FOR PUBLICATION
  UNITED STATES COURT OF APPEALS
       FOR THE NINTH CIRCUIT

ZUCCO PARTNERS, LLC; REX                    No. 06-35758
BOGGS; KENNARD MCADAM; GLEN                    D.C. No.
THOMAS,                                    CV-04-01390-AJB
              Plaintiffs-Appellants,
                v.                            ORDER
                                             AMENDING
DIGIMARC CORPORATION; BRUCE                 OPINION AND
DAVIS; E. K. RANJIT,                         AMENDED
             Defendants-Appellees.
                                             OPINION

        Appeal from the United States District Court
                 for the District of Oregon
         Anna J. Brown, District Judge, Presiding

                  Argued and Submitted
           August 26, 2008—Seattle, Washington

                  Filed January 12, 2009
                Amended February 10, 2009

  Before: Thomas G. Nelson, Michael Daly Hawkins, and
              Jay S. Bybee, Circuit Judges.

                  Opinion by Judge Bybee




                            1467
1472           ZUCCO PARTNERS v. DIGIMARC CORP.


                          COUNSEL

David F. Rees, Gary M. Berne, and Mark A. Friel, Stoll Stoll
Berne Lokting & Lokting P.C., Portland, Oregon; Lori G.
Feldman and Karen T. Rogers, Milberg Weiss LLP, New
York, New York, for the plaintiffs-appellants.

Barnes H. Ellis, Lois O. Rosenbaum, and Brad S. Daniels,
Stoel Rives LLP, Portland, Oregon, for the defendants-
appellees.


                           ORDER

   The opinion, originally filed January 12, 2009, slip op. 311,
is amended as follows:

   1. At slip op. 352, first full paragraph, lines 17-18, delete:
“must still be more compelling than an alternative innocent
explanation.” Replace with “must still be at least as compel-
ling as an alternative innocent explanation.”


                          OPINION

BYBEE, Circuit Judge:

   Zucco Partners, LLC and other named plaintiffs (collec-
tively, “Zucco”), on behalf of those who purchased publicly-
                 ZUCCO PARTNERS v. DIGIMARC CORP.                    1473
traded securities of Digimarc Corporation (“Digimarc” or “the
Company”) between April 22, 2003 and July 28, 2004, appeal
the District of Oregon’s dismissal of their Second Amended
Complaint, which alleges that Digimarc (and two of its offi-
cers, Bruce Davis and E. K. Ranjit) violated sections 10(b)
and 20(a) of the Securities Exchange Act of 1934 and the reg-
ulations promulgated thereunder, including Rule 10b-5. Zucco
contends that the district court erred in determining that its
complaint failed to allege a strong inference of scienter as
required by the Private Securities Litigation Reform Act
(“PSLRA”) because that court applied a more stringent stan-
dard than required by the Supreme Court’s recent decision in
Tellabs, Inc. v. Makor Issues & Rights, Ltd., 

(2007). Although we have previously evaluated the suffi-
ciency of such claims under the PSLRA by the standards of
In re Silicon Graphics Inc. Securities Litigation, 

(9th Cir. 1999), and In re Daou Systems, Inc. Securities Liti-
gation, 

(9th Cir. 2005), we have yet to fully
explain how the Court’s Tellabs decision relates to much of
our analysis under those cases.

   The district court determined that, pursuant to Daou, the
plaintiffs’ complaint failed to allege scienter with the requisite
particularity to survive dismissal under the PSLRA’s height-
ened pleading standard. Because we hold that the Court’s
decision in Tellabs does not materially alter the particularity
requirements for scienter claims established in our previous
decisions, but instead only adds an additional “holistic” com-
ponent to those requirements, we affirm the district court’s
dismissal of the complaint with prejudice and hold that Zucco
has failed to adequately plead a strong inference of scienter.1
   1
     Because we find that the district court correctly dismissed Zucco’s
claims for failure to plead scienter, we do not reach the questions of
whether the complaint adequately pleads loss causation, see 

, or whether certain statements relied upon by the complaint
as false representations are forward-looking statements protected from lia-
bility under the PSLRA’s “safe harbor” provision, 15 U.S.C. § 78u-5. See
Employers Teamsters Local Nos. 175 & 505 Pension Trust Fund v. Clorox
Co., 

, 1131-33 (9th Cir. 2004).
1474          ZUCCO PARTNERS v. DIGIMARC CORP.
                               I

   Accounting for the costs of internal software development
is not a simple task. In order to comply with Generally
Accepted Accounting Principles (“GAAP”), a company that
engages in internal software development projects must make
subtle differentiations between three stages of development
that determine whether expenditures incurred must be “ex-
pensed” (recorded immediately on the company’s financial
statement as a cost incurred) or “capitalized” (recorded as a
cost incurred in increments over several financial statements).
This distinction is important because if an expenditure is capi-
talized rather than expensed a company will (in the absence
of other factors) look more profitable in the short term (albeit
less profitable in the long term) and show a more consistent
pattern of reported income—because its expenditures are
spread out over a longer period of time. Under GAAP, if a
software development project is in the “preliminary project
stage,” wherein the company is evaluating development and
marketing alternatives; or in the “post-implementation/
operation stage,” in which the developed software is placed
into service, most expenditures related to the project must be
expensed. If, however, a project is in the “application devel-
opment stage,” in which management authorizes the project
and has settled on a comprehensive development and market-
ing strategy, most expenditures incurred must be capitalized.
Capitalized expenditures are amortized on a straight-line basis
over the estimated useful life of the software developed
(which, for a company like Digimarc, is generally three to
five years).

   According to Zucco, Digimarc, a fledgling Delaware cor-
poration headquartered in Oregon, whose business centers on
providing secure personal identification documents (such as
drivers licenses) based on digital watermarking technology,
purposefully manipulated its financial prospects by, inter alia,
capitalizing internal software development expenditures that
should have been expensed. Zucco’s compendious 130-page
              ZUCCO PARTNERS v. DIGIMARC CORP.             1475
Second Amended Complaint (“SAC”) claims that Digimarc
“used two primary accounting manipulations to deceptively
bolster Digimarc’s financial condition.” Namely, Digimarc
“capitalize[d] on its asset balance sheet ordinary payroll costs
that Digimarc paid to its software engineers and other
employees so that the Company could avoid recognizing these
expenses on its income statements.” Also, Digimarc allegedly
“fail[ed] to recognize ordinary expenses incurred by the Com-
pany” and instead “improperly moved or retained these
expenses in Digimarc’s inventory or property and equipment
accounts as purported ‘project development expenses.’ ” The
net effect of these manipulations, Zucco contends, was to
deceive investors into believing that the young corporation
had “turned the corner” from its early losses and had become
profitable.

   On September 13, 2004, Digimarc publicly announced that
it had erroneously accounted for internal software expendi-
tures and that due to these accounting errors it had likely
overestimated earnings for the previous six quarters. The Sep-
tember announcement listed the improper capitalization of
internal software development costs as the most likely source
of these accounting errors, and also cited “other project cost
capitalization accounting practices” of Digimarc’s ID Systems
division (acquired from Polaroid in December 2001 and
which represented 89 percent of the corporation’s revenue in
2003 and 2004) as containing potential errors that “may also
result in additional adjustments which may affect prior peri-
ods.” On September 13, Digimarc estimated these accounting
errors to “be in the range of approximately $1.2 million to
$2.0 million” and to possibly “require a restatement of prior
period financial statements.”

   Although the full extent of these accounting errors (approx-
imately $2.7 million in overstated earnings) was not revealed
until April 5, 2005, when Digimarc’s formal restatement was
issued, the corporation’s September 2004 announcement was
enough to trigger a number of class action lawsuits. Zucco,
1476           ZUCCO PARTNERS v. DIGIMARC CORP.
which had purchased fifty shares of Digimarc stock in March
2004 (at a price of $12.76 per share) filed a class action law-
suit in the District of Oregon fifteen days after the corpora-
tion’s public announcement, alleging that defendants
Digimarc, its Chief Executive Officer Bruce Davis, and its
former Chief Financial Officer E. K. Ranjit violated sections
10(b) and 20(a) of the Securities Exchange Act of 1934 and
its implementing regulations, including Rule 10b-5. Similar
suits, which followed on October 5th and 6th, were eventually
consolidated with Zucco’s action on December 16, 2004. Two
unrelated actions alleging violations of California corpora-
tions law, meanwhile, were filed in California state court on
October 19th, and subsequently re-filed in the District of Ore-
gon. See In re Digimarc Corp. Derivative Litig., ___ F.3d
___, No. 06-35838, 

(9th Cir. Dec. 11,
2008).

