EASTERBROOK, Chief Judge.

On July 22, 2004, Baxter International announced that it would restate the preceding three years' earnings in order to correct errors created by fraud at its subsidiary in Brazil. Managers there had conveyed the illusion of growth by reporting sales as having been made earlier than their actual dates; when it was no longer possible to accelerate revenue this way, the managers reported fictitious sales. Brazilian managers also failed to create appropriate reserves for bad debts. On the day the problem was announced, Baxter's common stock fell $1.48 per share, or 4.6% of its market price. (The parties have not used econometric methods to separate firm-specific changes from movements of the market as a whole, so we give raw numbers.) A few weeks later, when the formal restatement showed that the overstatement of profits was less serious than many investors initially feared, the price edged up. This is consistent with the pattern at other firms: a plan to restate earnings creates uncertainty that may be dispelled by the concrete results, but revelation that the firm's internal controls allowed the problem in the first place usually causes a persistent loss. See Zoe-Vonna Palmrose, Vernon J. Richardson Susan Scholz, Determinants of market reactions to restatement announcements, 37 J. Accounting Econ. 59 (2004).

Any restatement of a public company's financial results is likely to be followed by litigation. Multiple suits were filed in the wake of this one. These claims, which invoke § 10(b) of the Securities Exchange Act of 1934, 15 U.S.C. § 78j(b), and the SEC's Rule 10b-5, 17 C.F.R. § 240.10b-5, were consolidated, and a lead plaintiff was selected under the Private Securities Litigation Reform Act of 1995. Oddly, the district court never decided whether the litigation could proceed as a class action, though this subject has not been raised as an issue on appeal. (Although the PSLRA applies only to a "suit that is brought as a plaintiff class action", 15 U.S.C. § 78u-4(a)(1), the statute's rules apply whether or not the class is certified.)

The district court issued a series of opinions first dismissing the action, 2005 WL 1272271, 2005 U.S. Dist. LEXIS 12006 (N.D.Ill. May 25, 2005), then reinstating it, 2005 WL 2368795, 2005 U.S. Dist. LEXIS 21349 (N.D.Ill. Sept. 23, 2005), and finally dismissing it again with prejudice, 2005 WL 3542521, 2005 U.S. Dist. LEXIS 38011 (N.D.Ill. Dec. 22, 2005). The PSLRA provides that the complaint in a securities-fraud action must, "with respect to each act or omission alleged to violate this chapter, 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). That "required state of mind" is an intent to deceive, demonstrated by knowledge of the statement's falsity or reckless disregard of a substantial risk that the statement is false. See Ernst Ernst v. Hochfelder, 425 U.S. 185, 193, 96 S.Ct. 1375, 47 L.Ed.2d 668 (1976); SEC v. Jakubowski, 150 F.3d 675, 681 (7th Cir.1998). The district court ultimately ruled that the complaint fails to establish the required "strong inference" of scienter.

At the time this appeal was briefed and argued, Makor Issues Rights, Ltd. v. Tellabs, Inc., 437 F.3d 588 (7th Cir.2006), supplied this circuit's understanding of § 78u-4(b)(2). Shortly after argument, however, the Supreme Court granted certiorari in Tellabs, and we deferred action pending the Court's decision. The Supreme Court's opinion, Tellabs, Inc. v. Makor Issues Rights, Ltd., ___ U.S. ___, 127 S.Ct. 2499, 168 L.Ed.2d 179 (2007), establishes two propositions that govern this appeal. First, "[a] complaint will survive [a motion to dismiss] 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 alleged." 127 S.Ct. at 2510, footnote omitted. Second, in applying this standard, "the court must take into account plausible opposing inferences." 127 S.Ct. at 2502.

One upshot of the approach that Tellabs announced is that we must discount allegations that the complaint attributes to five "confidential witnesses" — one ex-employee of the Brazilian subsidiary, two ex-employees of Baxter's headquarters, and two consultants. It is hard to see how information from anonymous sources could be deemed "compelling" or how we could take account of plausible opposing inferences. Perhaps these confidential sources have axes to grind. Perhaps they are lying. Perhaps they don't even exist.

At oral argument, we asked when the identity of these five persons would be revealed and how their stories could be tested. The answer we received was that the sources' identity would never be revealed, which means that their stories can't be checked. Yet Tellabs requires judges to weigh the strength of plaintiffs' favored inference in comparison to other possible inferences; anonymity frustrates that process.

