WOOD, Circuit Judge.

After issuing a number of citations against Gunite Corporation for violations of occupational safety and health regulations, the Secretary of Labor failed to convince the Occupational Safety and Health Review Commission to uphold four of the charges. The Secretary has petitioned this court to reverse the Commission’s decision. We conclude that the Secretary is correct: the Commission’s decision is not supported by substantial evidence in the record and therefore the case must be remanded to the agency with instructions to affirm the citations.

Gunite’s foundry in Rockford makes brakes and wheels for heavy trucks. Its process involves melting scrap iron and then pouring the molten iron into molds created from a mixture of sand, clay, and water. The molds then pass along a series of interconnecting conveyor belts that transport and cool the iron pieces. As they move along the conveyer belts, the castings are shaken from the molds; in the process, dust containing respirable silica becomes airborne. The amount of this dust is enormous; the process uses some 400 tons of sand per hour. Breathing silica is dangerous for the foundry’s workers, as it can lead to silicosis, a deadly disease that primarily affects the lungs. The Occupational Safety and Health Ad *553 ministration (OSHA) has accordingly set ceilings called permissible exposure limits, or PELs, on the amount of silica that may be present in the air. See 29 C.F.R. § 1910.1000(c).

Gunite’s foundry was built in the first half of the twentieth century. From the start, it has been plagued with the problem of controlling the amount of silica dust escaping into the air. In 1977 and again in 1981, OSHA cited Gunite for violations of the silica PEL. That problem has intensified since the installation of the conveyer belt system in 1989. The plant manager described the initial installation of the conveyor belts as “a disaster.” In order to control the airborne dust, the plant first tried spraying water to keep the dust down. When that failed to make a difference, Gunite installed covers over the conveyer belts. They too were ineffective, even though they were still being used several years later when OSHA entered the picture. In 1990, one of Gunite’s insurers reported that employee exposure to silica exceeded a different measure, the “threshold limit value” set by the American Conference of Governmental Industrial Hygienists. Two upgrades later, Gunite still had too much silica in the air. Another insurer measured the air four times between June 1996 and March 1998 and found that foundry employees — including those at the positions listed in the citations at issue before us — were being exposed to levels of respirable silica in excess of OSHA’s PEL. That insurer, Kemper-NATLSCO, recommended in 1996 that Gunite require its employees to wear individual respirators until the company could implement feasible engineering and administrative controls to limit employee exposure. Gunite seems to have ignored that recommendation; two reports from Kem-per-NATLSCO in 1997 indicated that employees still were not being required to wear the individual respiratory protection. In 1996 and 1997, Gunite recorded three cases of silicosis in its OSHA logs. Gunite itself describes its efforts to deal with excess silica from 1991 through 1998 as involving four major engineering projects that together were intended to bring the foundry into compliance with the silica PEL and other federal regulations. The last of these, installation of new covers and a ventilation system over the conveyor belts, was planned and being implemented in 1998 during the OSHA inspection, though it did not become fully functional until March 1999.

Since 1971, OSHA regulations have required facilities with excess respirable silica to use engineering or administrative controls “whenever feasible” to attain compliance with the PEL. 29 C.F.R. § 1910.1000(e). Only when feasible engineering or administrative controls are insufficient to bring silica levels below the PEL may a company turn to individualized protective equipment to supplement those controls. Id. This “hierarchy of controls” privileges engineering and administrative controls because they “make respiratory protection automatic, while respirators are dependent on use and constant attention and are subject to human error.” American Iron & Steel Inst. v. OSHA, 182 F.3d 1261, 1269 (11th Cir.1999). See also Advance Bronze, Inc. v. Dole, 917 F.2d 944, 947 n. 2 (6th Cir.1990) (discussing the hierarchy in the context of lead standards).

