Opinion coneurring-in-part and dissenting-in-part filed by Circuit Judge NEWMAN.
STOLL, Circuit Judge.
SAS Institute, Inc. filed an inter partes review (“IPR”) petition with the Patent Trial and Appeal Board (“Board”) to review the patentability of ComplementSoft’s U.S. Patent No. 7,110,936. The Board instituted an IPR proceeding on some, but not all, of the ’936 patent claims challenged in SAS’s petition. The Board ultimately found all of the instituted claims, except for claim 4, unpatentable in view of the .prior art. SAS argues on appeal that the Board misconstrued a claim term and that the Board erred by not addressing in the final written decision claims SAS petitioned against, but that the Board did not institute as part of the proceeding. ComplementSoft cross-appeals two of the Board’s claim constructions. For the reasons below, we agree with the Board on all of the challenged constructions and determine that the Board did not need to address in its final written decision claims it did not institute. We also vacate the Board’s determination that claim 4 is patentable and remand so that the parties may address a new construction that the Board adopted in its final written decision after interpreting the claim differently before.
Background
I.
ComplementSoft is the assignee of the ’936 patent, issued September 19, 2006, and directed to an “Integrated Development Environment for generating and maintaining source code ... in particular, programmed in data manipulation languages.” ’936 patent col. 2 ll. 8-11. The patent characterizes a development environment as comprising a set of software tools allowing users to develop, edit, and debug software for a particular programming language or set of programming languages. Id. col. 1 ll. 32-48. The development environment contemplated by the ’936 patent utilizes a graphical user interface and is particularly designed for data manipulation languages, including SAS®, which is developed by the appellant. Id. col. 1 l. 64 — col 2 l. 3, col. 2 1. ll. The specification describes that the development environment of the ’936 patent serves three primary functions: (1) it allows users to locally edit code stored on a central server; (2) it detects a user’s programming language and parses code accordingly; and (3) it generates representative visualization of such code, which can be directly edited to effect a change to the underlying code. Id. col. 2 l. 8 - col. 3 1. 20.
The specification describes that four major components of the ’936 patent design environment are a document manager, an editor, a parser layer, and a visualizer. The document manager is a program that performs enhanced file management functions. Id. col. 6 ll. 22-42. The editor allows a user to edit and debug source code using standard text-editing functions. Id. col. 7 l. 3 — col. 8 l. 7. The parser layer examines source code, detects which programming language is being used in the code, and applies rules and logic corresponding to that programming language. Id. col. 9 ll. 38-53, col. 17 ll. 30-45. Finally, the visualizer works in conjunction with the parser layer to parse the code and display it graphically using icons connected with arrows. Id. col. 8 ll. 8-12.
[1344]*1344The ’936 patent discloses two types of visualizations, those for program flows and those for data flows. Program flow diagrams contain programming block icons, which represent sections of source code, linked with arrows that depict the overall flow of the program. Id. col. 2 ll. 38-40, col. 15 ll. 56-59. Figure 9 of the ’936 patent depicts a program flow.
[[Image here]]
Data flow diagrams, the other visualization disclosed by the ’936 patent, “are comprised of icons depicting data processing steps and arrows to depict the flow of the data through the program.” Id. col. 2 ll. 40-42. Figure 17 of the ’936 patent depicts a data flow.
[1345]*1345[[Image here]]
During prosecution, the patentee added the “data manipulation languages” limitation to the claims in response to a prior art rejection based on U.S. Patent No. 6,851,-107 to Coad (“the Coad patent”). The Coad patent generally describes a design environment for purely object-oriented programming languages, such as Java and C + +. In response to the patentee’s amendment, the examiner allowed the claims to issue, stating that the Coad patent does not disclose “that the detected language is a data manipulation language.” J.A. 528.1
The claims at issue in this appeal are independent claim 1 and dependent claim 4. They recite:
1. An integrated development environment, comprising:
a document manager for retrieving source code programmed using one of a plurality of types of data manipulation languages-,
an editor for displaying the retrieved source code and providing a means for a user to edit the retrieved source code;
a parser layer which detects the one of the plurality of types of data manipulation languages in which the retrieved source code is programmed and which activates rules and logic applicable to the detected one of the plurality of types of data manipulation languages; and
[1346]*1346a visualizer dynamically linked to the editor for displaying graphical representations of flows within the retrieved source code using the rules and logic applicable to the detected one of the plurality of types of data manipulation languages and activated by the parser, wherein the editor, parser layer and visualizer cooperate such that edits made to the source code using the editor are automatically reflected in the graphical representations of flows displayed by the visualizer and edits made to the graphical representations of flows in the visualizer are automatically reflected in the source code displayed by the editor. 4. The integrated development environment as recited in claim 1, wherein the graphical representations of data flows are expandable and collapsible.
