OPINION AND ORDER

RESTANI, Judge:

This matter comes before the court following the court's decision in Dongguan Sunrise Furniture Co. v. United States, 865 F.Supp.2d 1216 (CIT 2012) ("Dongguan I"), and Dongguan Sunrise Furniture Co. v. United States, 904 F.Supp.2d 1359 (CIT 2013) ("Dongguan II"), in which the court remanded Wooden Bedroom Furniture From the People's Republic of China: Final Results and Final Rescission in Part, 75 Fed.Reg. 50,992, 50,992 (Dep't Commerce Aug. 18, 2010) ("Final Results") to the U.S. Department of Commerce ("Commerce"). For the reasons stated below, the court finds that Commerce has complied with the court's instructions regarding the exclusion of Insular Rattan and Native Products' ("Insular Rattan") financial statement from use in the calculation of surrogate financial ratios, but that Commerce has not provided substantial evidence for its four partial adverse facts available ("AFA") rates assigned to Fairmont's unreported sales of dressers, armoires, chests, and nightstands. Thus, Commerce's second remand results are sustained in part and remanded in part.

BACKGROUND

The facts of this case have been documented in the court's previous opinions. See Dongguan I, 865 F.Supp.2d at 1224-25; Dongguan II, 904 F.Supp.2d at 1361-62. The court presumes familiarity with those decisions but briefly summarizes the facts relevant to this opinion. In the Final Results, Plaintiffs Dongguan Sunrise Furniture Co., Ltd., Taicang Sunrise Wood Industry Co., Ltd., Taicang Fairmont Designs Furniture Co., Ltd., and Meizhou Sunrise Furniture Co., Ltd. (collectively "Fairmont" or "Plaintiff") received a rate of 43.23%, which was calculated based on a rate of approximately 34% for reported sales and a partial adverse facts available ("AFA") rate of 216.01% for unreported sales. Final Results, 75 Fed.Reg. at 50,998; Dongguan I, 865 F.Supp.2d at 1234. In Dongguan I, the court sustained Commerce's application of a partial AFA rate when calculating Fairmont's overall dumping rate. Dongguan I, 865 F.Supp.2d at 1223-32. The court concluded, however, that Commerce's selected AFA rate of 216.01% was not supported by substantial evidence. Id. at 1233-34. The court stated that Commerce failed to demonstrate that the 216.01% rate, which was calculated in a new shipper review for a different entity during a different period of review ("POR"), was relevant and reliable for Fairmont. Id. at 1233-34.

On remand, Commerce determined four separate partial AFA rates, one for each of the four types of unreported products, by selecting the single highest CONNUM-specific margin below 216.01% for the corresponding reported product type. Dongguan II, 904 F.Supp.2d at 1362. Fairmont received a rate of 39.41%, which included partial AFA rates of 182.15% for armoires, 215.51% for chests, 134.42% for nightstands, and 183.52% for dressers, which resulted in an weighted-average rate of 39.41%. Id.

In Dongguan II, the court found that Commerce's selected AFA rates were not supported by substantial evidence. Id. at 1363-64. The court stated that Commerce had failed to demonstrate a "relationship between the AFA rates chosen and a reasonably accurate estimate of Fairmont's actual rate" because the rates were based on a minuscule percentage of Fairmont's actual sales. Id. at 1363-64. Additionally, the court noted that the weighted-average margin for the reported sales, which constituted the vast majority of Fairmont's POR sales, indicated that Fairmont's actual rate would be much lower than the AFA rates selected. Id. at 1364.

In its second redetermination, Commerce selected the single-highest CONNUM-specific margin below 216% where at least 0.04% of the total reported sales for that product type were dumped at or above the selected margin. Final Results of Second Redetermination Pursuant to Court Order (July 3, 2013) (Dkt. Entry No. 160) ("Second Remand Results") at 11. This approach yielded an overall rate of 41.75%, which included partial AFA rates of 189% for armoires, 161% for chests, 140% for nightstands, and 161% for dressers.2 Id.; Analysis Memorandum for the Final Results of Redetermination Pursuant to Second Court Remand in the 2008 Antidumping Duty Review of Wooden Bedroom Furniture from the People's Republic of China, at 27 (July 3, 2012) (Dkt. Entry 165-2) ("Analysis Memorandum"). Plaintiff argues the partial AFA rates are unsupported by substantial evidence and that Commerce erred in excluding Insular Rattan's financial statements from consideration.3 Pl. Fairmont's Cmts. on Commerce's Second Remand Decision ("Pl.'s Cmts.") at 1, 6. Intervenor Defendants continue to argue that 216.01% was the appropriate AFA rate to apply to all of Fairmont's unreported sales. AFMC's Cmts. Concerning Commerce's Final Results of Second Redetermination Pursuant to Court Order at 1. Defendant argues that the selected partial AFA rates and the exclusion of Insular Rattan's financial statement comply with the court's remand order. Def.'s Resp. to Fairmont's Remand Cmts. ("Def.'s Cmts.") at 3, 9.