   Although there was no question that Digimarc erroneously
capitalized expenditures that should have been expensed, the
plaintiffs had difficulty providing detailed allegations that the
defendants did so either intentionally or with deliberate reck-
lessness. Indeed, Zucco provided the district court with three
iterations of its allegations—none of which, according to that
court, was sufficient to survive a motion to dismiss. First,
after several additional named plaintiffs were added to its
consolidated class action, Zucco amended its original class
action complaint, adding significant detail to its formerly
skeletal allegations. This First Amended Complaint was filed
on May 16, 2005, on behalf of all those who purchased the
publicly traded securities of Digimarc between April 22, 2003
and July 28, 2004 (the “class period”), and alleged that Digi-
marc and the individual defendants engaged in the manipula-
tive accounting methods described above. Digimarc filed a
Federal Rule of Civil Procedure 12(b)(6) motion to dismiss
the First Amended Complaint, claiming that Zucco had failed
to satisfy the loss causation and scienter requirements of sec-
tion 10(b) of the Securities Exchange Act of 1934, as man-
dated by the Private Securities Litigation Reform Act, 15
              ZUCCO PARTNERS v. DIGIMARC CORP.             1477
U.S.C. § 78u-4. The district court granted this motion on
November 30, 2005. In its order dismissing the complaint, the
district court held that Zucco’s First Amended Complaint had
satisfied the loss causation pleading requirements, but had
failed to properly allege scienter. See Zucco Partners, LLC v.
Digimarc Corp., No. CV 04-1390-BR (D. Or. Nov. 30, 2005).

   The district court dismissed the complaint without preju-
dice, giving Zucco leave to amend. According to the district
court, the Second Amended Complaint was no better. After
that complaint was filed on January 17, 2006, Digimarc
responded with another motion to dismiss, contending that
Zucco had again failed to plead scienter adequately under the
PSLRA. This motion was granted on August 4, 2006, when
the district court dismissed the complaint with prejudice. See
Zucco Partners, LLC v. Digimarc Corp., 

(D. Or. 2006). After dismissal, Zucco filed a timely
appeal to this Court.

                               II

   Zucco argues that the district court failed to properly ana-
lyze its allegations of scienter under the standard recently
expounded by the Supreme Court in Tellabs, Inc. v. Makor
Issues & Rights, Ltd., 

(2007). We review
challenges to a dismissal for failure to state a claim under
Federal Rule of Civil Procedure 12(b)(6) de novo. Livid Hold-
ings, Ltd. v. Solomon Smith Barney, Inc., 

, 946
(9th Cir. 2005). Such review is generally limited to the face
of the complaint, materials incorporated into the complaint by
reference, and matters of which we may take judicial notice.
Metzler Inv. GMBH v. Corinthian Colls., Inc., 

,
1061 (9th Cir. 2008) (citing 

). In
undertaking this review, we will “accept the plaintiffs’ allega-
tions as true and construe them in the light most favorable to
plaintiffs,” Gompper v. VISX, Inc., 

, 895 (9th
Cir. 2002), and will hold a dismissal inappropriate unless the
plaintiffs’ complaint fails to “state a claim to relief that is
1478           ZUCCO PARTNERS v. DIGIMARC CORP.
plausible on its face.” Bell Atl. Corp. v. Twombly, 

, 1974 (2007). Where, as here, the district court dis-
misses the complaint without leave to amend, such prejudicial
dismissal is reviewed for abuse of discretion, see 

, and “is improper unless it is clear that the
complaint could not be saved by any amendment.” Livid

.

                                A

   [1] Section 10(b) of the Securities Exchange Act of 1934
makes it unlawful for “any person . . . [t]o use or employ, in
connection with the purchase or sale of any security registered
on a national securities exchange . . . any manipulative or
deceptive device or contrivance in contravention of such rules
and regulations as the Commission may prescribe as neces-
sary or appropriate in the public interest or for the protection
of investors.” 15 U.S.C. § 78j(b). One such rule promulgated
under the Act is SEC Rule 10b-5, which provides, inter alia,
“It shall be unlawful for any person . . . [t]o engage in any act,
practice, or course of business which operates or would oper-
ate as a fraud or deceit upon any person, in connection with
the purchase or sale of any security.” 17 C.F.R. § 240.10b-
5(c).

  Section 20(a) of the Act makes certain “controlling” indi-
viduals also liable for violations of section 10(b) and its
underlying regulations. Specifically, section 20(a) provides:

    Every person who, directly or indirectly, controls
    any person liable under any provision of this chapter
    or of any rule or regulation thereunder shall also be
    liable jointly and severally with and to the same
    extent as such controlled person to any person to
    whom such controlled person is liable, unless the
    controlling person acted in good faith and did not
    directly or indirectly induce the act or acts constitut-
    ing the violation or cause of action.
               ZUCCO PARTNERS v. DIGIMARC CORP.              1479
15 U.S.C. § 78t(a). Thus, a defendant employee of a corpora-
tion who has violated the securities laws will be jointly and
severally liable to the plaintiff, as long as the plaintiff demon-
strates “a primary violation of federal securities law” and that
“the defendant exercised actual power or control over the pri-
mary violator.” No. 84 Employer-Teamster Joint Council
Pension Trust Fund v. Am. W. Holding Corp. (“America
West”), 

, 945 (9th Cir. 2003) (quoting Howard
v. Everex Sys., Inc., 

, 1065 (9th Cir. 2000))
(quotation marks omitted); Paracor Fin., Inc. v. Gen. Elec.
Capital Corp., 

, 1161 (9th Cir. 1996). This
inquiry is normally an “intensely factual question.” Paracor

(quoting Arthur Children’s Trust v.
Keim, 

, 1396 (9th Cir. 1993)). Section 20(a)
claims may be dismissed summarily, however, if a plaintiff
fails to adequately plead a primary violation of section 10(b).
See In re Verifone Sec. Litig., 

, 872 (9th Cir.
1993). See, e.g., In re Metawave Commc’ns Corp. Sec. Litig.,

, 1087 (W.D. Wash. 2003).

   Five elements are required in order to prove a primary vio-
lation of Rule 10b-5. In particular, a plaintiff must demon-
strate “(1) a material misrepresentation or omission of fact,
(2) scienter, (3) a connection with the purchase or sale of a
security, (4) transaction and loss causation, and (5) economic
loss.” 

. At the pleading stage, a com-
plaint stating claims under section 10(b) and Rule 10b-5 must
satisfy the dual pleading requirements of Federal Rule of
Civil Procedure 9(b) and the PSLRA.

   Federal Rule of Civil Procedure 9(b) provides, “In alleging
fraud or mistake, a party must state with particularity the cir-
cumstances constituting fraud or mistake. Malice, intent,
knowledge, and other conditions of a person’s mind may be
alleged generally.” This requirement has long been applied to
securities fraud complaints. See Semegen v. Weidner, 

, 729, 734-35 (9th Cir. 1985). Accordingly, before
1995 we required “falsity” to be pled with particularity, and
1480           ZUCCO PARTNERS v. DIGIMARC CORP.
“scienter” to be alleged generally. See Ronconi v. Larkin, 

, 429 n.6 (9th Cir. 2001).

   [2] All securities fraud complaints since 1995, however, are
subject to the more exacting pleading requirements of the
PSLRA, which “significantly altered pleading requirements”
in securities fraud cases. 

(quotation
marks omitted). The PSLRA amended the Securities
Exchange Act to require that a complaint “plead with particu-
larity both falsity and scienter.” 

253
F.3d at 429). Thus, to properly allege falsity, a securities fraud
complaint must now “specify each statement alleged to have
been misleading, the reason or reasons why the statement is
misleading, and, if an allegation regarding the statement or
omission is made on information and belief, . . . state with
particularity all facts on which that belief is formed.” 

U.S.C. § 78u-4(b)(1))(quotation marks omitted).
To adequately plead scienter, the complaint must now “state
with particularity facts giving rise to a strong inference that
the defendant acted with the required state of mind.” 15
U.S.C. § 78u-4(b)(2). See also Ernst & Ernst v. Hochfelder,

, 193 (1976).

   [3] The Supreme Court recently defined “strong inference”
in Tellabs, concluding that a securities fraud complaint will
survive a motion to dismiss under Federal Rule of Civil Pro-
cedure 12(b)(6) “only if a reasonable person would deem the
inference of scienter cogent and at least as compelling as any
opposing inference one could draw from the facts 

(emphasis added). Thus, a court now
reviewing a complaint’s scienter allegations under the PSLRA
must “consider the complaint in its entirety, as well as other
sources courts ordinarily examine when ruling on Rule
12(b)(6) motions to dismiss, in particular, documents incorpo-
rated into the complaint by reference, and matters of which a
court may take judicial notice.” 

The court must
determine whether “all of the facts alleged, taken collectively,
give rise to a strong inference of scienter, not whether any
               ZUCCO PARTNERS v. DIGIMARC CORP.             1481
individual allegation, scrutinized in isolation, meets that stan-
dard.” 

“determining whether the pleaded
facts give rise to a ‘strong’ inference of scienter, the court
must take into account plausible opposing inferences.” 

is inherently comparative.” 

A court
must compare the malicious and innocent inferences cogniza-
ble from the facts pled in the complaint, and only allow the
complaint to survive a motion to dismiss if the malicious
inference is at least as compelling as any opposing innocent
inference. See 

See also Metzler 

.