Id. col. 18 ll. 19 — 43, 50-52 (emphases added).
II.
SAS petitioned for IPR of the ’936 patent, alleging that all sixteen of the patent’s claims were unpatentable as anticipated under 35 U.S.C. § 102 or as obvious under 35 U.S.C. § 103.2 The Board instituted IPR for claims 1 and 3-10 on obviousness grounds, but did not institute IPR for claims 2 and 11-16. Relevant to this appeal, the Board’s institution decision construed “data manipulation language” as “a programming language used to access data in a database, such as to retrieve, insert, delete, or modify data in the database,” and “graphical representations of flows within the retrieved source code” as “a diagram that depicts a map of the progression (or path) through the source code.” SAS Inst., Inc. v. ComplementSoft, LLC, IPR2013-00226, 2013 WL 8595939, at *4-6 (PTAB Aug. 12, 2013) (Institution Decision). The institution decision also interpreted “graphical representations of data flows” to mean “a depiction of a map of the path of data through the executing source code.” Id. at *12.
The Board’s final written decision concluded that claims 1, 3, and 5-10 of the ’936 patent were unpatentable as obvious in view of the prior art. At the same time, the Board found claim 4 patentable over the prior art. Particularly, the Board found that the prior art did not satisfy the “graphical representations of data flows” limitation in claim 4, which it newly construed to mean “a graphical representation comprised of icons depicting data processing steps and arrows to depict the movement of data through source code.”3 The Board’s construction of this purportedly unmet claim limitation differed from the interpretation the Board gave in its institution decision: “a depiction of a map of the path of data through the executing source code.” Institution Decision, 2013 WL 8595939, at *12. The final written decision did not review patentability of claims 2 and 11-16 for which the Board did not institute IPR. See SAS Inst., Inc. v. ComplementSoft, LLC, IPR2013-00226, 2014 WL 3885937, at *24 & n. 3 (PTAB Aug. 6, 2014) (Final Written Decision) (“Claims 2 and 11-16 are not at issue in this trial.”).
[1347]*1347SAS sought rehearing before the Board, which the Board denied. SAS then timely appealed to this court, and Complement-Soft timely cross-appealed. We have jurisdiction under 28 U.S.C. § 1295(a)(4)(A) and 35 U.S.C. § 141(c) to review the Board’s final written decision.
DISCUSSION
On appeal, SAS argues that the Board erred by construing “graphical representations of data flows” in claim 4 as “a graphical representation comprised of icons depicting data processing steps and arrows to depict the movement of data through source code.” SAS also argues that it was improper for the Board to change its interpretation of this claim term in the final written decision without affording the parties an opportunity to respond. SAS lastly argues that the Board’s final written decision is deficient for failing to address the patentability of all claims SAS included in its IPR petition, including those for which the Board did not institute IPR. Complem-entSoft cross-appeals, arguing that the Board erred in construing two terms in claim 1: “data manipulation language” as “a programming language used to access data in a database, such as to retrieve, insert, delete, or modify data in the database”; and “graphical representations of flows within the retrieved source code” as “a diagram that depicts a map of the progression (or path) through the source code.” We first address the parties’ claim construction arguments and then move to SAS’s remaining arguments.
“The ultimate construction of the claim is a legal question and, therefore, is reviewed de novo.” Info-Hold, Inc. v. Applied Media Techs. Corp., 783 F.3d 1262, 1265 (Fed. Cir. 2015). Further, claim construction based solely upon intrinsic evidence — meaning the patent claims, the patent specification, and the prosecution history — is a matter of law reviewed de novo. Teva Pharms. USA Inc. v. Sandoz, Inc., — U.S.-, 135 S.Ct. 831, 841, — L.Ed.2d - (2015). On the other hand, when extrinsic evidence is relied upon, we review the Board’s “underlying factual determinations involving extrinsic evidence for substantial evidence.” Microsoft Corp. v. Proxyconn, Inc., 789 F.3d 1292, 1297 (Fed. Cir. 2015) (citing Teva, 135 S.Ct. at 841-42).
