initial Determination] is issued * * *, on such matters as the [Board] determine[s] to be appropriate.” Id. § 803(c)(2)(A).

Section 803(c)(4) of Title 17, which is entitled “Continuing Jurisdiction,” separately authorizes the Board to sua sponte “issue an amendment to a written determination to correct any technical or clerical errors in the determination or to modify the terms, but not the rates, of royalty payments in response to unforeseen circumstances that would frustrate the proper implementation of such determination.” 17 U.S.C. § 803(c)(4). Any amendments must “be set forth in a written addendum to the determination that shall be distributed to the participants of the proceeding and shall be published in the Federal Register.” Id.; see also id. § 803(c)(6) (requiring the Librarian of Congress to make public corrections in the same manner as determinations).

So Section 803 identifies three ways in which the Board can revise Initial Determinations. It can (i) order rehearing “in exceptional cases” in response to a party’s motion, 17 U.S.C. § 803(c)(2)(A); (ii) correct “technical or clerical errors,” id. 52

§ 803(c)(4); and (iii) “modify the terms, but not the rates” of a royalty payment, “in response to unforeseen circumstances that would frustrate the proper implementation of [the] determination,” id. The Board’s rollout of an entirely new manner for calculating the streaming service revenue from bundled offerings fit none of those categories.

The Board’s material revision of the “Service Revenue” definition for bundled offerings does not fall within the Board’s rehearing authority under Section 803(c)(2)(A). We have that on no less an authority than the Board itself, which was explicit that it “did not treat the [Copyright Owners’] motion[]” to have the definition changed “as [a] motion[] for rehearing under 17 U.S.C. § 803(c)(2).” 84 Fed. Reg. at 1918 n.2. That is because the Copyright Owners’ motion did not “request[] a literal rehearing of evidence or legal argument.” Id. Nor could they have because, as the Board found, the Copyright Owners’ motion did “not meet [the] exceptional standard for granting rehearing motions” under Section 803(c)(2). J.A. 1251 (Board explaining that the Copyright Owners “failed to make even a prima facie case for rehearing under the [rehearing] standard”).

In a volte-face, the Board now defends its decision as an exercise of its rehearing authority. Oral Arg. Tr. 63:8; see Board Br. 74–75. No dice. It is well-trod ground at this point in our opinion that we may sustain agency action only for the reasons invoked by the agency at the time it took the challenged action. See Department of Homeland Security v. Regents of the Univ. of Cal., 140 S. Ct. 1891, 1909 (2020) (rejecting an agency’s attempt to rely in court on “impermissible post hoc rationalizations” to defend the legality of its action); Federal Power Comm’n v. Texaco, Inc., 417 U.S. 380, 397 (1974) (“We cannot accept appellate counsel’s post hoc rationalizations for agency action; for an agency’s order must be upheld, if at all, on the same basis articulated in the order by the agency itself.”) 53

(formatting modified); Grand Canyon Air Tour Coal. v. Federal Aviation Admin., 154 F.3d 455, 469 (D.C. Cir. 1998) (Administrative Procedure Act review confines courts to “the regulatory rationale actually offered by the agency during the development of the regulation, and not the post-hoc rationalizations of its lawyers.”) (citing cases). An equally forceful corollary is that we cannot sustain action on grounds that the agency itself specifically disavowed.

To be sure, the Board’s order added that, “[t]o the extent the [Board’s] actions could be considered a rehearing under 17 U.S.C. § 803(c)(2), the [Board] further resolve[s] [the] motion[] on the papers without oral argument.” J.A. 1251. One thing is clear from that passive-voice phrasing: Whoever it is that might be “consider[ing]” the decision a “rehearing,” it is not the Board. Although the Board can advance justifications for its decisions in the alternative, it cannot maintain in the same order both that the statutory rehearing standard has not been satisfied and (in the alternative) that the order could be considered as granting a rehearing. So that passing comment by the Board offers no legal justification for treating the Copyright Owners’ request as a successful motion for rehearing.

Neither was the Board’s new definition of Service Revenue for bundled offerings an exercise of its authority under Section 803(c)(4) to “correct any technical or clerical errors in the determination[.]” 17 U.S.C. § 803(c)(4). The Board does not even try to squeeze its substantive rewrite of the Service Revenue definition into that category. Quite the opposite, the Board admits that the new definition “represent[s] a departure” from the definition in the Initial Determination, and was a substantive swap designed to “mitigate” the alleged “problem” of the original definition leaving the interactive streaming service providers free to “obscure royalty-based 54

streaming revenue by offering product bundles that include music service offerings with other goods and services[.]” Board Br. 67–68 (internal quotation marks omitted). To that same point, the order itself labels the initial and new definitions “diametrically-opposed approaches to valuing bundled revenues.” J.A. 1266. Nothing technical or clerical about that.