The OSHA inspection leading to the citations involved in the Secretary’s petition took place between May and October of 1998. During the inspection, OSHA representatives took samples that showed that workers in four positions at the foundry were exposed to about 1.6 times the PEL for respirable silica in an eight-hour shift. OSHA assigned three members of its Health Response Team (HRT) to Gunite’s case and asked them to evaluate Gunite’s *554 administrative and engineering controls. The HRT came up with a list of proposed administrative and engineering controls that it concluded would alleviate the airborne silica problem. The team based its recommendations for engineering controls “on general principles of ventilation and industrial hygiene which have been shown to be effective in reducing contaminant levels in a variety of industries.” One control measure highlighted in the report was the use of “clean air islands,” which are devices that blow clean air at the at-risk employees; the fresh air creates a bubble around each employee that does not contain dangerous levels of silica. Other measures on the list included installing physical barriers to block the areas where the most dust was kicked up into the air and improving housekeeping and maintenance. Finally, the report mentioned the new system of covers for the conveyor belts that Gunite was in the process of implementing during the OSHA inspection, although it opined that the new system would solve the problem for only three of the four employee positions that were overexposed to silica.

Based on its investigation of the foundry, the Secretary issued three citations containing various items, each alleging violations of federal regulations. Among those were six items based on the sampling results and the HRT’s report alleging that Gunite had committed serious and willful violations of 29 C.F.R. § 1910.1000(c) (the air contaminant regulation) by exposing employees to respirable silica in amounts in excess of OSHA’s PEL and of 29 C.F.R. § 1910.1000(e) by failing to implement feasible engineering or administrative controls. Another item alleged a violation of 29 C.F.R. § 1910.134(e)(4) (1997) by failing to inspect to ensure proper respirator use. The Secretary also alleged a willful violation of 29 C.F.R. § 1910.95(g)(6) for failure to obtain annual audiograms. Of these charges, only four are at issue in this petition: items 8a and 8b of citation 1, which charge serious violations of § 1910.1000(c) and (e) for overexposing three “mold station” workers (a metal pourer, coreset/blowoff operator, and mold line technician) to res-pirable silica and for failing to determine and implement feasible administrative or engineering controls to achieve compliance with the PEL; and items 3a and 3b of citation 2, which charged willful violations of the same standards for overexposing a sprue pulloff operator, who works at a different location in the factory closer to the finishing process.

Gunite appealed the citations, contending before the administrative law judge (ALJ) that it should not be liable because it was already implementing a new system designed to alleviate the respirable silica problem and because of the availability of individual respirators, which it contended both alleviated overexposures and qualified as an administrative control. In the pre-hearing documents made part of the record by the ALJ, the Secretary designated two members of the HRT team — industrial hygienist Keith Motley and mechanical engineer Lee Hathon — as experts. They were expected to testify about their qualifications, their observations of Gunite’s foundry, and “administrative and engineering controls to reduce respirable silica” for the locations identified as having overexpo-sures, as well as the contents of the HRT report. Motley’s expertise included 12 years of experience as part of the HRT responsible for addressing respirable hazards, while Hathon had served 10 years on the HRT and had participated in investigations “at several foundries and other industries where airborne silica is a hazard.” Gunite objected to both Motley’s and Ha-thon’s testimony about actual silica levels as “not probative of exposures of the cited *555 employees to the cited levels of respirable silica dust.” Nevertheless, Gunite did “not deny that [Motley’s and Hathon’s] opinions are probative relative to the question of engineering controls for some of the cited work areas,” although the company reserved the right to disagree substantively with their opinions “in some respects.” At the hearing, the parties stipulated to the admission of the HRT report and agreed that the Secretary would not call the HRT members for direct examination, leaving their testimony in the form of the report itself and other pre-hearing filings made part of the administrative record. The Secretary, however, planned to and did present them for cross-examination.

Both Motley and Hathon were cross-examined about their qualifications, about their preparation of the HRT report, and about the recommended engineering controls. Motley testified about his tour of the plant, during which he was able to observe firsthand the problem areas and ventilation systems. When he asked to see the plans for the new ventilation system, he was shown the actual parts that Gunite was putting in place. Motley also described the team’s particular recommended solutions to the airborne silica problem, including a way generally to filter the plant’s air before recirculating it. He explained how clean air islands work and how they might alleviate the problems. Hathon’s testimony was similar. After answering questions about his own training and credentials, Hathon testified that he had previously examined at least seven foundries similar to Gunite’s in terms of size, age of the building, and products produced; that he had toured the areas of Gunite’s foundry with overexposures; and that the HRT report was designed specifically to address the problems of the Gunite foundry.