Claim construction seeks to ascribe the meaning to claim terms as a person of ordinary skill in the art at the time of invention would have understood them. Phillips v. AWH Corp., 415 F.3d 1303, 1312-14 (Fed. Cir. 2005) (en banc) (citing Vitronics Corp. v. Conceptronic, Inc., 90 F.3d 1576, 1582 (Fed. Cir. 1996)). In an IPR proceeding, claims are given their broadest reasonable interpretation in light of the specification. In re Cuozzo Speed Techs., LLC, 793 F.3d 1268, 1279 (Fed. Cir. 2015), cert. granted sub nom. Cuozzo Speed Techs., LLC v. Lee, — U.S. -, 136 S.Ct. 890, 193 L.Ed.2d 783 (2015). In construing terms, “the person of ordinary skill in the art is deemed to read the claim term not only in the context of the particular claim in which the disputed term appears, but in the context of the entire patent, including the specification.” Phillips, 415 F.3d at 1313. Indeed, the specification is “the single best guide to the meaning of a disputed term” and “[u]sually, it is dispositive.” Id. Thus, “claims ‘must be read in view of the specification, of which they are a part.’ ” Id. at 1315 (quoting Markman v. Westview Instruments, Inc., 52 F.3d 967, 979 (Fed. Cir. 1995) (en banc), aff'd, 517 U.S. 370, 116 S.Ct. 1384, 134 L.Ed.2d 577 (1996)).
A.
The Board ultimately construed “graphical representations of data flows” as “a [1348]*1348graphical representation comprised of icons depicting data processing steps and arrows to depict the movement of data through source code.” Final Written Decision, 2014 WL 3885937, at *10. In construing this term, the Board recognized that the specification did not use the claim term at all. The Board also recognized, however, that the specification spoke extensively about “data flow diagrams” and, in fact, defined them as comprising “icons depicting data processing steps and arrows to depict the movement of data through source code.” ’936 patent col. 2 ll. 40-42. After determining that there was no reason to conclude that the patentee meant something different between the terms “graphical representations of data flows” and “data flow diagrams,” the Board used the specification’s definition of data flow diagrams to construe graphical representations of data flows.
. We agree with the Bpard’s construction. SAS argues that because the Board’s construction is narrow, it cannot be the broadest reasonable interpretation of the claim term. This is not so. While we have endorsed the Board’s use of the broadest reasonable interpretation standard in IPR proceedings, we also take care to not read “reasonable” out of the standard. This is to say that “[e]ven under the broadest reasonable interpretation, the Board’s construction cannot be divorced from the specification and the record evidence, and must be consistent with the one that those skilled in the art would reach.” Proxyconn, 789 F.3d at 1298 (internal quotation marks omitted) (first quoting In re NTP, Inc., 654 F.3d 1279, 1288 (Fed. Cir. 2011); and then quoting In re Cortright, 165 F.3d 1353, 1358 (Fed. Cir. 1999)). The broadest reasonable interpretation here is that the claimed “graphical representation of a data flow” is commensurate with the “data flow diagram” described in the specification. The Board noted this, concluding that the specification suggests that the terms “data flow diagram” and “graphical representation of data flow” are interchangeable. Final Written Decision, 2014 WL 3885937, at *10; Rehearing Denial, IPR2013-00226, Paper No. 40, at *4. The only difference between these two phrases is the word “diagram” in the first phrase and the word “graphical representation” in the other. And a diagram is, in fact, a graphical representation. Because the specification explicitly defines data flow diagram, one of skill in the art having read the specification would apply this definition to graphical representations of data flows as well.