Nor did the Board’s order purport to “modify the terms * * * in response to unforeseen circumstances that would frustrate the proper implementation of [the Initial] [D]etermination.” 17 U.S.C. § 803(c)(4). The order never mentions Section 803(c)(4) or unforeseen circumstances as the basis for revamping the Service Revenue definition. As the Board agrees, its briefing to this court also did not explain what unforeseen circumstances permitted the term to be modified. See Oral Arg. Tr. 65:15–17 (Board agreeing it “did not specifically focus on unforeseen circumstances” in its briefing defending the revision).

Come oral argument, the Board attempted to explain that “the unforeseen circumstances would be that [it] [initially adopted] a [definition] that was not supported by the record, and that was in fact substantively unreasonable and would frustrate the proper implementation of their [determination].” Oral Arg. Tr. 61:20–24. It is hard to see how the need to ground the original definition in the record was an unforeseen circumstance. That is Administrative Law 101. See also 17 U.S.C. § 803(c)(3) (“A determination of the [Board] shall be supported by the written record[.]”). Anyhow, by this point, it should go without saying that we may not sustain the Board’s action based on its attorney’s theorizing at oral argument. See Regents, 140 S. Ct. at 1906–1909.

Brushing off the absence of any statutory authority for its action, the Board claims the inherent authority sua sponte to 55

make any “appropriate” substantive, J.A. 1251, or “fundamental” changes after the Initial Determination, Oral Arg. Tr. 62:20, that it believes serve “the interests of enhancing the clarity and administrability of the regulatory terms accompanying the [Final Determination].” J.A. 1251. To that end, the Final Determination explains that it treated the Copyright Owners’ request as a general motion under its regulations. 84 Fed. Reg. at 1918; see 37 C.F.R. § 303.4 (“A motion * * * must, at a minimum, state concisely the specific relief the party seeks from the [Board], and the legal, factual, and evidentiary basis for granting that relief[.]”) (formerly codified at § 350.4).

Granted, the Board has “considerable freedom to determine its own procedures.” SoundExchange, 904 F.3d at 61. But that flexibility must be exercised within the lines drawn by the authorizing statute. Congress’s decision to limit rehearing to “exceptional cases,” and to confine other post hoc amendments to cases involving “technical or clerical errors,” would be a nullity if the Board also had plenary authority to revise its determinations whenever it thought appropriate. The Board nowhere in its order or the Final Determination explains the source of its power to make “fundamental” changes under the authorizing statute, Oral Arg. Tr. 62:18–63:6, any time it deems such changes “appropriate,” J.A. 1251, even after the Initial Determination. The Board’s decision said nothing of the sort, and prior Board decisions are silent on that topic. And at oral argument, the Board was equivocal. See Oral Arg. Tr. 63:4–11 (Q: “Is that an inherent power, or is that what you’re putting under [Section 803(c)]? A: “I mean, I think it’s both, right? Sometimes it will fall under the [Section 803](c)(4) [authority], and sometimes it will fall under the [Section 803](c)(2) rehearing power. And I don’t think it’s necessary for this Court to address which one it is because I 56

think it could properly be understood as both.”) (formatting modified).

Vacillating gestures to uninvoked authority will not do. We must vacate the Final Determination’s bundled offering Service Revenue definition and remand for the Board either to provide “a fuller explanation of the agency’s reasoning at the time of the agency action[,]” or to take “new agency action” accompanied by the appropriate procedures. Regents, 140 S. Ct. at 1908 (formatting modified).

Because the Board failed to identify any legal authority for adopting the new Service Revenue definition, we have no occasion to address the Streaming Services’ separate argument that the definition was arbitrary, capricious, or unsupported by substantial evidence.

The Copyright Owners take exception to the Board’s definition of “Subscribers” as applied to student and family streaming plans, which affects the computation of the mechanical floor. Specifically, they object to treating (i) family plan subscriptions as 1.5 subscribers, regardless of the number of family members using the account, and (ii) student plans as 0.5 subscribers. 37 C.F.R. § 385.22(b) (2019); 84 Fed. Reg. at 1962; see Copyright Owners Br. 30.