   To adequately demonstrate that the “defendant acted with
the required state of mind,” a complaint must “allege that the
defendants made false or misleading statements either inten-
tionally or with deliberate recklessness.” 

-15 (citing Silicon 

). In Silicon
Graphics, we defined the “deliberate recklessness” standard,
noting that “we continue[ ] to view it as a form of intentional
or knowing 

. More specifically,
“although facts showing mere recklessness or a motive to
commit fraud and opportunity to do so may provide some rea-
sonable inference of intent, they are not sufficient to establish
a strong inference of deliberate recklessness.” 

Rather, the plaintiff must plead “a highly unreasonable omis-
sion, involving not merely simple, or even inexcusable negli-
gence, but an extreme departure from the standards of
ordinary care, and which presents a danger of misleading buy-
ers or sellers that is either known to the defendant or is so
obvious that the actor must have been aware of it.” 

(quoting Hollinger v. Titan Capital Corp., 

,
1569 (9th Cir. 1990); Sundstrand Corp. v. Sun Chem. Corp.,

, 1045 (7th Cir. 1977)) (quotation marks omit-
ted).

   [4] Although we have developed a set of rules to analyze
different types of scienter allegations, we recognize that Tel-
labs calls into question a methodology that relies exclusively
1482           ZUCCO PARTNERS v. DIGIMARC CORP.
on a segmented analysis of scienter. We read Tellabs to mean
that our prior, segmented approach is not sufficient to dismiss
an allegation of scienter. Although we have continued to
employ the old standards in determining whether, a plaintiff’s
allegations of scienter are as cogent or as compelling as an
opposing innocent inference, see, e.g., Metzler 

, we must also view the allegations as a
whole. See South Ferry LP, No. 2 v. Killinger, 

,
784 (9th Cir. 2008) (“Tellabs counsels us to consider the
totality of the circumstances, rather than to develop separately
rules of thumb for each type of scienter allegation.”). Thus,
following Tellabs, we will conduct a dual inquiry: first, we
will determine whether any of the plaintiff’s allegations,
standing alone, are sufficient to create a strong inference of
scienter; second, if no individual allegations are sufficient, we
will conduct a “holistic” review of the same allegations to
determine whether the insufficient allegations combine to
create a strong inference of intentional conduct or deliberate
recklessness.

                               B

   The SAC relies on several types of factual allegations to
plead the requisite intentional or deliberately reckless con-
duct, including (1) statements of six confidential witnesses,
(2) Digimarc’s April 5, 2005 restatement of earnings, (3) the
resignations of Ranjit, two members of the accounting depart-
ment, and the corporation’s auditing firm during the class
period, (4) statements made in filing the corporation’s
Sarbanes-Oxley certifications, (5) the compensation packages
of the individual defendants, (6) the stock sales of the individ-
ual defendants occurring during the class period, and (7) a pri-
vate placement by the corporation during the class period. We
address each of these allegations in turn, and then, as Tellabs
instructs, consider the allegations collectively to determine
whether the complaint as a whole raises a strong inference of
scienter.
              ZUCCO PARTNERS v. DIGIMARC CORP.             1483
                               1

   Zucco first alleges scienter with respect to Digimarc’s
improper capitalization of internal software development
costs, including capitalization of the payroll costs of software
engineers and other personnel. The SAC contends that Digi-
marc and its senior management deliberately capitalized inter-
nal software development costs, including payroll costs, that
they knew should have been expensed under GAAP. This
improper accounting, according to the SAC, occurred when
Digimarc capitalized, rather than expensed, internal software
development costs for software projects in the “preliminary
stage” (where a company is in the process of evaluating alter-
natives for the software development) and in the “post-
implementation/operation stage” (when the company places
the software into service). To support this allegation of
scienter, the SAC relies primarily on the statements of six
confidential witnesses.

   According to Confidential Witness # 2 (“CW2”), in early
2003 Digimarc (and CFO Ranjit in particular) “eliminated
internal control processes intended to prevent the improper
capitalization of payroll expenses.” CW2 states that in early
2003, he was personally instructed by Val Ford, who reported
to Ranjit, to “reprogram the time and expense software to
eliminate the required supervisor or manager approval of
time” and to give administrative assistants in the finance
department authority to enter time for software engineers.
Moreover, CW2 reports that Confidential Witness # 1
(“CW1”) sent an email to Ranjit about the changes to the
time-reporting software, and received in return an email say-
ing “that it was what Ranjit wanted and that he expected CW1
to take care of it.” Confidential Witness # 4 (“CW4”), who
worked at the company earlier, recalls that this accounting
behavior was typical—specifically, that “decisions about what
to capitalize were made either by highly-placed finance
employees, or made by corporate-level employees at Digi-
marc and Digimarc ID Systems.”
1484          ZUCCO PARTNERS v. DIGIMARC CORP.
   Confidential Witness # 3 (“CW3”) indicates that Jennifer
Walden, a Digimarc financial analyst, told him/her that Indra
Paul, the President of Digimarc’s ID Systems business unit,
instructed employees “to assign more payroll time to projects
that were at a point where the payroll costs could be capital-
ized, even if the employee’s time was not spent working on
the projects that were appropriate for capitalization.” Addi-
tionally, CW3 states that several unnamed project managers
told him/her that they were upset that their time was being
assigned improperly, and that this was occurring “merely so
that this time could be capitalized and not recognized as
expenses, thereby improperly and falsely boosting the Com-
pany’s reported income.”

   Confidential Witness # 5 (“CW5”) recounts that, following
his/her departure from Digimarc, “he/she had discussions with
employees who still worked at the Company, and heard that
time was being changed and hours reassigned improperly,
resulting in more payroll expense being capitalized improp-
erly as assets.” CW5 also reports hearing “once or twice in
mid-2003, from finance people,” that some expenses not
directly related to a program were being charged to that pro-
gram. Confidential Witness # 6 (“CW6”) reports he/she “un-
derstood that some employees were improperly capitalizing
labor on contracts.”

   CW1 states that Susan Scacchi, a former ID Systems con-
troller who reported directly to Ranjit, told him/her that, Ran-
jit once ordered Scacchi to make last-minute journal entries
that increased the amount of payroll that would be capitalized.
CW1 reports that Scacchi objected, and later left as a result
of this incident—and that Martin Day, then Digimarc’s Con-
troller, told CW1 that the journal entries were made despite
Scacchi’s objections. CW6, in corroboration, describes how
Scacchi worked at Digimarc for only six weeks, and says that
Scacchi announced publicly that she was leaving because her
supervisors “asked her to do some things she believed were
unethical.”
              ZUCCO PARTNERS v. DIGIMARC CORP.           1485
   Finally, CW1 reports that Ranjit, at the close of every
accounting period, required the IT Department to copy all the
data from the corporation’s Great Plains accounting system
into spreadsheets, so that management could analyze the data.
CW1 asserts that the “real purpose” of this data extraction
was to allow management to determine what “adjustments”
were necessary to meet analysts’ expectations.

   The SAC also employs confidential witnesses to support its
allegations of scienter with respect to Digimarc’s improper
capitalization of inventory and fixed assets. According to
Zucco, the corporation misrepresented its inventory numbers
by (1) using an improperly low “scrap rate” to calculate
inventory lost over the relevant period, (2) creating separate
Access databases to track inventory that could be manipulated
by Digimarc’s officers, (3) booking purchases to projects that
would require the purchases to be capitalized, rather than
expensed, and (4) leaving obsolete and overvalued inventory
on its balance sheet.

   The SAC alleges that “Digimarc purchased raw materials
that went through several production processes with numer-
ous vendors before [they were] ready for use, and that each
production process resulted in a loss of some of the materials
being processed.” When accounting for this loss of materials,
the SAC contends that the “scrap rate” used to estimate the
loss of raw materials was “unreasonably low.” CW1 states
that he/she calculated the “correct” scrap rate (60 percent),
met with a consultant for the company’s accounting system
and Rahoul Banerja (then Digimarc’s ID Systems’ Vice Presi-
dent of Finance), and informed both that “the Company had
a problem with ‘seriously overvalued inventory’ ” and that
“the Company was not accounting for scrap properly.” CW1
also recalls meeting frequently with senior management to
discuss the scrap rate, and sending senior management a
detailed email describing why his/her scrap rate was prefera-
ble to Digimarc’s current scrap rates (between five and thirty
percent).
1486          ZUCCO PARTNERS v. DIGIMARC CORP.
   Second, the SAC alleges that Digimarc set up “databases to
track inventory separate from the Company’s Great Plains
accounting system,” and then used these databases, rather
than the Great Plains system, to record its inventory values at
the end of each quarter. According to CW1 and CW2, Ranjit
and others in charge of these separate databases used them to
improperly increase the amount of inventory reported by
Digimarc. CW1 and CW2 claim that “Ford and other finance
department personnel munged [sic] the data in the separate
Access databases at the end of the month” to boost its
reported income. Also, CW6 reports that Banerjea was “ac-
cessing Digimarc’s inventory accounts on his own and manip-
ulating the value of the Company’s inventory.” CW6 recalls
that Banerjea met frequently with Ranjit and other manage-
ment to discuss “the Company’s problems accounting for
inventory,” and that Ranjit “had to have known what was
going on with respect to the Company’s inventory accounting
manipulation.”