What the specification also makes clear is that program flows differ from data flows. The Board correctly noted that the specification consistently distinguishes these flows. See Rehearing Denial, IPR2013-00226, Paper No. 40, at *5 (citing ’936 patent col. 2 ll. 38-42, col. 8 ll. 8-14, col. 16 ll. 6-30); see also ’936 patent abstract, col. 2 ll. 33-38, col. 3 ll. 3-5. The specification crystallizes this distinction when it describes a user having to “toggle between the program flow and the data flow display” in the visualizer. ’936 patent col. 15 ll. 50-56. The figures also show that program flows differ from data flows. Compare id. Fig. 9, with id. Fig. 17. To the extent that SAS argues that the specification describes a representation of a program flow as depicting the flow of data, we disagree that one of skill in the art would equate this with a graphical representation of a data flow, given the specification’s consistent disjunction of the two flows. Thus, it is not in error for the Board’s construction of “graphical representation of data flow” in claim 4 to exclude program flows.
The structure of the claims also lends support to the Board’s construction. See Phillips, 415 F.3d at 1314 (“Other claims of the patent in question, both asserted [1349]*1349and unasserted, can also be valuable sources of enlightenment as to the meaning of a claim term”). Independent claim 1 recites broadly “graphical representations of flows,” and dependent claims 2 and 3 recite more specifically that the flows are “data flows” or “program flows,” respectively. ’936 patent col. 18 ll. 19-49. This structure is consistent with the specification, which first discloses visualizations generally and then immediately defines program flow diagrams and data flow diagrams as distinct things. See id. col. 2 ll. 33-42. We therefore reject SAS’s argument that the Board construed the term “graphical representations of data flows” too narrowly.
B.
On cross-appeal, ComplementSoft first challenges the Board’s construction of “data manipulation language” as “a programming language used to access data in a database, such as to retrieve, insert, delete, or modify data in the database.” ComplementSoft argues that this construction should limit the claim term to program languages that are solely purposed for creating datacentric programs and which provide direct access to a database. Particularly, ComplementSoft would have us exclude object-oriented languages altogether — even when implemented with embedded data manipulation code using Structured Query Language (“SQL”) — because the prior art Coad patent the pat-entee distinguished during prosecution discussed object-oriented languages. Com-plementSoft makes this argument under a theory of prosecution history disclaimer.
The prosecution history of a patent, though “less useful for claim construction purposes” than the claim language and written description, plays various roles in resolving uncertainties about claim scope. Phillips, 415 F.3d at 1317. “[F]or prosecution disclaimer to attach, our precedent requires that the alleged disavowing actions or statements made during prosecution be both clear and unmistakable.” Omega Eng’g, Inc., v. Raytek Corp., 334 F.3d 1314, 1325-26 (Fed. Cir. 2003). “Where the alleged disavowal is ambiguous, or even ‘amenable to multiple reasonable interpretations,’ Cordis Corp. v. Medtronic AVE, Inc., 339 F.3d 1352, 1359 (Fed. Cir. 2003), we have declined to find prosecution disclaimer.” Avid Tech., Inc. v. Harmonic, Inc., 812 F.3d 1040, 1045 (Fed. Cir. 2016) (citing Omega Eng’g, 334 F.3d at 1325 (“[W]e have thus consistently rejected prosecution statements too vague or ambiguous to qualify as a disavowal of claim scope.”); Rheox, Inc. v. Entact, Inc., 276 F.3d 1319, 1327 (Fed. Cir. 2002)).
Prosecution history disclaimer does not apply in this case, at least not as Com-plementSoft would have it. There is simply nothing in the prosecution history to suggest that SQL, which the ’936 patent specification specifically identifies as .a data manipulation language, embedded within another programming language does not satisfy the “data manipulation language” limitation. It is true that ComplementSoft added the “data manipulation language” limitation to avoid the Coad patent, but the Coad patent never discloses or suggests embedded data manipulation coding. At most, the patentee disclaimed object-oriented programming languages without any data manipulation components. There is no clear and unmistakable evidence that the patentee disclaimed anything more.
The Board also examined the specification and found that it did not limit data manipulation languages to those providing direct access to a database. In other words, the Board found that the specification does not exclude embedding database access within object-oriented language code. We agree. And because the specification and prosecution history did not con-[1350]*1350elusively resolve construction, we also agree that it was appropriate for the Board to rely on dictionaries and expert testimony to aid its construction. Those sources lend support to the Board’s construction.
