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

                                1

     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.

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, 

, 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., 

, 1004 (D.C.
Cir. 2014); see also 17 U.S.C. § 101; 

. 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, 

, 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

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.

     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, 

, 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 

. 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).

                               2

     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.

no
settlement emerges, the Board presides over a contested royalty
ratemaking proceeding. See

person that the
Board determines has a “significant interest in the proceeding”
may participate. See

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.

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);

(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

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.”

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

                               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.

But the Board still
had to adjudicate the rates and terms for the mechanical license
as applied to permanent downloads and physical phonorecords.

     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.

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

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

with

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

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.

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.

                               1

     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.

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.

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.

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.

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.

                                 2

      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.

                               3

     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

                               4

     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.

     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.

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.

      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.

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.

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, 

, 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.

                                  A

     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

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., 

, 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.

                               1

    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., 

, 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.

                               2

     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., 

, 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 

& 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, 

; see also
Landgraf v. USI Film Products, 

, 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, 

, 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 

).
                               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 

. A provision “does not operate retrospectively merely
because it is applied in a case arising from conduct antedating
[its] enactment.”

–270). Instead, considerations of “fair notice,
reasonable reliance, and settled expectations” of the parties
also carry weight.

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 

(emphasis
added); see Columbia Gas Transmission Corp. v. FERC, 

, 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.

                                B

     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.

                                1

     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., 

, 1083
(D.C. Cir. 2016) (quoting Public Serv. Comm’n of Ky. v.
FERC, 

, 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., 

, 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., 

, 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., 

,
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, 

, 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

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.”);

(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%.

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

(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.

                               2

     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.

                               3

     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.,

, 755 (D.C. Cir. 2009) (quoting East Ky. Power
Coop. v. FERC, 

, 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.

                               a

     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 

. 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.

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.

For the other two years, the Board
simply averaged the values selected for the first two years.

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 

. 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.

     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 

(“[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.”

That gap had empowered
sound recording rightsholders to extract that surplus amount in
their negotiations with the interactive streaming services.

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.

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.

took
the conservative approach of “viewing [Watt’s] lowest
figure * * * as an upper bound[.]”

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.

This involved estimating the ratio of
sound recording royalties to musical works royalties in an
unconstrained market.

     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.”

(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

. 
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

                               b

     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.

                               c

     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.

                                C

     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.

                                1

     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.

                               2

      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.

                               1

     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.

                               2
    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.

                                3

     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.

                                4
     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.