   According to several confidential witnesses, these Access
databases were used despite reassurances from Microsoft (the
creator of the Great Plains accounting software) that the Great
Plains system was producing accurate numbers. CW1 reports
meeting with Ranjit and other senior management several
times in Fort Wayne, Indiana, and discussing the inventory
valuation and the use of the Great Plains system. CW1 also
recounts talking over the telephone with a Microsoft represen-
tative about the Great Plains system while Ranjit and Baner-
jea observed. During this phone conversation, “the Microsoft
consultant walked through the process of how the inventory
numbers were created in the Great Plains system and con-
firmed that the numbers were correct.”

   The SAC additionally alleges that Digimarc booked pur-
chases of materials “to the wrong project so that the purchases
could be capitalized, rather than expensed, thereby improving
the Company’s bottom line.” Specifically, CW3 reports that
Claudia Rao, a former staff accountant in the ID Systems
              ZUCCO PARTNERS v. DIGIMARC CORP.           1487
division, told CW3 that “she was directed by senior manage-
ment (who were in turn commanded by Defendant Ranjit) to
improperly book [materials] purchases in an account for a
drivers licence program that was still in the [application
development] stage so that the purchase expense could be
capitalized.” Rao also told CW3 that Rao “was never
instructed to close out projects with capitalized expenses and
therefore the research and development expenses that were
being capitalized were improperly accumulating on the Com-
pany’s balance sheet.”

    Finally, Zucco contends that Digimarc left obsolete and
overvalued inventory on its balance sheet, thereby over-
reporting the value of inventory Digimarc purportedly owned.
According to CW1, who early in 2004 became Director of
Supply Chain Management, “the Company had approximately
$2 million in book value of cameras and laminate inventory
in its warehouse that was obsolete and substantially overva-
lued on the Company’s balance sheet.” Also, CW1 states that
Digimarc reported $130,000 of inventory value on its books
for worthless laminate inventory that had been shredded.
CW1 says he/she was specifically ordered by Ranjit “not to
write down obsolete inventory because, according to Ranjit
. . . writing down obsolete inventory would result in the Com-
pany missing market expectations.”

                             ***

   As we have explained in Daou, a complaint relying on
statements from confidential witnesses must pass two hurdles
to satisfy the PSLRA pleading requirements. First, the confi-
dential witnesses whose statements are introduced to establish
scienter must be described with sufficient particularity to
establish their reliability and personal knowledge. 

(citing Nursing Home Pension Fund, Local
144 v. Oracle Corp., 

, 1233 (9th Cir. 2004); Sil-
icon 

). Second, those statements
which are reported by confidential witnesses with sufficient
1488              ZUCCO PARTNERS v. DIGIMARC CORP.
reliability and personal knowledge must themselves be indica-
tive of scienter. See 

   [5] The first prong of this two-part confidential witness test
analyzes whether a complaint has provided sufficient detail
about a confidential witness’ position within the defendant
company to provide a basis for attributing the facts reported
by that witness to the witness’ personal knowledge. In Daou,
we explicitly adopted this approach as an amalgam of the
“Second Circuit’s standard for evaluating personal sources of
information” and “the First Circuit’s suggested criteria for
assessing reliability of confidential 

. Under this approach, therefore, a complaint may rely on
confidential witnesses in two situations. Where a complaint
relies on both confidential witnesses and other factual infor-
mation, such as documentary evidence, the plaintiffs “ ‘need
not name their sources as long as the latter facts provide an
adequate basis for believing that the defendants’ statements
were false.’ ” 

v. Kasaks, 

,
314 (2d Cir. 2000)).

  [6] Where as here, however, such additional evidence is
absent, confidential witness statements may only be relied
upon where the confidential witnesses are described “with
sufficient particularity to support the probability that a person
   2
     Some circuits have questioned whether Tellabs requires a stricter stan-
dard for evaluating the sufficiency of securities fraud complaints relying
on confidential witnesses. See Ind. Elec. Workers Pension Trust Fund
IBEW v. Shaw Group, Inc., 

, 535 (5th Cir. 2008) (“Following
Tellabs, courts must discount allegations from confidential sources.”);
Higginbotham v. Baxter Int’l, Inc., 

, 756-57 (7th Cir. 2007)
(Tellabs requires courts to “discount allegations that the complaint attri-
butes to . . . ‘confidential witnesses’ ” because “it is hard to see how infor-
mation from anonymous sources could be deemed ‘compelling’ or how we
could take account of plausible opposing inferences” when the confiden-
tial witnesses could be lying or nonexistent). Because we conclude the
complaint here fails the PSLRA’s strong inference requirement even under
our standard in Daou, we need not decide whether Tellabs requires us to
further “discount” the use of confidential witness statements.
               ZUCCO PARTNERS v. DIGIMARC CORP.             1489
in the position occupied by the source would possess the
information alleged.” 

, 216 F.3d at 314).
Accordingly, the complaint must provide an adequate basis
for determining that the witnesses in question have personal
knowledge of the events they report. 

re
Cabletron Sys., Inc., 

, 29 (1st Cir. 2002)). To
determine whether the complaint has done so, we look to
“ ‘the level of detail provided by the confidential sources, the
corroborative nature of the other facts alleged (including from
other sources), the coherence and plausibility of the allega-
tions, the number of sources, the reliability of the sources, and
similar indicia.’ ” 

311 F.3d at 29-30).

    Although the SAC describes the confidential witnesses’ job
titles and employment information with ample detail to satisfy
Daou’s requirement that a complaint make apparent a confi-
dential witnesses’ position within the defendant corporation,
we conclude that the SAC fails to allege with particularity
facts supporting its assumptions that the confidential wit-
nesses were in a position to be personally knowledgeable of
the information alleged.

   As in Daou, the plaintiffs here “describe the confidential
witnesses with a large degree of specificity”—the SAC “num-
ber[s] each witness and describe[s] his or her job description
and responsibilities.” 

CW1 is described as “Direc-
tor of the IT Department of Digimarc’s ID Systems” who “re-
port[ed] directly to . . . Ranjit.” He is alleged to have “worked
extensively on Digimarc’s accounting system software” and
on Digimarc’s attempt to “track[ ] and value[ ] inventory.”
CW2 is identified as an anonymous IT employee, who
“worked extensively on Digimarc’s implementation of the
Great Plains accounting system, and had direct knowledge of
how Digimarc accounted for its inventory during the [relevant
time period].” CW3 was purportedly a “financial analyst in
Digimarc’s ID Systems finance department” who “worked
closely with other finance department employees involved in
creating forecasts of Digimarc’s financial performance” and
1490            ZUCCO PARTNERS v. DIGIMARC CORP.
“worked extensively attempting to resolve the Company’s
serious accounting problems regarding its inventory in 2004.”3
CW4, according to the SAC, was a Human Resources man-
ager until 2002 who reported to John Munday, a former ID
Systems President. CW5 was ostensibly the ID Systems Con-
troller who reported to Ranjit from December 2001 until mid-
2002. Finally, CW6 was the Vice President of Supply Chain
Management at the ID Systems unit from 2002 until early
2005. He was based in Fort Wayne, Indiana, and was “in-
volved in tracking inventory, and purchasing and distribution
of materials.” Such descriptions are undoubtedly sufficient
under Daou.

   [7] The SAC, however, does not provide the requisite par-
ticularity to establish that certain statements of these confi-
dential witnesses are based on the witnesses’ personal
knowledge. See 

(validating the con-
fidential source statements in a securities fraud complaint
only when they “are not conclusory allegations of fraud, but
specific descriptions of the precise means through which it
occurred, provided by persons said to have personal knowl-
edge of them”). Some of the confidential witnesses were sim-
ply not positioned to know the information alleged, many
report only unreliable hearsay, and others allege conclusory
assertions of scienter. These allegations are not sufficient to
raise a strong inference of scienter because they demonstrate
that the confidential witnesses are not reliable. See 

.

  Two of the witnesses, CW4 and CW5, were not employed
by Digimarc during the time period in question and have only
secondhand information about accounting practices at the cor-
  3
    Although the SAC does not reveal the exact job titles of CW2 and
CW3, this is not fatal to the complaint. In Daou, we held that plaintiffs
had pled facts sufficient to meet the PSLRA standard when plaintiffs
relied on six confidential witnesses described in several cases simply by
business department and general duties. 

.
                  ZUCCO PARTNERS v. DIGIMARC CORP.                      1491
poration during that year. CW4 was a human-resources
employee who, even while employed at Digimarc, had no
firsthand knowledge of the workings of the finance or corpo-
rate departments. He supplies no basis for his claim that “de-
cisions about what to capitalize were made . . . by highly-
placed finance employees.” Likewise, although CW5 was the
ID Systems Controller for at least six months between 2001
and 2002, and may have had some personal knowledge about
the inner workings of Digimarc’s accounting and financial
reporting at that time, the only facts he reports to indicate
scienter are hearsay statements from anonymous finance per-
sonnel employed by Digimarc in mid-2003. The lack of detail
in CW5’s allegations (he claims only that sometime in 2003,
“some expenses not directly related to a program were
charged to that program” and that “[employee] time was
being changed,” and fails to provide specifics or dates for
these activities) further support the conclusion that CW5’s
allegations are not reliable enough to support the SAC’s alle-
gations of scienter. See 

.

