MEMORANDUM *

Defendant Norman Taylor appeals the district court’s imposition of a 140-month *16sentence following Taylor’s unconditional plea of guilty to one count of distribution of cocaine. Because the parties are familiar with the factual and procedural history of this case, we will not recount it here.

I

In this case, the district court determined drug quantity by employing the “preponderance of the evidence” standard. The district court erred in doing so. When a sentencing factor has an extremely disproportionate effect on the sentence, the government must prove the sentencing enhancement by clear and convincing evidence. United States v. Jordan, 256 F.3d 922, 930 (9th Cir. 2001); United States v. Romero-Rendon, 220 F.3d 1159, 1160 (9th Cir. 2000). Viewing the evidence in the light most lenient to Taylor, the government had proof beyond a reasonable doubt of 2 grams of cocaine base-an amount that would have resulted in a base offense level of 20. This base level results in a sentencing range of 41-51 months. In contrast, the uncorroborated hearsay evidence of the sole testifying witness to Taylor’s drug trafficking exposed Taylor to a base offense level of 32, and a corresponding sentencing range of 151-188 months. See U.S.S.G. § 2D1.1. In other words, the increase resulting from the witness’s testimony had a strongly disproportionate impact-a threefold increase-on his sentence.

We have imposed the clear and convincing evidentiary standard in other cases where the effect of a sentencing finding had a similar impact. See Jordan, 256 F.3d at 929 (holding that failure to apply clear and convincing evidence standard was plain error when district court increased offense level by nine levels and sentence range increased from 70-87 months to 151-188 months); United States v. Munoz, 233 F.3d 1117, 1127 (9th Cir. 2000) (applying clear and convincing standard where sentencing factor increased defendants’ individual sentencing ranges from 12-18 months to 41-51 months); United States v. Mezas de Jesus, 217 F.3d 638, 643 (9th Cir. 2000) (applying heightened standard where sentencing factor increased range from 21-27 months to 57-71 months). Thus, given the substantially increased sentence sought by the government, the court should have required proof of drug quantity by clear and convincing evidence.

II

Although the district court erred by employing the incorrect evidentiary standard, that does not end our analysis: we must also assess whether the error was harmless. If the record contains evidence sufficient for us to conclude, by applying the clear and convincing standard, that the district court’s determination of drug quantity was correct, then we may affirm. Romero-Rendon, 220 F.3d at 1163. However, the record in this case is not sufficient for us to make that conclusion.

No witness with personal knowledge of the facts testified at sentencing. The only person who testified was the probation officer, Leo Rivers, who compiled the pre-sentence report. He based his estimate entirely on the uncorroborated statements made by a cooperating witness, Manual Galvan, who had been detained on murder charges. Rivers had never met or interviewed Galvan, but relied on reports by other officers who had interviewed him. Rivers admitted that another defendant, Owen Fifer, had been the primary target of the investigation, “[s]o there wasn’t a lot of information regarding drug distribution with Mr. Taylor.”

Rivers calculated quantity based on estimates provided by Galvan, without the benefit of any direct evidence or information. Galvan speculated on the amount of *17drugs Taylor may have distributed based on an estimate of the amount of drugs he brought from California for a group of dealers and how he thought the drugs might have been distributed. He had no direct knowledge of how the drugs he brought back were actually distributed, and no direct knowledge as to the actual drug quantity properly attributable to Taylor. Galvan’s estimates of undistributed drug quantity were also not internally consistent.2 Further, Officer Rivers agreed on cross-examination that it was completely possible that Galvan had exaggerated the amounts attributable to Taylor. Perhaps in recognition of this, Rivers then “arbitrarily” reduced the estimated amount.

The government argues that there was evidence supporting the finding of drug quantity in the trial of co-defendant Fifer, and the district court referenced the trial at the sentencing hearing.3 However, the trial transcript was not introduced at the sentencing hearing, the government provided no specific citations to the record, and a review of the transcript of the Fifer trial does not support the government’s assertion. In fact, no one testified to Taylor’s association with any quantity of drugs; the trial transcript “support” the government cites in its briefs, quite simply, does not exist. In short, Galvan’s state-merits and estimates were not corroborated in any way by the trial testimony of Taylor’s co-defendants — even assuming it was properly considered by the sentencing court.

Given, among other considerations, that the determination of drug quantity was based on the uncorroborated hearsay estimate of a cooperating witness without direct personal knowledge, and that the government admitted that its final determination of drug quantity was based on an arbitrary reduction, we cannot say that the record of this case supports the district court’s finding by clear and convincing evidence. Thus, the sentence must be vacated and the case remanded for resen-tencing.

Ill

The district court did not clearly err in imposing a two level increase for possession of a firearm during a drug-trafficking crime pursuant to U.S.S.G. § 2Dl.l(b)(l). Application Note 3 to § 2Dl.l(b)(l) explains that “[t]he enhancement for weapon possession reflects the increased danger of violence when drug traffickers possess weapons.” The enhancement may be applied “if the weapon was present, unless it is clearly improbable that the weapon was *18connected with the offense.” Id. In applying this enhancement, a district court “need not find a connection between the firearm and the offense. If it finds that the defendant possessed the weapon during the commission of the offense, the enhancement is appropriate.” United States v. Lopez-Sandoval, 146 F.3d 712, 714 (9th Cir. 1998) (citation and internal quotation marks omitted). A defendant “need not be carrying a weapon when he is arrested.” Id. Furthermore, for the purposes of the firearm enhancement, a district court can properly examine “all the offense conduct, not just the crime of conviction.” United States v. Willard, 919 F.2d 606, 610 (9th Cir. 1990) (citing USSG § 1B1.3(a)(2)).