Not that anonymity is possible in the long run. There is no "informer's privilege" in civil litigation. Defendants are entitled to learn in discovery who has relevant evidence, and to obtain that evidence. Indeed, plaintiffs are obliged by Fed.R.Civ.P. 26(a)(1)(A) to provide defendants with the names and addresses of all persons "likely to have discoverable information that the disclosing party may use to support its claims or defenses". Concealing names at the complaint stage thus does not protect informers from disclosure (and the risk of retaliation); it does nothing but obstruct the judiciary's ability to implement the PSLRA.

This does not mean that plaintiffs must reveal all of their sources, as one circuit has required. See In re Silicon Graphics Inc. Securities Litigation, 183 F.3d 970, 985 (9th Cir.1999). A complaint is not a discovery device. Our point, rather, is that anonymity conceals information that is essential to the sort of comparative evaluation required by Tellabs. To determine whether a "strong" inference of scienter has been established, the judiciary must evaluate what the complaint reveals and disregard what it conceals.

Decisions such as In re Cabletron Systems, Inc., 311 F.3d 11, 24 n. 6, 28-31 (1st Cir.2002), which countenance the use of confidential sources to satisfy the PSLRA, analogize to unnamed informants in affidavits for search warrants. See Illinois v. Gates, 462 U.S. 213, 103 S.Ct. 2317, 76 L.Ed.2d 527 (1983). That analogy shows the problem. A complaint passes muster under Tellabs "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 alleged." That is a higher standard than "probable cause," which (the court stressed in Gates) does not entail a more-likely-than-not threshold. No decision of which we are aware concludes that anonymous accusers can demonstrate that scienter is "at least as [likely] as any opposing inference one could draw from the facts alleged."

It is possible to imagine situations in which statements by anonymous sources may corroborate or disambiguate evidence from disclosed sources. Informants sometimes play this role in applications for search warrants. Because it is impossible to anticipate all combinations of information that may be presented in the future, and because Tellabs instructs courts to evaluate the allegations in their entirety, we said above that allegations from "confidential witnesses" must be "discounted" rather than ignored. Usually that discount will be steep. It is unnecessary to say more today.

Plaintiffs do not proffer concrete evidence that anyone at Baxter's headquarters in the United States knew of the shenanigans in Brazil until May 2004. Nor do plaintiffs contend that the knowledge of the senior managers in Brazil should be imputed either to Baxter or any of its managers in the United States. Perhaps they omit this line of argument because Baxter Brazil was a subsidiary rather than a division of Baxter International, but it does not matter why the argument has been omitted. It is enough to say that we need evaluate only arguments concerning managers at the firm's headquarters, none of whom participated in the scheme at the Brazilian subsidiary.

Plaintiffs describe, as the fraudulent statements, the income and earnings figures in Baxter's 2004 10-K report filed on March 12, 2004 (covering financial information through the end of 2003), and its 10-Q report filed on May 10, 2004 (covering financial information for the first quarter of 2004). These documents contain false information that the Brazilian subsidiary reported to Baxter's headquarters. Plaintiffs assert that by March 12 or May 10 Baxter's senior managers (the ones who signed the reports) knew the Brazilian data to be false, or were recklessly indifferent to its accuracy. Plaintiffs also maintain that Baxter's senior managers knew (and failed to disclose) that the Brazilian subsidiary had inadequate financial controls, and that even if Baxter did not learn of the Brazilian fraud until after May 10 it should not have waited until July 22 to disclose the problem.

Yet the complaint offers no reason at all to infer knowledge (or reckless indifference) before March 12, 2004, and very little reason to infer knowledge (or reckless indifference) before May 10, 2004. Although the complaint asserts that managers at Baxter's headquarters knew about the contracts reported by the Brazilian subsidiary — they appeared in a computerized database available throughout the firm — there is a big difference between knowing about the reports from Brazil and knowing that the reports are false. The complaint documents the former but not the latter.