   A majority of the confidential witnesses base their knowl-
edge on vague hearsay, which is not enough to satisfy Daou’s
reliability standard.4 For example, CW3 reports that Jennifer
Walden, a Digimarc financial analyst, told him/her that Indra
Paul, the President of Digimarc’s ID Systems business unit,
  4
    We agree with Zucco that, under the Court’s test in Tellabs, which
explicitly rejected any standard which would “transpose to the pleading
stage the test that is used at the summary judgment and judgment-as-a-
matter-of-law 

n.5 (quotation marks omitted),
the fact that a confidential witness reports hearsay does not automatically
disqualify his statement from consideration in the scienter calculus. See

(noting that “the rigorous standards for pleading
securities fraud do not require a plaintiff to plead evidence.”). However,
a hearsay statement, while not automatically precluded from consideration
to support allegations of scienter, may indicate that a confidential wit-
nesses’ report is not sufficiently reliable, plausible, or coherent to warrant
further consideration under Daou. See Daou, 411 at 1015 (quoting

).
1492             ZUCCO PARTNERS v. DIGIMARC CORP.
instructed employees “to assign more payroll time to projects
that were at a point where the payroll costs could be capital-
ized, even if the employee’s time was not spent working on
the projects that were appropriate for capitalization.” This tri-
ple hearsay, accompanied by no details describing which proj-
ects or what employees were affected, is not detailed enough
to pass muster under Daou. Likewise, CW1’s statement that
Susan Scacchi, a former ID Systems controller who reported
directly to Ranjit, told him/her that, “at the close of the first
quarter of 2003 . . . Ranjit ordered [Scacchi] to make last-
minute journal entries that improperly increased the amount
of payroll that was capitalized” is based on at least one level
of hearsay, and includes no details specifying the nature of
these entries. CW1’s additional hearsay statements (that Mar-
tin Day, then Digimarc’s Controller, told CW1 that the jour-
nal entries were made despite Scacchi’s objections, and that
Raoul Banerjea told CW1 that Digimarc had eliminated
inventory reserves at the bequest of Ranjit to meet market expec-
tations5) are similarly vague and unreliable, as are CW3’s
statements that Claudia Rao, a Digimarc staff accountant, told
him/her that she was instructed by Ranjit to book purchases
“improperly.” Even more unreliable are suggestive statements
by several confidential witnesses that some employees were
upset about their time being reassigned, that corporate officers
“munged” the financial data through separate spreadsheets at
the end of each month, and that Banerjea was “accessing
   5
     It is also notable that despite the many allegations in the SAC relating
to Digimarc’s fraudulent use of inventory reserves the corporation did not
restate inventory reserves when it formally published its restatement on
April 5, 2005. Since we find that Zucco has failed to properly allege
scienter, however, we need not also analyze whether the falsity of Zucco’s
inventory reserve representations is pled with the requisite particularity.
See 

(“A securities fraud complaint must now
‘specify each statement alleged to have been misleading, the reason or rea-
sons why the statement is misleading, and, if an allegation regarding the
statement or omission is made on information and belief, the complaint
shall state with particularity all facts on which that belief is formed.’ ”
(quoting 

)).
              ZUCCO PARTNERS v. DIGIMARC CORP.             1493
Digimarc’s inventory accounts on his own and manipulating
the value of the Company’s inventory.” None of these confi-
dential witness statements establishes the witnesses’ personal
knowledge or reliability by recounting the particulars of the
alleged transgressions.

   Finally, several confidential witnesses report only conclu-
sory assertions about the defendants’ scienter. We have previ-
ously cautioned that such assertions are usually insufficient,
standing alone, to adequately allege scienter. See In re Worlds
of Wonder Securities Litigation, 

, 1426-27 (9th
Cir. 1994). For example, CW6 claims that Ranjit “had to have
known what was going on with respect to the Company’s
inventory accounting manipulation,” and CW3 contends that
certain project managers knew that the controls on the time-
entry system were altered “merely so that this time could be
capitalized and not recognized as expenses, thereby improp-
erly and falsely boosting the Company’s reported income.”
These generalized claims about corporate knowledge are not
sufficient to create a strong inference of scienter, since they
fail to establish that the witness reporting them has reliable
personal knowledge of the defendants’ mental state.

   The few allegations that have the requisite level of particu-
larity to withstand the first prong of the Daou confidential
witness test fail to demonstrate the deliberate recklessness
required to survive the second prong. Instead, these remaining
allegations demonstrate only that there was some disagree-
ment within the corporation over its accounting processes,
and not that Digimarc’s management was deliberately reck-
less in its capitalization of certain software development costs
in violation of GAAP. See Silicon 

.

   CW2 recounts that he/she was personally instructed by
Ford (and Ranjit) to “re-program the time and expense soft-
ware to eliminate the required supervisor or manager approval
of time” and to allow “administrative assistants working in
the finance department [to] enter time for the [software] engi-
1494          ZUCCO PARTNERS v. DIGIMARC CORP.
neers, rather than having the engineers enter their time direct-
ly.” But, as the district court concluded, “it is questionable
whether it is inherently improper to allow finance personnel
access to the payroll system in order to categorize engineers’
time spent on software development as ordinary expenses or
capital expenses.” Zucco 

.
There is nothing so necessarily nefarious about a time-entry
system supervised by the finance department to suggest that
an inference of deliberate recklessness in such a situation is
equally as cogent and as compelling as an innocent explana-
tion. Indeed, this allegation demonstrates only “motive and
opportunity,” which, without more, is not enough to establish
a cogent and compelling inference of scienter. See DSAM
Global Value Fund v. Altris Software, Inc., 

, 389
(9th Cir. 2002) (quoting Silicon 

(“To allege a strong inference of deliberate recklessness,
Appellants must state facts that come closer to demonstrating
intent, as opposed to mere motive and opportunity.”) (internal
quotation marks omitted)).

   CW1, who was the Director of the IT Department, reports
from personal knowledge that Ranjit, at the close of every
accounting period, required the IT Department to copy all the
data from the corporation’s Great Plains accounting system
into spreadsheets, so that management could analyze the data.
But the existence of the separate databases, which were alleg-
edly used in preparing Digimarc’s financials, indicates only
motive and opportunity—not scienter. In fact, the SAC sup-
plies facts indicating that management sincerely believed the
numbers generated by the Great Plains system were incorrect
and wanted to confirm these numbers through the use of sepa-
rate databases. The SAC describes several meetings wherein
Digimarc’s management openly questioned the formulae used
by the Great Plains system to calculate inventory levels, and
at least one telephone call between management and a Micro-
soft representative to resolve the perceived issues. At best,
CW1’s account establishes “disagreement and questioning
within [Digimarc] about the [accounting numbers],” and fails
              ZUCCO PARTNERS v. DIGIMARC CORP.             1495
to demonstrate that “[Digimarc’s] external auditors counseled
against the practice or that [management] admitted or was
aware that [using the separate databases] was improper.” Met-
zler 

. This is far from the deliber-
ate, conscious recklessness required for a strong inference of
scienter under the PSLRA.

   Likewise, the SAC’s many specific allegations about
improper inventory scrap rates and obsolete inventory only
demonstrate disagreement within Digimarc over the proper
percentage of raw materials that should have been recorded
on its balance sheets. CW1 alleges that his scrap rate of sixty
percent was “correctly calculated,” but that management at
Digimarc refused to implement such a high (or individual-
ized) scrap rate, and instead applied a uniform scrap rate of
between five and thirty percent. CW1 specifically recollects
sending an email to management about his preferred scrap
rates. Mere knowledge of alternative scrap rates, however, or
disagreement among employees with regard to the proper
scrap rate, is not enough to establish a cogent or compelling
scienter allegation—especially where, as here, there is no
indication that Digimarc’s management acted with deliberate
recklessness in choosing the lower uniform scrap rates.
Indeed, nothing in the SAC suggests using the chosen scrap
rate violated GAAP. See Metzler 

. Likewise, CW1’s statements about specific obsolete
inventory held within the corporation do not present a cogent
and compelling inference of scienter. Obsolete inventory may
be retained for a variety of reasons, including to satisfy the
repair demands of prior customers.