The Board explained that the assigned valuations match how the interactive streaming services themselves generally price those programs, with family plans set at 1.5 times the normal subscription rate and student plans at 0.5 times the normal subscription rate. See 84 Fed. Reg. at 1961–1962. The Board reasoned that this practice of “marketing reduced rate subscriptions to families and students” was sensibly “aimed at 57

monetizing a segment of the market with a low [willingness to pay] (or ability to pay) that might not otherwise subscribe at all” to a streaming service. Id.

The Copyright Owners’ sole argument is that “the record lacks evidence to support [the] factual premise” that “students and families have a low willingness to pay” for digital music. Copyright Owners Br. 30. That is wrong.

As a reviewing court, we ask only whether the Board’s determination that students and families have a lower willingness (or ability) to pay is “supported by substantial evidence on the record as a whole.” Arkansas v. Oklahoma, 503 U.S. 91, 113 (1992). That is not a high evidentiary bar to clear: “It means—and means only—such relevant evidence as a reasonable mind might accept as adequate to support a conclusion.” Biestek, 139 S. Ct. at 1154; see also Settling Devotional Claimants, 797 F.3d at 1115 (applying “the highly deferential lens of substantial evidence review”). The Board’s finding about the willingness (and ability) of students and families to pay is grounded in substantial record evidence.

For starters, the testimony of multiple witnesses during the ratemaking proceeding supports the Board’s factual findings. For example, Spotify’s expert, Dr. Leslie Marx, specifically touted the greater efficiency attained by offering student and family plans given those groups’ lower willingness to pay for streaming services. J.A. 435–436 (Testifying about the benefits of having “a way for low willingness to pay consumers to access music, for example, student discounts, family discounts[,] * * * where low willingness to pay consumers can still access music in a way that still allows more monetization of that provision of that service.”); see also J.A. 1449 (The continuation of different subscriber offerings “provide[s] an efficient avenue for expanding listening and generating profits 58

from consumers with low willingness to pay,” specifically groups “such as students * * * with a higher elasticity of demand for streaming.”); J.A. 1454–1455 (“[E]conomic efficiency” is furthered “by offering terms, such as student and family discount plans, under which users with a lower [willingness to pay] can participate in the service.”).

Several streaming service providers similarly testified to the benefits of offering student and family plan discounts. One described an internal study that his service had conducted demonstrating “that[,] while a large number of students would not pay [the full monthly price]” for a streaming service, “they would be willing to pay [half of it].” J.A. 1446. He also explained that the study “showed that the additional revenue from students who would sign up with the reduced price but wouldn’t have signed up without it” was greater than “the lost revenue from students who would be willing to pay for the higher price[.]” J.A. 1446.

Another provider explained that family plans have proven helpful to access “younger members of the family [who] don’t have a credit card, don’t have a payment method, are not really in a position to afford a [full price monthly] plan.” J.A. 450– 451 (also explaining that, “for students, it is really more of a value proposition because someone who is going to school is quite often not working and still loves music”). That provider also testified specifically that such individuals have a lower willingness to pay, and that the discounted offerings “allow [the Services] to get more people into the ecosystem to be participants of the subscription service,” and to eventually be funneled into full-priced subscribers. J.A. 451.

Several other streaming service providers testified to similar effect. See, e.g., J.A. 1365 (“[I]t is unlikely that a family of four is going to purchase four separate streaming 59

service subscription plans to the tune of $40 per month, particularly with the widespread availability of fully licensed (and unlicensed) free music,” so “[f]amily subscription plans provide a financial boon for the entire ecosystem[.]”); id. (not offering a family discount plan could lead to a family sharing an individual account at only $10 a month, rather than $15 a month); J.A. 1367 (Students access licensed music for free through platforms like YouTube, “[a]nd the specter of digital piracy still looms[,]” so “[d]iscounted student subscription plans allow [the Services] to” convert “non-paying listeners to paying listeners[,]” “benefit[ing] [copyright owners] by way of increased royalties[.]”); J.A. 416–417 (“So students who have a smaller budget, as long as they are still students, having a student plan that is at a discount, it allows them to be a paying customer, teaches them about paying for music, builds that habit, and then when they graduate and enter the workforce[,] * * * they upgrade.”).