JURISDICTION AND STANDARD OF REVIEW

The court has jurisdiction pursuant to 28 U.S.C. § 1581(c) (2006). The court will not uphold Commerce's final determination in an antidumping duty review if it is "unsupported by substantial evidence on the record, or otherwise not in accordance with law." 19 U.S.C. § 1516a(b)(1)(B)(i).

DISCUSSION

Plaintiffs argue that Commerce's selected partial AFA rates are not supported by substantial evidence and are not in accordance with law because the rates are not reasonably accurate estimates of Fairmont's actual rate for the unreported sales. Pl.'s Cmts. at 1. Fairmont argues the court should remand the AFA rates for the same reasons cited in Dongguan I and Dongguan II, namely, the "huge divergence" between the rate calculated for the reported sales and the AFA rates, and the use of an insufficiently small percentage of sales to determine the AFA rates. Id. Additionally, Fairmont argues Commerce's stated rationale, that Fairmont made sales of the unreported product types "at prices that could have resulted in margins similar to the selected partial AFA rates," fails because the gross prices corresponding to the sales tied to the specific CONNUM-margins selected are significantly smaller than the average gross price of the unreported sales. Id. at 2-6. Plaintiffs argue, therefore, that there is no evidence that the unreported sales were made at prices that could have resulted in the rates selected. Id.

When a party has failed to act to the best of its ability, Commerce, "in reaching the applicable determination under this subtitle, may use an inference that is adverse to the interests of that party in selecting from among the facts otherwise available." 19 U.S.C. § 1677e(b). The statute permits Commerce to rely on an adverse inference, but the adverse inference does not replace Commerce's obligation to base its determinations on substantial evidence. See 19 U.S.C. § 1516a(b)(1)(B)(i). Accordingly, even an AFA rate must be supported by substantial evidence. See Gallant Ocean (Thail.) Co. v. United States, 602 F.3d 1319, 1325 (Fed.Cir.2010). "Substantial evidence requires Commerce to show some relationship between the AFA rate and the actual dumping margin." Id. In other words, Commerce must demonstrate that its selected AFA rate is "a reasonably accurate estimate of the respondent's actual rate, albeit with some built-in increase intended as a deterrent to non-compliance." Id. at 1323 (citing F.lli De Cecco Di Filippo Fara S. Martino S.p.A. v. United States, 216 F.3d 1027, 1032 (Fed.Cir.2000)).

Even when a higher AFA rate is available, and can provide a stronger deterrent to non-compliance, "Commerce may not select unreasonably high rates having no relationship to the respondent's actual dumping margin." Id. (noting the purpose of an AFA rate is to provide an incentive to cooperate, not to impose "punitive, aberrational, or uncorroborated margins" (quoting De Cecco, 216 F.3d at 1032)); Yangzhou Bestpak Gifts & Crafts Co. v. United States, 716 F.3d 1370, 1379-80 (Fed.Cir.2013) ("An overriding purpose of Commerce's administration of antidumping laws is to calculate dumping margins as accurately as possible."). Although Commerce may include a built-in increase for deterrence, Commerce cannot impose a deterrence factor far beyond the amount sufficient to deter respondents from future non-compliance. Gallant Ocean, 602 F.3d at 1324.

Here, Commerce relied on the same methodology as in the first Remand Results, with one modification:4 Commerce increased the percentage of sales relied on to at least 0.04% by quantity of the reported sales by product type, to align with the percentage relied upon in Ta Chen. Second Remand Results at 11 (citing Ta Chen Stainless Steel Pipe, Inc. v. United States, 298 F.3d 1330 (Fed.Cir.2002)). Commerce did not, however, make a material increase in the percentage of sales relied on to determine the rate for armoires and nightstands, because the AFA rates for these products were based on more than 0.04% of sales in Commerce's first redetermination. See Dongguan II, 904 F.Supp.2d at 1363 n.3.5 Accordingly, Commerce again selected the single highest CONNUM-specific margin below 216% for these two products. See Analysis Memorandum at Attach. 3. The increase in percentages for dressers and chests were marginal and resulted in the selection of the second highest margin below 216% for dressers and the third highest for chests. See id. Commerce, therefore, has not attempted to significantly change the selected rates or its methodology and instead attempts to support its original determinations with what it suggests are new explanations. Second Remand Results at 9 (noting the methodology applied was also used in the first redetermination).