According to the pre-sentence report, Galvan observed Taylor in possession of a plastic-gripped dark-colored semi-automatic handgun equipped with a laser sight during the course of his drug trafficking activities. This statement, partially corroborated by other sources, was sufficient to justify the adjustment.

IV

Taylor argues that his due process rights were violated by exposing him to a sentence in excess of the statutory maximum in violation of Apprendi v. New Jersey, 530 U.S. 466, 120 S.Ct. 2348, 147 L.Ed.2d 435 (2000). Given that argument, we deferred submission of this case pending the issuance of the en banc opinion in United States v. Buckland, 277 F.3d 1173 (9th Cir. 2002) (en banc), because Buckland involved Apprendi issues. However, there is nothing in Buckland that would alter the analysis of this case. Taylor’s due process rights, as articulated in Apprendi, were not violated. Taylor was sentenced to prison for 140 months, less than the lowest possible statutory maximum for cocaine distribution, which is 20 years. See 21 U.S.C. § 841(b)(1)(C). Thus, even if the district court’s finding of drug quantity increased the prescribed maximum to which Taylor was exposed, he “was not, for purposes of Apprendi, prejudiced by the error because the sentence he received fell well below the statutory maximum for the offense to which he pleaded guilty.” United States v. Scheele, 231 F.3d 492, 497 n. 2 (9th Cir. 2000).

V

Because the district court applied the incorrect standard of proof at sentencing concerning drug quantity and the error cannot be shown on this record to be harmless, we must reverse the judgment of the district court and remand for re-sentencing. In doing so, we do not limit the district court’s sentencing discretion on remand, nor do we confine the parties to the existing record. United States v. Matthews, 278 F.3d 880, 889-90 (9th Cir. 2002) (en banc).

REVERSED AND REMANDED FOR RE-SENTENCING.

concurring in part and dissenting in part.

I concur in Parts III and IV but in my view, the district court did not err in calculating drug quantity.

First of all, Taylor does not dispute that the standard of proof is preponderance of the evidence. This is for good reason. The calculation of quantity here involved charged conduct; it is not part of a sentencing enhancement, or uncharged relevant conduct, that may call for proof by a clear and convincing standard. That distinguishes each of the cases upon which the majority relies. United States v. Jordan, 256 F.3d 922 (9th Cir. 2001) (enhancement for firearm possession and abduction to escape); United States v. Romero-Ren-*19don, 220 F.3d 1159 (enhancement based on prior aggravated felony); United States v. Munoz, 233 F.3d 1117 (9th Cir. 2000) (uncharged conduct); United States v. Mexas de Jesus, 217 F.3d 638 (9th Cir. 2000) (same). Even so, Taylor argues that instead of a quantity that put him at base level 32, he should have had a base level of 28. With a three-level decrease for acceptance of responsibility, plus two for use of a firearm, that would have given him an offense level of 27. At criminal history category III, Taylor’s preferred guideline range was 87-108 months. Instead, his range at level 32 was 151-188. This isn’t a severe difference at all.

Regardless, the Probation Officer relied on the two grams of crack sold in a controlled buy on March 5, 13, and 19, 1998, together with Galvan’s statement that he made two-three trips per month for twelve months delivering “at a bare minimum” one kilogram of cocaine to Taylor, Fifer and someone else — which he believed they divided equally. Rivers assumed 24 trips. Attributing one third to Taylor for each trip, this comes to eight kilograms, which Rivers then arbitrarily decreased by one third to 5.33 kilograms to give Taylor the benefit of the doubt that Galvan might be inflating his numbers. I am not firmly convinced that the district court incorrectly found this rehable, because field agents who knew about the entire operation thought Galvan’s version was credible, and Galvan showed that he — not Taylor — was believable in that Galvan identified Taylor while Taylor said that he didn’t know Gal-van.

While it might have been plausible for the court to accept Taylor’s submission that Galvan’s information should have been discounted by one half instead of by one third, his estimate is not equahy plausible; it is equahy arbitrary. Although a fifty percent discount had been used in United States v. August, 86 F.3d 151 (9th Cir. 1996), there is no authority requiring it or indicating that fifty percent is a sufficiently cautious discount but thirty-three percent is not. In any event, fifty percent of the high-end Galvan estimate would not benefit Taylor any more than one third of his low-end estimate.

Finally, Taylor sought no evidentiary hearing at which he could have — but did not — impeach Galvan’s submission. In these circumstances, I can’t say that the district court got it wrong. Galvan was in the ballpark, and his estimate was discounted to take account of exaggeration and then again (by one third) to take account of the obligation to be cautious and err in favor of the defendant. I would, therefore, affirm.