The Copyright Owners object that the testimony was too “speculative” and “conclusory” to support the Board’s decision. Copyright Owners Reply Br. 3. They also point to a study that reached a different conclusion from those witnesses’ testimony, asserting that “[c]ollege students are more willing to pay for music streaming services than non-students.” J.A. 503 (emphasis added).

The Board’s decision needed only to be grounded in substantial evidence, not undisputed evidence. See Settling Devotional Claimants, 797 F.3d at 1117 ( “[A]ll that matters is that we cannot say that the [Board] lacked substantial evidence” in reaching its conclusions.). Finding facts based on the weight and credibility of the evidence falls squarely within the Board’s expertise, and the Copyright Owners have offered no plausible basis for this court to “displace” the Board’s “choice between two fairly conflicting views” of the record 60

evidence. Universal Camera Corp. v. NLRB, 340 U.S. 474, 488 (1951); see Settling Devotional Claimants, 797 F.3d at 1115.

D

Finally, we turn to songwriter George Johnson’s objections to the Board’s mechanical license royalty rates and terms. While thoughtfully presented, none of his arguments succeed.

Johnson’s opening argument is that the Board erred by approving the industry-wide Subpart A settlement. That settlement continued the prior mechanical royalty rate agreed to in 2006—the greater of 9.1 cents per song or 1.75 cents per minute of playing time (or fraction thereof)—for physical phonorecords, permanent digital downloads, and ringtones.

Johnson is the only party that objected to adoption of that settlement agreement. He argues that, instead of continuing those rates, the Board should have adjusted for “unrecognized inflation” the 2-cent mechanical rate originally established in 1909, so that the mechanical rate for Subpart A would be roughly 50 cents. Johnson Br. 13; see also Oral Arg. Tr. 34:6– 15 (Johnson explaining that his inflation argument “is not an all-or-nothing request[,]” and that he would welcome any inflation-adjusted increase of a prior mechanical rate).

While adopting such an inflation-based approach to rate setting might well have been a reasonable option, that is not enough to prevail under the deferential Administrative Procedure Act standard of review. See Department of Commerce v. New York, 139 S. Ct. 2551, 2569 (2019) (Judicial review under the Administrative Procedure Act is narrow, and 61

“[w]e may not substitute our judgment for that of the [agency], but instead must confine ourselves to ensuring that [it] remained within the bounds of reasoned decisionmaking[.]”) (formatting modified); see also Settling Devotional Claimants, 797 F.3d at 1115. The only questions are whether the law required such an inflation adjustment or whether it was unreasonable to omit it. The record establishes neither of those.

Nothing in the Copyright Act compelled the Board to include an inflation adjustment. Rather, the Copyright Act empowers the Board to adopt rates and terms reached in an “agreement * * * among some or all of the participants in a proceeding” as long as (i) the Board affords parties to the proceeding “an opportunity to comment on the agreement and object to its adoption” (and those that would be bound by the terms an opportunity to comment on the agreement); and (ii) the agreement provides a “reasonable basis for setting statutory terms or rates.” 17 U.S.C. § 801(b)(7)(A) (2012) (emphasis added).

Johnson fails to explain why mechanically adjusting prior rates and terms for inflation (from 1909 or otherwise) was the only reasonable approach for the Board to take, or why accepting the parties’ negotiated continuation of the 2006 rates here was unreasonable. As the Board explained, the rates and terms adopted by the settlement were “negotiated on behalf of the vast majority of parties that historically have participated in [ratemaking proceedings] before the [Board].” 82 Fed. Reg. at 15,298. Those parties, including copyright owners like the National Music Publishers’ Association and the Nashville Songwriters Association International, represented “individual songwriters and publishers[,]” and so could be expected to protect their economic self-interest. Id. While Johnson disagreed “[f]rom the perspective of an independent songwriter,” he did not identify any “evidence to support his 62

argument that the representative negotiators [were] engaged in anti-competitive price-fixing at below-market rates.” Id.

For those reasons, the Board reasonably concluded that the stakeholder-negotiated prices continued to reflect “market value”—that is, what “a willing buyer and a willing seller would pay, with neither party being under any compulsion to bargain.” 82 Fed. Reg. at 15,298–15,299. Those representative “parties clearly concluded that the rates and terms were acceptable to both sides[,]” and Johnson presented insufficient evidence and arguments for the Board “to determine that the agreed rates and terms [were] unreasonable.” Id. at 15,299.

Johnson separately argues that the Board erred by allowing “limited download[s]” without compensation to the copyright owners.