   The lone fact in the SAC reported by a confidential witness
that is in any way indicative of scienter is too contradictory
to be compelling. CW1 specifically reports that he/she was
directly ordered by Ranjit in 2004 “not to write down obsolete
inventory because, according to Ranjit . . . writing down obso-
lete inventory would result in the Company’s missing market
expectations.” Although this statement, considered in isola-
1496          ZUCCO PARTNERS v. DIGIMARC CORP.
tion, might be enough to demonstrate scienter, it is notable
that Digimarc did write down significant amounts of obsolete
inventory in 2004. CW1’s statement, therefore, is simply
incongruous with Digimarc’s public actions alleged in the
complaint. As we have recently noted, “a plaintiff cannot
avoid dismissal by reliance on an isolated statement that
stands in contrast to a host of other insufficient allegations.”
Metzler 

. Especially as here,
where a single statement indicative of scienter is contradicted
by readily available physical evidence, it is impossible to con-
clude that the statement creates an inference of scienter suffi-
ciently cogent or compelling to survive under Tellabs.

   [8] As a whole, the SAC’s plethora of confidential witness
statements fail to create a an inference of scienter more cogent
or compelling than an alternative innocent inference. 

. The complaint’s only specific allegation
that could be used to infer scienter (that Ranjit ordered CW1
to write-down inventory) is contradicted by other physical
evidence. The remaining confidential witness statements are
either not indicative of scienter or so vague and of such unre-
liable origin as to be unpersuasive.

                               2

   In addition to the number of confidential witnesses Zucco
relies upon in its Second Amended Complaint, it also con-
tends that certain of Digimarc’s public actions support an
inference of scienter—first among which is the issuance of
the restatement of earnings on April 5, 2005.

   In general, the mere publication of a restatement is not
enough to create a strong inference of scienter. In particular,
we have previously found inadequate complaints alleging that
“facts critical to a business’s core operations or an important
transaction generally are so apparent that their knowledge
may be attributed to the company and its key officers.” In re
Read-Rite Corp. Sec. Litig., 

, 848 (9th Cir. 2003)
              ZUCCO PARTNERS v. DIGIMARC CORP.             1497
(quotation and emphasis omitted). See also DSAM Global
Value 

(“Thus, mere allegations that an
accountant negligently failed to closely review files or follow
GAAP cannot raise a strong inference of scienter.”).

   Recently, however, we have recognized two exceptions to
this general rule, and have found bare allegations of falsely
reported information probative under certain narrow condi-
tions. See South 

(summarizing the
exceptions). Specifically, falsity may itself be indicative of
scienter where it is combined with “allegations regarding a
management’s role in the company” that are “particular and
suggest that the defendant had actual access to the disputed
information,” and where “the nature of the relevant fact is of
such prominence that it would be ‘absurd’ to suggest that
management was without knowledge of the matter.” 

(quotation marks and citation omitted).

   [9] The first exception permits general allegations about
“management’s role in a corporate structure and the impor-
tance of the corporate information about which management
made false or misleading statements” to create a strong infer-
ence of scienter when these allegations are buttressed with
“detailed and specific allegations about management’s expo-
sure to factual information within the company.” 

To satisfy this standard, plaintiffs might include in their com-
plaint “specific admissions from top executives that they are
involved in every detail of the company and that they moni-
tored portions of the company’s database,” 

411 F.3d at 1022-23), a specific admission from a top execu-
tive that “ ‘[w]e know exactly how much we have sold in the
last hour around the world,’ ” 

), or other particular “details about the defen-
dants’ access to information within the company.” 

SAC fails to allege such particular details.
Although the SAC includes allegations that senior manage-
ment (and Ranjit, in particular) closely reviewed the account-
1498          ZUCCO PARTNERS v. DIGIMARC CORP.
ing numbers generated by Digimarc each quarter (through the
use of the Access databases), and that top executives had sev-
eral meetings in which they discussed quarterly inventory
numbers, this is not enough to satisfy the narrow exception to
Read-Rite. Allegations that Digimarc’s management had
access to the purportedly manipulated quarterly accounting
numbers, or that the management analyzed the inventory
numbers closely, do not support the inference that manage-
ment was in a position to know that such data was being
manipulated. Nothing in the complaint suggests that Ranjit
had access to the underlying information from which the
accounting numbers were derived.

   [11] The second exception to Read-Rite permits an infer-
ence of scienter where the information misrepresented is read-
ily apparent to the defendant corporation’s senior
management. Where the defendants “must have known” about
the falsity of the information they were providing to the pub-
lic because the falsity of the information was obvious from
the operations of the company, the defendants’ awareness of
the information’s falsity can be assumed. See Berson v.
Applied Signal Tech., Inc., 

, 987-89 (9th Cir.
2008). Nevertheless, reporting false information will only be
indicative of scienter where the falsity is patently obvious—
where the “facts [are] prominent enough that it would be
‘absurd to suggest’ that top management was unaware of
them.” 

(quoting America 

n.21). In Berson we found that the defendant company’s mis-
representation of the status of stop-work orders was enough
to infer scienter when the four stop-work orders had respec-
tively “halted between $10 and $15 million of work on the
company’s largest contract with one of its most important
customers,” “halted $8 million of work,” “caused the com-
pany to reassign 50-75 employees,” and “required [Defen-
dant] to complete massive volumes of paperwork.” See 

n.5 (quotation marks omitted).

   [12] In this case, unlike in Berson, the alleged misrepresen-
tations do not concern especially prominent facts. In particu-
               ZUCCO PARTNERS v. DIGIMARC CORP.              1499
lar, Digimarc admitted in its restatement that “certain costs
had been erroneously capitalized because either a portion of
such costs did not qualify for capitalization (e.g., costs related
to the preliminary or post-implementation stages [of a soft-
ware project]) or the project itself did not qualify as internal-
use software (e.g., costs related to a non-software project).”
Because these misrepresentations are largely definitional, the
falsity of the original representations would not be immedi-
ately obvious to corporate management. For example, man-
agement would have to be aware that a software development
project had reached the “post-implementation” stage or that
computer engineers were working on software that did not
qualify for “internal use.” There is no indication that the dif-
ferences between the “preliminary project” stage, the “appli-
cation development” stage, and the “post-implementation”
stage of a software project would be operationally visible to
executives not intimately connected with the development
process and GAAP’s definitions. Although the difference
between an “internal software project” and a more general
research and development software project might be slightly
more apparent to the lay observer, some research and devel-
opment projects might be closely integrated with software
development. In any case, there is no indication that the
reclassification of a development project along these lines
would, as in Berson, “cause [Digimarc] to reassign 50-75
employees” or “complete a massive volume of paperwork.”

are not the type that qualifies for
the narrow exception to the general rule that falsity alone can-
not create a strong inference of scienter.

                                3

   Zucco also claims that the resignation of several members
of Digimarc’s financial department can be used to infer
scienter. The SAC points to the resignation of (1) Digimarc’s
CFO Ranjit, in early 2004, (2) two Digimarc controllers
(Scacchi, after only six weeks of work, and Diana King, in
late 2003); and (3) KPMG as Digimarc’s independent
1500           ZUCCO PARTNERS v. DIGIMARC CORP.
accounting firm on June 14, 2005, shortly after the restate-
ment was issued as evidence that Digimarc was conscious of
the accounting misrepresentations.

   [13] Although resignations, terminations, and other allega-
tions of corporate reshuffling may in some circumstances be
indicative of scienter, the resignations at issue here are not so
numerous or suspicious as to raise such an inference. Where
a resignation occurs slightly before or after the defendant cor-
poration issues a restatement, a plaintiff must plead facts
refuting the reasonable assumption that the resignation
occurred as a result of restatement’s issuance itself in order
for a resignation to be strongly indicative of scienter. See,
e.g., In re U.S. Aggregates, Inc. Sec. Litig., 

, 1074 (N.D. Cal. 2002) (“Plaintiff can point to no partic-
ularized allegation refuting the reasonable assumption that
[defendant’s employee] was fired simply because the errors
that lead to the restatement occurred on his watch or because
he failed adequately to supervise his department.”). Here, the
resignation of KPMG as Digimarc’s independent accounting
firm a month after the restatement was issued is not surprising
—it had just been partially responsible for the corporation’s
failure to adequately control its accounting procedures. This
is not enough to support a strong inference of scienter.

   For other resignations occurring during the relevant time
period, a plaintiff must allege sufficient information to differ-
entiate between a suspicious change in personnel and a benign
one. Mere conclusory allegations that a financial manager
resigns or retires during the class period or shortly before the
corporation issues its restatement, without more, cannot sup-
port a strong inference of scienter. Cf. Silicon 

(holding that stock sales of individual defendants
are only indicative of scienter where they are “dramatically
out of line with prior trading practices” (quoting In re Apple
Computer Sec. Litig., 

, 1117 (9th Cir. 1989))).
Absent allegations that the resignation at issue was uncharac-
teristic when compared to the defendant’s typical hiring and
               ZUCCO PARTNERS v. DIGIMARC CORP.               1501
termination patterns or was accompanied by suspicious cir-
cumstances, the inference that the defendant corporation
forced certain employees to resign because of its knowledge
of the employee’s role in the fraudulent representations will
never be as cogent or as compelling as the inference that the
employees resigned or were terminated for unrelated personal
or business reasons.