Another witness, Julia Evans, an OSHA compliance officer, testified that an administrative control such as employee rotation likely would have eliminated the overexposure. Evans also testified that Gunite’s planned improvements likely would solve the silica problem at three of the four employee stations.

Finally, Gunite’s own witness, Leroy Ca-tor, the 50-year veteran employee in charge of the abatement process, testified that “[c]lean air islands are probably effective and I don’t question that.” However, he also said that they are difficult to implement because of temperature control issues, and that they had not been recommended by the outside engineers working with Gunite. Instead, those engineers recommended systemic approaches that would improve the air for many employees rather than individualized approaches. He admitted, however, that clean air technology was used elsewhere in the foundry, near the pouring line.

The ALJ affirmed the four citations (as well as the others not at issue here), finding that “except for clean air islands, Gun-ite has not challenged [the HRT’s] recommendations.” The ALJ also found that Gunite’s future plans to solve the problem did not relieve the company of liability. Likewise, the use of respirators did not alleviate Gunite’s obligation to implement systemic administrative or engineering controls that would make individual respirators unnecessary.

Gunite appealed the ALJ’s determination to the Commission, which by a divided vote affirmed some of the citations, but vacated the four now before us. The majority found that the Secretary had failed to prove that the proposed engineering and administrative controls would produce a “significant reduction” in respirable silica:

*556 The Secretary’s case for establishing technological feasibility rests primarily on OSHA’s HRT report and supporting testimony by compliance officer Evans and HRT members Lee Hathon and Keith Motley. Neither compliance officer Evans nor the HRT members were qualified as experts. The HRT report identified deficiencies in Gunite’s controls and recommended additional controls, including general ventilation to reduce plantwide levels of air contaminant and specific controls to address areas where sampling results showed employee exposure in excess of the PEL.... We conclude that the evidence of record as a whole is insufficient to prove that the controls suggested by the Secretary would produce a significant reduction in airborne respirable silica in the foundry. Because neither compliance officer Evans nor any of the HRT members were presented by the Secretary as expert witnesses, the record lacks sufficient evidence to establish that the proposed controls were technologically feasible. Moreover, the testimony failed to quantify the expected or anticipated amount of silica dust reduction. At most, the HRT report provided a list of control technologies for Gunite to experiment with in the hope that some of them or some combination of them would reduce employee exposure to some undefined levels.

The Commission then found that because the Secretary had failed to show a technologically feasible engineering control, the use of respirators by employees was sufficient. (Somewhat inexplicably, the Commission found that the evidence of the respirator use on the day the air was tested was sufficient to vacate those items, even though elsewhere it affirmed a separate item by finding that “Gunite’s lax enforcement of respirator use in the foundry constituted willfulness.”)

The dissenting commissioner found that the HRT report was “comprehensive” and that it recommended a number of feasible administrative and engineering controls. She also concluded that the witnesses’ credentials made their testimony “sufficiently reliable” and that Gunite had failed to challenge the witnesses’ expertise. Furthermore, she wrote, “[T]he fact that the Secretary did not present the HRT members as experts does not diminish the probative value of their testimony.... To the extent that the majority would find dispos-itive the lack of ‘expert’ testimony from the Secretary in order to meet her burden, and since this issue has not been previously briefed, I would remand.”