By way of explanation, 37 C.F.R. § 385.31(a)–(c) sets the royalty rate for the mechanical license at “zero” in three circumstances. First, this rate applies where a record company that owns a sound recording and has a license to use the musical work authorizes a streaming service to play a particular song without cost (usually for a limited period of time or for a limited number of plays) to promote the song, artist, or album. Those are referred to as “Promotional Offerings.” 37 C.F.R. § 385.31(a) (2019).

Second, the zero mechanical license applies when the playing of the song is part of a “Free Trial Offering[]” of the streaming service, and the service “receives no monetary consideration” from the user. 37 C.F.R. § 385.31(b) (2019). 63

Third, the mechanical license zeroes out when the customer has already purchased the song and is simply playing it through an online digital locker run by the streaming service. See 37 C.F.R. § 385.31(c) (2019); see also 84 Fed. Reg. at 1955. Those are referred to as “Purchased Content Locker Services.” 37 C.F.R. § 385.31(c) (2019).

The Board concluded that it was reasonable in setting the royalty rate for determining the mechanical license “to distinguish promotional or non-revenue producing offerings from” the general “revenue-producing offerings” provided by the streaming services. 84 Fed. Reg. at 1955. With respect to Limited Downloads, 37 C.F.R. § 385.31(a) (2019), the Board emphasized that “[r]ecord companies distributing promotional recordings bear responsibility, if any there be, for the licensing of the embodied musical work.” 84 Fed. Reg. at 1955. As for Free Trial Offerings, 37 C.F.R. § 385.31(b) (2019), the Board reasoned that they were offered by the Services “to entice free users to become paying subscribers after the free trial period.” 84 Fed. Reg. at 1955. And for Purchased Content Locker Services, the customer has already purchased the song and the Streaming Services have already drawn revenue from—and paid royalties for—the purchase price of the song. 37 C.F.R. § 385.31(c) (2019). So additional plays by the purchaser were “free to the [purchaser] and produce[d] no revenue for the Service[s].” 84 Fed. Reg. at 1955.

In those limited circumstances where the Streaming Services gained no revenue from their offering, the Board concluded, it was reasonable not to demand that the streaming service providers “pay mechanical musical works royalties.” 84 Fed. Reg. at 1955. To balance things out, the Board simultaneously prohibited the Streaming Services from “deduct[ing] the costs of those service offerings from [their] 64

revenue, for purposes of calculating royalties payable on a percent of service revenue.” Id.

Johnson fails to explain why the Board’s adoption of those limited and economically balanced exceptions to the generally governing mechanical rates was unreasonable under the circumstances.

Johnson next asserts that the Board erred by not requiring a “BUY Button” on all streaming service platforms. Johnson Br. 14–15. Under Johnson’s proposal, interactive streaming service providers “would be required to include a buy button” alongside songs available for streaming “that allows customers to voluntarily buy or purchase a work as a permanent paid digital download.” 84 Fed. Reg. at 1924 (internal quotation marks omitted).

But the Board reasonably explained that, as relevant here, its role is statutorily confined to establishing royalty rates and terms. See 17 U.S.C. § 115 (2012); 84 Fed. Reg. at 1924. So while the Board recognized that “Services may install a ‘buy button’ if they wish,” the Board itself had no authority to “mandate that service business innovation[.]” 84 Fed. Reg. at 1924. Nor was it clear “what purpose that button would serve other than to alert consumers to the possibility of buying a song they happen to stream[,]” a fact of which the Board believed consumers were “already aware.” Id.

Finally, Johnson asks this court to “re-design” the entire rate structure because it is “based on a faulty business model for streaming.” Johnson Br. 15–16. He proposed a system in which customers “buy the song or the album for a few dollars, 65

then stream all they want at the nano-penny rate[.]” Johnson Br. 15.

That is, perhaps, a fine option for the Board to consider. Which it did here. 84 Fed. Reg. at 1925 & n.23. But, fatally, Johnson does not explain why such a system is compelled under the Copyright Act; why customers who have purchased songs and then play them in the future should still be incurring royalty rates; or how the Board or this court would have the authority to effectively eliminate the asserted “faulty business model for streaming.” Johnson Br. 15. For those reasons, Johnson’s objection on this ground provides no valid basis for setting aside the Board’s actions.

IV

For the foregoing reasons, we affirm in part and vacate and remand to the Board in part for further proceedings consistent with this opinion.

So ordered.