Commerce argues the selected rates are supported by substantial evidence for three reasons: (1) the rates reflect Fairmont's commercial reality because they are based on Fairmont's own POR data and Fairmont experienced transaction-specific dumping margins at or above the selected margins; (2) the percentage of sales relied on are consistent with the percentage relied on in Ta Chen; and (3) any broader base or average of sales would eviscerate the adverse inference. Second Remand Results at 6-15, 20-24. The court notes that it has previously rejected the first two reasons. Dongguan I, 865 F.Supp.2d at 1234; Dongguan II, 904 F.Supp.2d at 1363-65. Commerce, however, articulates some new reasoning, which the court addresses.6

1. Commercial Reality

Commerce argues it need not rely on a larger percentage of sales to justify its selected rates because an AFA rate must reflect only a company's "commercial reality" and does not have to be an estimate of Fairmont's "average" experience. Second Remand Results at 21. Commerce defines "commercial reality" as any dumping margin, including individual transaction-specific margins, that occurred during the POR. Id. at 9 ("Each transaction ... is properly considered part of the company's commercial reality."); id. at 14 (defining commercial reality to refer to any transaction that was commercially viable). Because Commerce defines commercial reality to refer to any margin experienced by Fairmont, the term "commercial reality" adds nothing to Commerce's analysis, and Commerce's position is that an AFA rate is supported by substantial evidence as long as it is calculated from the respondent's own POR data. Second Remand Results at 11 (arguing an AFA rate is reasonable if it is tied to the respondent's actual sales and the rates here are inherently so because they are derived from actual sales).

Commerce's position is inconsistent with the Federal Circuit's directives in Gallant Ocean, which require Commerce to provide substantial evidence demonstrating a rational relationship between the AFA rate selected and a reasonably accurate estimate of the respondent's actual rate. 602 F.3d at 1325. Reliance on a single CONNUM-specific margin based on less than 0.38% of the relevant and verified record evidence, without more, does not demonstrate a reasonably accurate estimate of Fairmont's actual rate, regardless of whether the margin was calculated during the POR. This is because a single sale or CONNUM-specific margin is not generally representative of a respondent's actual rate. See Gallant Ocean, 602 F.3d at 1324 (noting that "one sale by itself does not always rise to the level of substantial evidence"). In administrative reviews, Commerce determines a respondent's rate by calculating the weighted-average of the respondent's dumping margins, see, e.g., 19 C.F.R. § 351.212(b)(1) (noting that Commerce generally determines assessment rates after a review by calculating a weighted-average of the respondent's dumping margins), and thus, a single dumping margin calculated during a review is not indicative of a respondent's overall rate, especially when that dumping margin is based on extremely small percentages (in terms of value and quantity) of reported sales. Commerce's methodology, in which the analysis stops once Commerce identifies an individual dumping margin that occurred during the POR, is not a reasonable attempt to estimate Fairmont's actual rate. See Second Remand Results at 21 (rejecting Fairmont's proposal that Commerce rely on a larger percentage of reported sales because "commercial reality" is not defined by a company's average or typical experience).

Commerce argues Gallant Ocean's directives are not applicable here because Gallant Ocean was a corroboration case based on secondary information, whereas Commerce has relied on Fairmont's own POR data here.7 Id. at 23 ("We ... ford that Gallant Ocean does not apply to the facts present here."). The statutorily imposed corroboration requirement for secondary information in 19 U.S.C. § 1677e(c) did not grant Commerce a carte blanche to use primary information8 to impose unreasonable AFA rates having no relationship to the respondent's actual rate. Congress imposed the corroboration requirement for secondary information because it recognized that the "temptation by Commerce to overreach reality in seeking to maximize deterrence" would be at its highest. Gallant Ocean, 602 F.3d at 1323. Thus, the corroboration requirement was necessary to ensure Commerce would not "overreach reality" and apply "unreasonable" rates. Id. (requiring Commerce to determine AFA rates consistent with the statutory goals of accuracy and fairness, and not distinguishing between secondary and primary information). The corroboration requirement for secondary information did not eliminate Commerce's general obligation to determine AFA rates that are supported by substantial evidence, nor does it permit Commerce to overreach reality and impose unreasonable rates as long as primary information is used. Thus, even when Commerce determines an AFA rate from information obtained during the current review, Commerce's determination must be supported by substantial evidence on the record, meaning that the record as a whole demonstrates the rate is a reasonably accurate estimate of respondent's actual rate, albeit with a built-in increase for deterrence. See id. at 1325 (noting that Commerce did not support its determination with substantial evidence when "a large body of reliable information" suggested the application of a much lower rate). Commerce's reliance on individual CONNUM-specific margins, therefore, is insufficient until Commerce provides substantial evidence to demonstrate that these particular margins represent a reasonably accurate estimate of Fairmont's actual rate, albeit with the built-in increase for non-compliance.