ComplementSoft lastly argues that “access” to data in a database is not enough to make a programming language a “data manipulation language.” As the Board noted, ComplementSoft’s own expert and a named inventor on the ’936 patent provided testimony that contradicts this argument. Thus, we agree with the Board that this argument is not compelling. For these reasons, we agree with the Board’s construction of “data manipulation languages.”
ComplementSoft’s second construction challenge relates somewhat to its first. ComplementSoft argues that when the Board construed “graphical representations of flows within the retrieved source code” to mean “a diagram that depicts a map of the progression (or path) through the source code,” it failed to limit the graphical representations to only depicting segments of source code that manipulate data.
Because we have already determined that a “data manipulation language” is not limited to languages that only perform data manipulation, we likewise do not inject a similar limitation into claim 1. Doing so would run counter to the disclosure in the specification. Figures 9, 19, and 20, for instance, are graphical representations of flows. Included within these representations is an icon for a “Print” operation, which is not a data manipulation step. Thus, we agree with the Board’s construction and do not insert an additional limitation that all depicted steps must relate to data manipulation.
As noted above, we agree with the Board’s ultimate construction of the “graphical representations of data flows” claim term. The Board’s procedure for arriving at this construction, however, gives us pause.
In its institution decision, the Board’s claim construction section indicated that “data flow diagram” means “a map of the path of data through the executing source code.” Institution Decision, 2013 WL 8595939, at *5. The Board equated this interpretation of “data flow diagram” with the claim limitation “graphical representations of data flows” when it denied institution of a prior art ground. Particularly, the Board concluded that SAS had not demonstrated that the prior art ground disclosed “a depiction of a map of the path of data through the executing source code” and thus had not shown that the ground “discloses a graphical representation of a data flow.” Id. at *11-12 (internal quotation marks omitted).
ComplementSoft filed its patent owner’s response and identified “a diagram that depicts a map of the path of data through the executing source code” as the Board’s construction for the term “graphical representations of data flows.” While it argued that the Board misconstrued the “data manipulation language” term, it did not similarly argue that the Board misconstrued graphical representations of data flows. SAS’s reply took issue with the construction’s' inclusion of the term “executing,” but suggested no modifications other than to remove this term from the construction. The parties did not ask for a revised construction of “graphical representations of data flows” at the oral hearing.
The Board’s final written decision acknowledged that “the parties directly disagree regarding only the construction of the term ‘data manipulation language.’” [1351]*1351Final Written Decision, 2014 WL 3885937, at *3. Nonetheless, the Board newly construed “graphical representations of data flows” as “a graphical representation comprised of icons depicting data processing steps and arrows to depict the movement of data through source code,” id. at *10, which varies significantly from its initial interpretation of the term as “a map of the path of data through the executing source code.” Institution Decision, 2013 WL 8595939, at *12. In denying SAS’s request for rehearing, the Board concluded that the new construction did not prejudice SAS because SAS could have made construction arguments for the term in its IPR petition. Rehearing Denial, Paper No. 40, at *3-4 n.1.
We disagree with the Board’s approach. As we have noted, IPR proceedings are formal administrative adjudications subject to the procedural requirements of the Administrative Procedure Act '(“APA”). See Dell Inc. v. Acceleron, LLC, 818 F.3d 1293, 1298 (Fed. Cir. 2016); Belden Inc. v. Berk-Tek LLC, 805 F.3d 1064, 1080 (Fed. Cir. 2015); see also Dickinson v. Zurko, 527 U.S. 150, 154, 119 S.Ct. 1816, 144 L.Ed.2d 143 (1999). One such APA provision is that “[p]ersons entitled to notice of an agency hearing shall be timely informed of ... the matters of fact and law asserted.” 5 U.S.C. § 554(b)(3); see Dell, 818 F.3d at 1298. SAS; as the petitioner, is entitled to-this procedural protection in this instance. Although in the past we have discussed § 554(b)(3) with respect to the protection it provides to the patent owner, the provision is not so limited in an instituted IPR proceeding. First, the APA provides that this protection applies to “[p]ersons entitled to notice of an agency hearing.” 5 U.S.C. § 554(b)(3). In an IPR proceeding, this class of persons includes the petitioner. See 35 U.S.C § 316(a)(10) (directing the PTO to promulgate regulations “providing either party — i.e., petitioner or patent owner — “with the right to an oral hearing as part of the proceeding”); 37 C.F.R. § 42.70 (providing that “[a] party may request oral argument” before the Board). Moreover, affording petitioners with the benefit of § 554(b)(3) is appropriate because petitioners are not disinterested parties in an IPR proceeding. Rather, petitioners stand to lose significant rights in an instituted IPR proceeding because of the estoppel effects that trigger against them if the Board issues a final written decision. See 35 U.S.C. § 315(e).