This court has jurisdiction under 29 U.S.C. § 660(b) to review on the Secretary’s petition “any final order” of the Commission. The Commission’s “function is to act as a neutral arbiter.” Cuyahoga Valley Ry. Co. v. United, Transp. Union, 474 U.S. 3, 7, 106 S.Ct. 286, 88 L.Ed.2d 2 (1985) (per curiam). Where the Commission reverses an ALJ, it is the Commission’s order alone that is reviewed. See Marshall v. West Point Pepperell, Inc., 588 F.2d 979, 984 (5th Cir.1979). In reviewing the Commission’s conclusions in enforcement actions, we follow the dictates of the Administrative Procedure Act, 5 U.S.C. § 701 et seq., and review interpretations of law with deference to determine only whether they are “arbitrary or capricious” or contrary to law. See 5 U.S.C. § 706; Sierra Resources, Inc. v. Herman, 213 F.3d 989, 992 (7th Cir.2000); Caterpillar, Inc. v. Reich, 111 F.3d 61, 62 (7th Cir.1997). Where, however, a dispute exists between the Secretary and the Commission about an interpretation of OSHA or regulations adopted under its umbrella, *557 the Secretary’s interpretation is entitled to deference if it “sensibly conforms to the wording and purpose” of the relevant provisions. Martin v. OSHRC, 499 U.S. 144, 150-51, 111 S.Ct. 1171, 113 L.Ed.2d 117 (1991) (quoting Northern Indiana Pub. Serv. Co. v. Porter Cty. Chapter of Izaak Walton League of America, Inc., 423 U.S. 12, 15, 96 S.Ct. 172, 46 L.Ed.2d 156 (1975)). In reviewing the facts, we uphold the Commission if its findings are “supported by substantial evidence on the record considered as a whole.” 29 U.S.C. § 660(a); see Sierra Resources, 213 F.3d at 992. Our deference to the Commission includes deference to its credibility determinations, except in extraordinary circumstances. Id.

The Secretary bears the burden of proof in demonstrating the feasibility of administrative and engineering controls “when the compliance remedy is based upon a very general statutory or regulatory command that does not describe for the employer any specific methods for compliance.” Faultless Div., Bliss & Laughlin Indus., Inc. v. Sec’y of Labor, 674 F.2d 1177, 1189 (7th Cir.1982); Modern Drop Forge Co. v. Sec’y of Labor, 683 F.2d 1105, 1113 (7th Cir.1982). As the Commission itself has noted, however, “[t]he test of whether administrative and/or engineering controls are technologically feasible is whether the controls are ‘achievable’ and capable of producing a significant reduction in exposure to air contaminants.” Sec’y of Labor v. G & C Foundry Co., 17 O.S.H. Cas. (BNA) 2137, 1997 WL 447196, *4 (Rev. Comm’n 1997). Although a significant reduction is required, it is not necessary to show that the control would achieve full compliance. Id. “The Secretary need not [] propose or prove the feasibility of a detailed abatement program. [She] need only show that some controls are feasible in an employer’s plant.” Sec’y of Labor v. Great Falls Tribune Co., 5 O.S.H. Cas. (BNA) 1443, 1977 WL 8017, *3-4 (Rev. Comm’n 1977) (internal citation omitted). There is no magic percentage of reduction that is required; all that is required is that the administrative or engineering control be systemic — so that, unlike with respirators, an individual employee’s mistake cannot eviscerate her protection — and that it produce a significant reduction in silica.

In this case, the Commission found that the Secretary failed to carry her burden on feasibility, but it is hard to discern why it came to this conclusion. On one reading of the opinion, it seems that the Commission might be saying that the Secretary failed either to present any evidence of feasibility or that she failed to present necessary expert evidence of feasibility, and thus failed to meet her burden of proof. This, at least, is the way one might understand the portions of the Commission’s opinion stating that “neither compliance officer Evans nor any of the HRT members were presented by the Secretary as expert witnesses.... ” and “the Secretary offered no expert testimony in attempting to meet her burden of proof.” In the alternative, the opinion might be interpreted as finding that the Secretary’s witnesses lacked expertise, as where the Commission writes that “[n]either compliance officer Evans nor the HRT members were qualified as experts.” Or the Commission might have been faulting the Secretary for failing to jump through some procedural hoops in presenting her evidence or designating her experts.