Commerce attempts to support its selected rates by noting that Fairmont experienced transaction-specific margins at or above the selected margins. Second Remand Results at 12, 22.9 Margins calculated from individual transactions of all products, however, are even less relevant to a reasonably accurate estimate of respondent's actual rate for each product type than CONNUM-specific margins. The dumping margin of one sale is not probative of a respondent's actual rate, especially when the quantity or value of that sale, relative to the total amount of sales, is minimal. Instead of increasing the percentage of sales relied on to determine the AFA margins, Commerce uses a smaller denominator (the Hospitality Division sales) to achieve a higher percentage and argue that 0.14% to 0.33% of sales can be considered indicative of the unreported sales. Commerce's methodology purports to rely on sales that are not "atypical" or "subject to unusual terms of sale." Second Remand Results at 12. Commerce, however, attempts to justify the selected AFA rates by comparing them to transaction-specific margins over 216%, including margins exceeding 1,500%. Id. at 22; see Analysis Memorandum at Attach. 13. Although Commerce notes there is no evidence to suggest these extremely high margins are aberrant, the absence of such evidence is not substantial evidence that these transactions are relevant to Fairmont's actual rate, and margins this high have not been shown to be within the mainstream of Fairmont's reported sales.

Moreover, it appears that under Commerce's methodology, any CONNUM-specific margin would be supported by substantial evidence because all CONNUM-specific margins are based on Fairmont's own POR data and correspond to at least some transaction-specific margins. For example, under Commerce's reasoning Commerce could select a CONNUM-specific margin for armoires of 336.72%, 189.33%, 2.04% or -157.22% because all of these margins were commercially viable during the POR and are indicative of at least some of Fairmont's individual transaction. See Analysis Memorandum at Attach. 10. There must be some rational explanation, however, as to why Commerce selected 189.33% as the margin to represent a reasonable estimate of Fairmont's actual rate for armoires. Commerce avoids this rational explanation and instead relies exclusively on its ability to draw an adverse inference to select the highest rate possible.10 Without a rational explanation linking the chosen AFA rates to Fairmont's actual rate, as opposed to merely linking the rate to individual POR dumping margins, Commerce selected an "unreasonably high rate[] having no relationship to [Fairmont's] actual dumping margin," contrary to the Federal Circuit's directives. See Gallant Ocean, 602 F.3d at 1323; PAM, S.p.A. v. United States, 582 F.3d 1336, 1340 (Fed.Cir.2009) (noting that although "the possibility of a high AFA margin creates a powerful incentive to avoid dumping and to cooperate in investigations, there is a limit to Commerce's discretion.").

2. Adverse Inference

Commerce states it cannot rely on a larger percentage of sales because this would eliminate the application of an adverse inference. Second Remand Results at 8-9, 14, 21 (rejecting the options of relying on 34%, the rate for the reported sales, relying on a larger percentage of reported sales, or relying on an average of a larger percentage of the reported sales by product type because all of these options would eviscerate the adverse inference). Commerce creates a false choice between determining rates based on less than 0.5% of reported sales by product type and rates based on 100% of all reported sales.11 This range obviously provides Commerce with a number of options to determine a reasonable estimate of Fairmont's actual rate based on a larger percentage of reported sales. Instead, Commerce has cherry-picked from the available data in order to achieve the highest-rate possible, which is both results-oriented and unreasonable. Additionally, although Commerce notes that the selected AFA rates are large enough to deter future non-compliance, Commerce provides no explanation or evidence to suggest that a lower rate would not also be sufficient. Such an explanation is particularly warranted here because the selected AFA margins are many times higher than the weighted-average margin for the reported sales, which indicates that the deterrence factor applied is far beyond the amount necessary to defer future non-compliance. See Gallant Ocean, 602 F.3d at 1324 (noting a rate over five times the highest rate imposed on similar products is far beyond an amount sufficient to deter future non-compliance). Even if Commerce relied on a larger percentage of sales and discovered the resulting rates were insufficient to deter future non-compliance, it could add an additional amount to the rate to achieve the desired deterrence factor. Commerce provides no explanation as to what the deterrence factor is in this case and why the particular amount applied is necessary, and thus, Commerce has no support for its position that there is absolutely no alternative margin or increased percentage of sales that could impose a sufficient deterrence factor.