Knowledge does not come until May at the earliest. One of the "confidential sources" supposedly told defendant Brian P. Anderson, Baxter's Chief Financial Officer from February 1998 through June 2004, about the Brazilian deceit during the first half of May 2004. (We refer to this allegation not because we think this "confidential source" reliable but to show that even by plaintiffs' lights the evidence is slim.) And during a discussion with analysts following the announcement on July 22, Robert Parkinson, Jr., Baxter's Chief Executive Officer, related that headquarters had been notified of the fraud by an employee in Brazil "sometime in the May time frame." These two tidbits do not supply a "compelling" demonstration that anyone who signed the 10-Q report filed on May 10 had actual knowledge that the data from Brazil had been tampered with. The most one can say is that, sometime during May 2004, Baxter learned enough to lead a reasonable person to conduct an investigation. That is exactly what Baxter did during the next two months, demonstrating a pursuit of truth rather than reckless indifference to the truth. Knowing enough to launch an investigation (Baxter could not simply assume that the initial report of bad news was accurate) is a very great distance from convincing proof of intent to deceive.

On April 29, 2004, Brazil's national government accused Baxter's subsidiary there of participating in a cartel to raise the price of blood products. Plaintiffs say that this should have alerted the parent to the fraud before the 10-Q report of May 10. We don't get it. Accusations differ from proof; business executives are not charged with "knowing" the truth of whatever any public official anywhere in the world may assert. Anyway, antitrust offenses have no apparent link to fraud. Cartels improve the profitability of the participants; proof that managers in Brazil were working hard to generate profits (if illegal ones) would not imply that they were also reporting nonexistent sales. What's more, the charge of antitrust conspiracy was public knowledge. If it were enough to demonstrate fraud, then plaintiffs and other investors could have drawn that inference themselves, and the price of Baxter's stock would have declined immediately. The securities laws do not require firms to "disclose" information that is already in the public domain. See Wielgos v. Commonwealth Edison Co., 892 F.2d 509, 517 (7th Cir.1989).

Plaintiffs insist that Baxter must have known of the deceit in April, because during that month Anderson and Carlos del Salto (one of Baxter's senior vice presidents) sold shares of the company's stock. Anderson realized about $1.5 million and del Salto $4.4 million. Surely they knew that something was amiss, plaintiffs insist. That inference is neither compelling nor cogent, given that plaintiff's own complaint identifies May 2004 as the first month in which the bad news reached Baxter's headquarters. Even if Anderson and del Salto had got wind of some problem, why would that induce them to sell? The restated financials showed that the fraud in Brazil had increased Baxter's reported net income by about $33 million over a three-year period, or some 1.5% of Baxter's operating profits. Reasonable executives need not see a 1.5% change as substantial; indeed, securities lawyers often use a 5% change as a rule-of-thumb approach to what is "material," and 1.5% is less than a third of that. Cf. Basic Inc. v. Levinson, 485 U.S. 224, 108 S.Ct. 978, 99 L.Ed.2d 194 (1988); TSC Industries, Inc. v. Northway, Inc., 426 U.S. 438, 96 S.Ct. 2126, 48 L.Ed.2d 757 (1976).

The complaint did not allege (nor do plaintiffs argue on appeal) that the sales by Baxter's senior managers as a whole were abnormally high during the period before public disclosure of the Brazilian problem. Managers sell stock all the time (Anderson's sale was part of a planned program, though del Salto's was not); if two of Baxter's senior managers sell blocs during the average month, plaintiffs can't get any mileage out of what happened in April. One possible inference is that the absence of sales by other managers who would have been in the know (had news of the fraud reached HQ) implies that nothing was thought to be out of the ordinary in April 2004. Because Tellabs instructs us to consider all potential inferences, and not just those that favor plaintiffs, the absence of any demonstration that April 2004 was an unusual period for managerial sales means that the complaint lacks the required "strong" demonstration of scienter.

Plaintiffs' remaining theories are even weaker. Take the claim that Baxter "knew" the financial controls in Brazil to be inadequate yet failed to disclose this to investors until July 22. Hindsight is the only basis of the proposed inference — and, as the Court observed in Tellabs, citing a famous opinion by Judge Friendly, there is no "fraud by hindsight." 127 S.Ct. at 2508, quoting from Denny v. Barber, 576 F.2d 465, 470 (2d Cir.1978). A report by Baxter's audit committee accompanying the restated financials in August 2004 concluded that the reporting system used in Brazil had not been up to the task of preventing the fraud. That's no news; by definition, all frauds demonstrate the "inadequacy" of existing controls, just as all bank robberies demonstrate the failure of bank security and all burglaries demonstrate the failure of locks and alarm systems. It does not follow from this that better financial-tracking systems would help shareholders. Spending $50 million to stop a $33 million fraud is no bargain. Indeed, no system is so foolproof that it cannot be evaded. Top managers at any firm can affect how financial results are reported, and it was the top managers in Brazil who engineered this deceit.