   In this case, the SAC alleges that Digimarc’s CFO Ranjit
retired “just prior to the disclosure of Digimarc’s improper
accounting and lack of financial controls during his tenure.”
The complaint does not indicate whether Ranjit was nearing
retirement age, whether he left to pursue other opportunities,
or even the length of his tenure. Thus the bare fact of Ranjit’s
retirement cannot support Zucco’s allegations of scienter.
Similarly, the SAC’s allegations that Scacchi and King (two
Digimarc controllers) resigned during the class period are not
enough, absent particular facts about Digimarc’s hiring and
firing of controllers during the class period, to create a com-
pelling inference of scienter. Although Zucco contends that
both left because they believed management was unethical,
these accounts are based on vague hearsay allegations and are
not specific enough to extract a strong inference of scienter
from otherwise mundane turnover in the corporation’s finan-
cial department.

                                4

    The SAC also alleges that boilerplate Sarbanes-Oxley certi-
fications signed by the individual defendants, Davis and Ran-
jit, are strongly indicative of scienter. Specifically, pursuant to
section 302(a) of the Sarbanes-Oxley Act, a company’s “prin-
cipal executive officer or officers and the principal financial
officer or officers” must certify the accuracy and reliability of
its quarterly financial reports. See 15 U.S.C. § 7241(a). A
signing officer must certify that he has reviewed the report,
that based on his knowledge the report “does not contain any
untrue statement of a material fact or omit to state a material
1502             ZUCCO PARTNERS v. DIGIMARC CORP.
fact necessary in order to make the statements made . . . not
misleading,” and that the report and any information included
within the report “fairly present in all material respects the
financial condition and results of operations of the issuer
. . . .” 15 U.S.C. § 7241(a)(1), (2), (3). Moreover, the officer
must certify that he is “responsible for establishing and main-
taining internal controls,” 15 U.S.C. § 7241(a)(4)(A), and that
he has “evaluated the effectiveness of the issuer’s internal
controls” within the past ninety days and has “presented in the
report [his] conclusions about the effectiveness of [the corpo-
ration’s] internal controls based on [his] evaluation as of that
date.” 15 U.S.C. § 7241 (a)(4)(B)-(D). Zucco alleges that
compliance with section 302(a)—and in particular, specific
language in Digimarc’s “Controls and Procedures” section of
its Form 10-Q for the second quarter of 2003 announcing its
compliance with that section—is indicative of scienter.6

   The SAC alleges that the following language in the “con-
trols and procedures” section of Digimarc’s Form 10-Q for
the second quarter of 2003, certified by Davis and Ranjit, can
raise an inference of scienter:

     As of June 30, 2003 our management evaluated,
     under the supervision and with the participation of
     our Chief Executive Officer and Chief Financial
  6
    Zucco also erroneously alleges that the omission of certain language in
Digimarc’s March 31, 2004 10-Q, which was present in its March 31,
2003 10-Q, is indicative of scienter. The language in question read:
    The registrant’s other certifying officer and I have indicated in
    this quarterly report whether or not there were significant
    changes in internal controls or in other factors that could signifi-
    cantly affect internal controls subsequent to the date of our most
    recent evaluation . . .
As the district court correctly pointed out, however, this language was
omitted from subsequent certifications by governmental mandate (through
amendments to the SEC rules which became effective August 14, 2003).
Thus, the changes cannot be indicative of scienter. See Zucco 

.
              ZUCCO PARTNERS v. DIGIMARC CORP.             1503
    Officer, the effectiveness of the design and operation
    of our disclosure controls and procedures. Based on
    this evaluation, our Chief Executive Officer and
    Chief Financial Officer concluded that our disclosure
    controls and procedures are effective in timely alert-
    ing them to material information required to be
    included in this report. Our management also evalu-
    ated, under the supervision and with the participation
    of our Chief Executive Officer and Chief Financial
    Officer, any charge that occurred in our internal con-
    trol over financial reporting during the fiscal quarter
    ended June 30, 2003. No such change materially
    affected, or is reasonably likely to materially affect,
    our internal control over financial reporting.

Although this language is mandated by the Securities and
Exchange Commission, Zucco argues that at least one district
court, see In re Lattice Semiconductor Corp. Sec. Litig., No.
CV04-1255-AA, 

, at *18 (D. Or. Jan. 3,
2006), has held that such language gives rise to a strong infer-
ence of scienter (once the language’s falsity is shown).

   [14] Boilerplate language in a corporation’s 10-K form, or
required certifications under Sarbanes-Oxley section 302(a),
however, add nothing substantial to the scienter calculus. Our
sister circuits to rule on such questions have unanimously
agreed that allowing Sarbanes-Oxley certifications to create
an inference of scienter in “every case where there was an
accounting error or auditing mistake made by a publicly
traded company” would “eviscerat[e] the pleading require-
ments for scienter set forth in the PSLRA.” Garfield v. NDC
Health Corp., 

, 1266 (11th Cir. 2006); accord
In re Ceridian Corp. Sec. Litig., 

, 248 (8th Cir.
2008); Ind. Elec. Workers’ Pension Trust Fund IBEW v. Shaw
Group, Inc., 

, 545 (5th Cir. 2008); Central
Laborers’ Pension Fund v. Integrated Elec. Servs. Inc., 

, 555 (5th Cir. 2007). We have recently joined these
circuits and have ruled that “Sarbanes-Oxley certifications are
1504           ZUCCO PARTNERS v. DIGIMARC CORP.
not sufficient, without more, to raise a strong inference of
scienter.” Glazer Capital Mgmt., LP v. Magistri, ___ F.3d
___, No. 06-16899, 

at *9 (9th Cir. Nov.
26, 2008). Thus, we reject Zucco’s invitation to undermine
the PSLRA’s distinct requirements for pleading falsity and
scienter, and hold that Zucco’s Sarbanes-Oxley certifications
are not enough to create a strong inference of scienter and do
not make Zucco’s otherwise insufficient allegations more
compelling by their presence in the same complaint.

                                5

   The SAC additionally alleges that Davis’ and Ranjit’s exec-
utive compensation packages, which included significant
bonuses tied to Digimarc’s financial performance, are indica-
tive of scienter because they demonstrate that the individual
defendants had a pecuniary motive to inflate the corporation’s
financial performance. In particular, the SAC notes that
Davis’ compensation was being evaluated during the class
period, that his bonus was “based in part on Digimarc’s ‘oper-
ating profit,’ ” and that based on the financial results reported
in 2003, Davis received a bonus of $113,000. Zucco’s com-
plaint also alleges that Davis and Ranjit were rewarded for
Digimarc’s 2003 financial performance with substantial stock
option grants: specifically, on January 2, 2004, Digimarc
granted Davis 110,000 stock options and Ranjit 25,000 stock
options (which effectively doubled the amount of shares Ran-
jit beneficially owned).

   A strong correlation between financial results and stock
options or cash bonuses for individual defendants may occa-
sionally be compelling enough to support an inference of
scienter. See America 

. In America
West, we noted that because “none of the [defendant’s] execu-
tive officers received options awards in 1997 . . . [but defen-
dant] awarded [thousands of options to executive officers] in
March 1998 [for performance allegedly increased by misrep-
resentations] . . . a strong inference of scienter can be inferred
              ZUCCO PARTNERS v. DIGIMARC CORP.             1505
from Plaintiffs’ allegations.” Id.; see also 

(“[P]ersonal financial gain may weigh heavily in
favor of a scienter inference.”).

   Zucco, however, has failed to provide the particularity we
found persuasive in America West. Although Zucco’s SAC
alleges that Davis and Ranjit received bonuses and stock
option grants in part based on Digimarc’s financial perfor-
mance, there is no allegation indicating how intimately the
bonuses were tied to the company’s financials. In America
West, we found it significant that the individual defendants’
compensation was based “principally” on the defendant com-
pany’s financial 

. The complaint
at issue in America West established this fact by comparison
of the individual defendants’ prior year’s compensation with
the year in question, noting that while “none of the executive
officers received option awards in 1997 for the previous
year,” in the year in question “America West awarded Franke
350,000 options . . . [and] awarded 110,000 options to Good-
manson, 35,000 options to Parker, and 20,000 options to
Garel in March 1998.” 

Zucco’s SAC makes only the bare assertion that
executive-level bonuses were “based in part” on Digimarc’s
financial performance—the complaint fails to provide specifi-
cally, with comparisons to prior years bonuses, the correlation
between Davis’ and Ranjit’s compensation and Digimarc’s
bottom line. Such “generalized assertions of motive, without
more, are inadequate to meet the heightened pleading require-
ments of Silicon Graphics” and Tellabs. Lipton v. Pathogene-
sis Corp., 

, 1038 (9th Cir. 2002). If simple
allegations of pecuniary motive were enough to establish
scienter, “virtually every company in the United States that
experiences a downturn in stock price could be forced to
defend securities fraud actions.” 

v. IMC-
ERA Group, Inc., 

, 54 (2d Cir. 1995).
1506           ZUCCO PARTNERS v. DIGIMARC CORP.
                               6

   Next, Zucco alleges that the individual defendants’ stock
sales during the class period are strongly indicative of
scienter. The SAC notes several sales during the class period,
alleging that Davis sold approximately 4.5% and Ranjit sold
approximately 48% of their “total personal Digimarc stock
holdings, including options,” during the period between May
23, 2003 and May 4, 2004. During this period, according to
the SAC, Davis sold 38,750 shares of stock for a total of
$610,375.00, and Ranjit sold 30,000 shares of stock for a total
of $392,750.00. The SAC also alleges that Davis sold a sub-
stantially greater percentage of his stock options that were “in
the money” during the class period (those exercisable options
that had a stock price above the market price of Digimarc’s
stock).