3. Substantial Evidence

None of Commerce's new explanations justify its reliance on the minuscule percentages of sales, and the court concludes that the selected AFA rates are not supported by substantial evidence for the reasons above and for the same reasons explained in Dongguan II. 904 F.Supp.2d at 1363-65. The court finds it necessary, however, to clarify what it means by substantial evidence on the record because Commerce appears to have not understood the previous two orders. See, e.g., Second Remand Results at 8 (stating that Commerce understands the court to mean that only the weighted-average margin for all of the reported sales, regardless of product type, can reflect the "commercial reality" of Fairmont).

Generally, Commerce's ability to determine a reasonably accurate estimate of a respondent's actual rate is hampered by the lack of record data, hence the need for an AFA rate. Because of this difficulty, which arises from the respondent's own failure to cooperate, Commerce's discretion in determining an AFA rate is particularly great, and Commerce need only act reasonably in light of the record evidence, or lack thereof. See De Cecco, 216 F.3d at 1032 (noting Commerce's discretion is particularly great when determining AFA rates). Commerce's discretion, however, is not unbounded. Id. Here, Commerce abused its discretion when it decided that the reported sales by product type represent the most reliable and probative evidence as to Fairmont's unreported sales and then chose to ignore 99% of this record evidence. Unlike other AFA cases, which generally lack record evidence relating to the particular company, or of sales of a particular product or customer, the record here contains verified sales data relating to the same type of products, from the same producer, and during the same period of time as the unreported sales. See Second Remand Results at 8-10 (noting that Fairmont's experience with reported sales of the four product types are more relevant to the unreported sales than other sales data on the record).

Despite its recognition that the reported and verified sales data for armoires, chests, dressers, and nightstands are the most appropriate data from which to determine the partial AFA rates for each unreported product type, id. at 10, and its recognition that the more sales relied on, the more support for a selected rate, id. at 14, Commerce ignored the majority of the reported and verified information and instead relied on an extremely small percentage of these sales. For example, for reported sales of armoires, the record contains 66 CONNUM-specific margins under 216%, ranging from 189.33% to -157.22%. Analysis Memorandum at Attach. 3. Commerce, however, based the armoire AFA rate on a single CONNUM-specific margin, which was based on 13 out of 3,885 armoire sales, or 0.33% of total armoire sales by quantity with margins under 216%. Id. at Attach. 3, 11. As explained above, reliance on such a minuscule amount of sales provides very little insight into Fairmont's "actual rate" for the unreported armoires and ignores a large amount record evidence suggesting a much lower rate. See id. at Attach. 3 (approximately a third, by quantity, of reported armoire sales with margins under 216% are dumped at margins below 40%). Such a small percentage is particularly unhelpful here because the record evidence demonstrates that Fairmont experienced a wide range in prices, products, and dumping margins, even within each of the four product types. See, e.g., id. at Attach. 3 (the U.S. net price for reported sales of armoires range from $85 to $2174, normal value ranges from $218 to $1410, and the dumping margins range from 189.33% to -157.22%). Given that Commerce declined to consider the very evidence it identified as most indicative of Fairmont's actual rate for the unreported sales, and given that the disregarded record evidence suggests a reasonably accurate estimate of Fairmont's actual rate would be much lower, Commerce's determinations are not supported by substantial evidence.

CONCLUSION

The court has twice informed Commerce that without a rational explanation as to why a small percentage of sales can be considered representative of Fairmont's actual rate, especially given the large amount of record evidence available and the diversity in Fairmont's products and sales, Commerce's reliance on minuscule percentages of sales to determine the partial AFA rates does not constitute substantial evidence. Commerce failed to provide a rational explanation in two attempts, and a third attempt is unnecessary. On remand, Commerce is directed to rely on a significant portion of the available evidence, already identified by Commerce as relevant and reliable, to determine the partial AFA rates. Commerce cannot substitute the adverse inference for substantial evidence demonstrating that the selected rates are related to Fairmont's actual rate. Commerce may impose an increase to deter future non-compliance, but it must demonstrate that the amount does not go substantially beyond what is necessary to deter non-compliance.

In all other respects, the Second Remand Results are sustained. Commerce shall file its remand determination with the court within 60 days of this date (November 4, 2013). The parties shall have 30 days thereafter to file objections (December 4, 2013), and the Government will have 15 days thereafter to file its response (December 19, 2013).