We have interpreted § 554(b)(3) in the context of IPR proceedings to mean that “ ‘an agency may not change theories in midstream without giving respondents reasonable notice of the change’ and ‘the opportunity to present argument under the new theory.’ ” Belden, 805 F.3d at 1080 (quoting Rodale Press, Inc. v. FTC, 407 F.2d 1252, 1256-57 (D.C. Cir. 1968)); see also Dell, 818 F.3d at 1300-01 (holding that the Board, in relying on factual assertions the petitioner introduced for the first time at the oral hearing, violated § 554(b)(3) because the patent owner did not have a meaningful opportunity to respond). That maxim applies in this fact-specific circumstance. What concerns us is not that the Board adopted a construction in its final written decision, as the Board is free to do, but that the Board “change[d] theories in midstream.” Belden, 805 F.3d at 1080. SAS focused its argument on the Board’s institution decision claim interpretation, a reasonable approach considering ComplementSoft agreed with this interpretation in its patent owner’s response and never suggested that the Board adopt the construction that eventually materialized in the final written decision. It is difficult to imagine either party anticipating that already-interpreted terms were actually moving targets, and it is thus unreasonable to expect that they would have briefed or argued, in the alternative, hypothetical constructions not asserted by their oppo[1352]*1352nent. This is especially true for SAS, considering the strict fifteen page limit for its reply to the patent owner’s response. See 37 C.F.R. § 42.24(c)(1) (2012).4
Finally, to be clear, it is uncertain whether SAS will ultimately be able to show unpatentability of the ’936 patent claim 4 even under the construction of “graphical representations of data flows” that the Board adopted and that we agree with. That is not for us to decide today, but for the Board to examine in the first instance after hearing from the parties on the new construction. See Dell, 818 F.3d at 1301 (remanding to the Board for further proceedings upon finding § 554(b)(3) not satisfied).
m.
SAS also argues that the Board erred by not addressing in the final written decision every ’936 patent claim SAS challenged in its IPR petition. The Board’s final written decision, rather, addresses patentability of only those claims for which the Board instituted an IPR proceeding. SAS’s argument, however, is foreclosed by our recent decision in Synopsys, Inc. v. Mentor Graphics Corp., 814 F.3d 1309 (Fed. Cir. 2016).
Synopsys presented the same question that SAS raises here: Must a final written decision by the Board address every patent claim challenged in an IPR petition?5 The petitioner argued, as does SAS, that the text of the final written decision statutory subsection, 35 U.S.C. § 318(a), compels the Board to address every petition-challenged' claim. We found, however, “no statutory requirement that the Board’s final decision address every claim raised in a petition for inter partes review. Section 318(a) only requires the Board to address claims as to which review was granted.” Id. at 1316-17. We found it significant that § 318(a) describes “claims challenged by the petitioner,” whereas the institution decision statutory subsection, 35 U.S.C. § 314, describes “claims challenged in the petition.” We reasoned that the differing language implies a distinction between the two subsections such that § 318(a) does not foreclose the claim-by-claim approach the Board adopted there and in this case. Further, we upheld the validity of a PTO-promulgated regulation authorizing the claim-by-claim approach. Id. at 1316 (validating 37 C.F.R. § 42.108, which “authorize[s] the review to proceed on all or some of the challenged claims”).
Accordingly, we reject SAS’s argument that the Board must address all claims challenged in an JPR petition in its final written decision.
ConolusioN
For the foregoing reasons, we agree with the Board regarding the challenged constructions and conclude that SAS’s argument that the Board must address all claims from the IPR petition in the final written decision is foreclosed by Synopsys. We vacate the Board’s patentability determination of claim 4 and remand so that the [1353]*1353parties may address the Board’s construction of “graphical representations of data flows.”
AFFIRMED-IN-PART, VACATED-IN-PART, AND REMANDED