   As we have previously articulated, “[a]lthough ‘unusual’ or
‘suspicious’ stock sales by corporate insiders may constitute
circumstantial evidence of scienter, insider trading is suspi-
cious only when it is ‘dramatically out of line with prior trad-
ing practices at times calculated to maximize the personal
benefit from undisclosed inside information.’ ” Silicon

(citing Apple 

) (internal citation and quotation marks omitted).
Among three factors that must be considered to determine
whether stock sales raise a strong inference of deliberate reck-
lessness are: “(1) the amount and percentage of shares sold by
insiders; (2) the timing of the sales; and (3) whether the sales
were consistent with the insider’s prior trading history.” 

   [16] As the district court correctly noted, the SAC “fail[s]
to provide any information on the trading history of Davis or
Ranjit for purposes of comparison to the stock sales at issue.”
Zucco 

. For individual
defendants’ stock sales to raise an inference of scienter, plain-
tiffs must provide a “meaningful trading history” for purposes
              ZUCCO PARTNERS v. DIGIMARC CORP.             1507
of comparison to the stock sales within the class period. See
In re Vantive Corp. Sec. Litig., 

, 1095-96 (9th
Cir. 2002). Even if the defendant’s trading history is simply
not available, for reasons beyond a plaintiff’s control, the
plaintiff is not excused from pleading the relevant history. See

(noting that “[b]ecause [the defendant] joined Van-
tive four months into the class period, he has no relevant trad-
ing history,” and thus finding that “[b]ecause [the defendant]
had no trading history, we cannot conclude that his trades
were out of line with his past practice”); 

; Silicon 

(rejecting an
inference of scienter when a defendant sold over 75.3 percent
of stock holdings during the class period, because the defen-
dant was “legally forbidden to trade” for a long period before
the class period and thus had no meaningful trading history).
There is no indication that Tellabs has altered our pleading
standard based on suspicious stock sales. See Metzler Invest-

(reaffirming the tripartite Silicon
Graphics test after Tellabs). Thus, since there is no allegation
within the SAC that Davis and Ranjit’s stock sales, though
significant, are inconsistent with their usual trading patterns,
no inference of scienter can be gleaned from Zucco’s stock
sale assertions.

                               7

   [17] Similarly, the SAC’s allegations about Digimarc’s pri-
vate placement of stock during the class period do not contain
enough relevant comparative history to create a strong infer-
ence of scienter. Zucco alleges that on August 25, 2003, Digi-
marc “completed a private placement offering whereby it sold
1,785,996 units to institutional and accredited investors, rais-
ing net proceeds of $23.5 million for Digimarc.” Each unit
included one share of stock and a warrant to purchase 0.15
shares of common stock at an exercise price of $14 per share.

  Although “corporate acquisitions” may, when combined
with “specific allegations of deliberate accounting misfea-
1508           ZUCCO PARTNERS v. DIGIMARC CORP.
sance,” create a strong inference of scienter, see 

, mere generalized assertions about “routine busi-
ness objectives, without more” cannot support such an infer-
ence. See 

. To create a strong
inference of scienter, therefore, the corporate stock sales must
be significant enough and uncharacteristic enough to cast
doubt on the defendant company’s motives. In Daou, for
example, we held that a company’s eleven stock-funded
acquisitions during the class period were at least partially
indicative of scienter where the company “exchang[ed] over
6.6 million shares” of its stock and, if the stock had been
properly valued, would have been required to exchange
19,642,865 more shares of its stock to accomplish the same

. The stock placement in this
case is far less disproportionate than in Daou. Moreover,
Zucco has failed to allege that the Digimarc’s stock placement
was in any way inconsistent with the corporation’s traditional
business practices. See Silicon 

.
Thus, the SAC’s allegations about Digimarc’s private place-
ment do not create a strong inference of scienter.

                                C

   Although none of the SAC’s allegations of scienter is indi-
vidually cogent or compelling enough to survive under the
PSLRA, we must also “consider the complaint in its entirety”
to determine whether “all of the facts alleged, taken collec-
tively, give rise to a strong inference of scienter.” 

. As we have recently articulated, “Tellabs per-
mits a series of less precise allegations to be read together to
meet the PSLRA requirement.” South 

.
Accordingly, even “[v]ague or ambiguous allegations are now
properly considered as a part of a holistic review when con-
sidering whether the complaint raises a strong inference of
scienter.” 

this holistic review, however,
we must also “take into account plausible opposing infer-
ences” that could weigh against a finding of scienter. 

. Even if a set of allegations may create an
               ZUCCO PARTNERS v. DIGIMARC CORP.             1509
inference of scienter greater than the sum of its parts, it must
still be at least as compelling as an alternative innocent expla-
nation.

   [18] Although the allegations in this case are legion, even
together they are not as cogent or compelling as a plausible
alternative inference—namely, that although Digimarc was
experiencing problems controlling and updating its account-
ing and inventory tracking practices, there was no specific
intent to fabricate the accounting misstatements at issue here.
Instead, the facts alleged by Zucco point towards the conclu-
sion that Digimarc was simply overwhelmed with integrating
a large new division into its existing business. The SAC notes
that Digimarc in 2001 significantly expanded the scope of its
business by acquiring its ID Systems unit from Polaroid. This
acquisition eventually mandated the integration of several
accounting systems into the new Great Plains system. As the
SAC reports, Digimarc’s 2004 Form 10-K admitted that this
integration resulted in problems including inadequate training
of personnel in the new Great Plains system, duplicate record-
ing of purchases, and other classification errors. It is more
plausible that Digimarc’s management was unable to control
the accounting processes within the corporation during this
integration than that it was systematically using accounting
manipulations to make the company seem slightly more finan-
cially successful. As a result, we hold that the district court
did not err when it dismissed Zucco’s Second Amended Com-
plaint for failure to sufficiently allege scienter under Federal
Rule of Civil Procedure 9(b) and the PSLRA.

                               III

   Because the district court dismissed the complaint with
prejudice, we must finally consider whether the district court
abused its discretion in refusing to grant Zucco leave to
amend its complaint. See 

. Under
Federal Rule of Civil Procedure 15(a)(2), federal courts are
instructed to “freely give leave [to amend] when justice so
1510           ZUCCO PARTNERS v. DIGIMARC CORP.
requires.” A district court, however, may in its discretion deny
leave to amend “due to ‘undue delay, bad faith or dilatory
motive on the part of the movant, repeated failure to cure defi-
ciencies by amendments previously allowed, undue prejudice
to the opposing party by virtue of allowance of the amend-
ment, [and] futility of amendment.’ ” Leadsinger, Inc. v. BMG
Music Publ’g, 

, 532 (9th Cir. 2008) (quoting
Foman v. Davis, 

, 182 (1962)). Since the district
court determined that any attempt to amend would be futile,
“we will affirm the district court’s dismissal on this basis if
it is clear, upon de novo review, that the complaint could not
be saved by any amendment.” 

and cita-
tion omitted). As here, where the plaintiff has previously been
granted leave to amend and has subsequently failed to add the
requisite particularity to its claims, “[t]he district court’s dis-
cretion to deny leave to amend is particularly broad.” Read-

(quoting 

).

   [19] The district court, when dismissing the First Amended
Complaint, held that Zucco had failed to satisfy the scienter
requirements of the PSLRA with respect to its allegations
based on confidential witness statements and stock sales. The
fact that Zucco failed to correct these deficiencies in its Sec-
ond Amended Complaint is “a strong indication that the plain-
tiffs have no additional facts to plead.” 

. Accordingly, the district court did not err when it dis-
missed the SAC with prejudice, since it was clear that the
plaintiffs had made their best case and had been found want-
ing. See Metzler 

(upholding a
dismissal with prejudice where, inter alia, the deficiencies at
issue “persisted in every prior iteration of the [complaint]”).

                                IV

   The allegations of scienter in the SAC, though voluminous,
are not pled with the particularity required to survive a Fed-
eral Rule of Civil Procedure 12(b)(6) dismissal under the
standards enumerated in Federal Rule of Civil Procedure 9(b)
              ZUCCO PARTNERS v. DIGIMARC CORP.            1511
and the PSLRA. Instead, the plaintiffs in this case assume that
compiling a large quantity of otherwise questionable allega-
tions will create a strong inference of scienter through the
complaint’s emergent properties. Although Tellabs instructs
us to view such compilations holistically, even such a com-
prehensive perspective of Zucco’s complaint cannot transform
a series of inadequate allegations into a viable inference of
scienter. We therefore affirm the district court’s dismissal of
the Second Amended Complaint with prejudice.

  AFFIRMED.