In September 2004 Baxter hired Deloitte Touche to beef up financial controls in Brazil, upgrading the subsidiary's system to the level required in the United States by the Sarbanes-Oxley Act. Drawing any inference from this would be incompatible with Fed.R.Evid. 407, which provides that subsequent remedial measures may not be used as evidence of liability. Quite independent of Rule 407, changing the accounting protocols does not show-that earlier ones were recognized as deficient. For all the complaint reveals, improving the financial controls in Brazil is not cost-justified and has been undertaken only as a public-relations measure, or to forestall future litigation, the cost of which easily can exceed the losses attributable to fraud. Overstating profits by $33 million does not "cost" shareholders $33 million; bogus "paper profits" differ from money out of pocket; the actual loss — funds stolen by the dishonest managers plus unnecessary volatility in stock prices and the costs of diversification designed to protect against surprises at a single firm — may have been modest. See generally John C. Coates IV, The Goals and Promise of the Sarbanes-Oxley Act, 21 J. Econ. Perspectives 91, 101-08 (Winter 2007). The judgment of Congress, reflected in the PSLRA, is that it is better to curtail baseless litigation by dismissing unfounded complaints than by hiring teams of accountants.

As for the contention that Baxter should have disclosed the news in June 2004 or the first half of July, rather than on July 22: what rule of law requires 10-Q reports to be updated on any cycle other than quarterly? That's what the "Q" means. Firms regularly learn financial information between quarterly reports, and they keep it under their hats until the time arrives for disclosure. Silence is not "fraud" without a duty to disclose. See Basic and, e.g., Dirks v. SEC, 463 U.S. 646, 103 S.Ct. 3255, 77 L.Ed.2d 911 (1983). The securities laws create a system of periodic rather than continual disclosures. See Gallagher v. Abbott Laboratories, 269 F.3d 806, 810 (7th Cir.2001); Grassi v. Information Resources, Inc., 63 F.3d 596, 599 (7th Cir. 1995). See also Gregory S. Porter, What Did You Know and When Did You Know It?: Public Company Disclosure and the Mythical Duties to Correct and Update, 68 Fordham L.Rev. 2199 (2000).

Gallagher distinguishes between a duty to update disclosures by adding the latest information and a duty to correct disclosures false when made. The Securities Exchange Act of 1934 may require the latter, though not the former, Gallagher holds. Plaintiffs maintain that their claim falls on the correction side of this line; after all, the information from Brazil was false, so the annual and quarterly statements also were false when released. But this does not mean that correction must occur as soon as the statements have been questioned. Prudent managers conduct inquiries rather than jump the gun with half-formed stories as soon as a problem comes to their attention. Baxter might more plausibly have been accused of deceiving investors had managers called a press conference before completing the steps necessary to determine just what had happened in Brazil.

Taking the time necessary to get things right is both proper and lawful. Managers cannot tell lies but are entitled to investigate for a reasonable time, until they have a full story to reveal. See Stransky v. Cummins Engine Co., 51 F.3d 1329 (7th Cir.1995); In re Burlington Coat Factory Securities Litigation, 114 F.3d 1410, 1430-31 (3d Cir.1997) (Alito, J.). After all, delay in correcting a misstatement does not cause the loss; the injury to investors comes from the fraud, not from a decision to take the time necessary to ensure that the corrective statement is accurate. Delay may affect which investors bear the loss but does not change the need for some investors to bear it, or increase its amount.

We have so far refrained from mentioning appellants' opening argument: that the district judge, having reconsidered his initial order dismissing the complaint, should not have re-reconsidered and dismissed it a second time. The law of the case bars re-reconsideration, plaintiffs maintain. We held this argument for last because it is frivolous. No matter what the district judge should have done as a matter of local procedure, the only question on appeal is whether the complaint is adequate. See, e.g., Santamarina v. Sears, Roebuck Co., 466 F.3d 570, 571-72 (7th Cir.2006). The law of the case never blocks a higher court from inquiring into the subject and making the right decision. It would be absurd for this court to remand for further proceedings, only to reverse at the end of the case because the complaint flunked the PSLRA. A district court ought not be reversed for getting to a legally required outcome by a needlessly roundabout means. None of appellants' other arguments requires comment.

AFFIRMED