Our problem with the Commission’s opinion is twofold. First, it did not adequately explain why it concluded that the Secretary failed to satisfy her burden. It is neither proper nor feasible for us to fill in the blanks with our own guesses. Second, even if we somehow succeeded in *558 constructing a rationale for the Commission’s result, we conclude that its factual conclusions are contrary to the “substantial evidence on the record considered as a whole.” Finally, nothing in the record hints at any purely procedural failing in the Secretary’s handling of the case. The Commission generally follows the Federal Rules of Civil Procedure, 29 C.F.R. § 2200.2(b), but it does so in a manner designed to “secure an expeditious, just and inexpensive determination of every case.” 29 C.F.R. § 2200.2(c). It is the role of the ALJ, among other things, to “[r]ule upon offers of proof and receive relevant evidence” and “introduce into the record documentary or other evidence”; the judge may also ask the parties to state their positions on issues. 29 C.F.R. § 2200.67(d), (j), (k). Witnesses generally must testify under oath or affirmation and must be subject to cross-examination, 29 C.F.R. § 2200.69, and the Federal Rules of Evidence govern the proceedings. 29 C.F.R. § 2200.71. Gunite has not pointed to any particular problem under those rules, and so we conclude that the Commission could not have rejected the ALJ’s opinion on this ground.

The other three possibilities all, in one way or the other, attack the sufficiency of the evidence the Secretary presented to support the citations. In applying the substantial evidence rule, our point of reference is the Commission’s opinion. As we have said before, “[w]e cannot uphold a decision by an administrative agency ... if ... the reasons given by the trier of fact do not build an accurate and logical bridge between the evidence and the result.” J.C. Penney Co. v. NLRB, 123 F.3d 988, 995 (7th Cir.1997) (quoting Sarchet v. Chater, 78 F.3d 305, 307 (7th Cir.1996)). See also Blakes ex rel. Wolfe v. Barnhart, 331 F.3d 565, 569 (7th Cir.2003).

If we interpret the Commission’s decision as holding that the Secretary utterly failed to present any evidence of feasibility that could satisfy her burden of proof, we can easily reject that conclusion as contrary to the record as a whole. This would be obvious if the Secretary had presented Hathon and Motley as witnesses on direct, but the method of proceeding to which all parties agreed does not change the result. The parties stipulated that the HRT report would be admitted into evidence, and it was. Then, as agreed, the Secretary presented both Motley and Hathon for cross-examination. Their testimony on cross covered much of the same material that a direct examination normally would have covered — their credentials, their inspection of the foundry, and the basis for their report and conclusions. Furthermore, the ALJ specifically made part of the record the pre-trial pleadings, see 5 U.S.C. § 556(e), which also covered much of the same ground. Even the Commission acknowledged that “Gunite challenged only the feasibility of clean air islands” and not the many other recommended controls, which included basic housekeeping, employee hygiene, and employee rotation. Indeed, the parties agreed that the already planned — but not implemented — improvements would significantly decrease the silica exposure for the molding line employees. In the absence of any rebuttal or specific evidence about nonfeasibility from Gunite, the Secretary’s evidence is more than sufficient to carry her burden. See, e.g., Sec’y of Labor v. Smith Steel Casting Co., 15 O.S.H. Cas. (BNA) 1001, 1991 WL 81019, at *10 (Rev. Comm’n 1991) (finding sufficient evidence when the testimony of the Secretary’s single witness on technological feasibility was undisputed). In administrative litigation as elsewhere, if both sides agree on a matter, such as the fact that the planned changes would significantly decrease the *559 level of silica in the foundry’s atmosphere, there is no requirement that evidence be introduced to support that fact or conclusion. A contrary rule would bog down the administrative process with unnecessary evidence on uncontested facts. The feasibility inquiry is governed by “realism and common sense.” Sec’y of Labor v. Sherwin-Williams Co., 11 O.S.H. Cas. (BNA) 2105, 1984 WL 34904, *7 (Rev. Comm’n 1984) (quoting Donovan v. Castle & Cooke Foods, a Div. of Castle & Cooke, Inc., 692 F.2d 641, 650 (9th Cir.1982)). While the Commission majority tried to distinguish Sherwin-Williams and Castle & Cooke Foods by suggesting that expert testimony may be unnecessary only where the feasibility challenged is economic (as it was in those cases) rather than technological (as in Gunite’s case), we see no principled distinction between the two issues. “Realism and common sense” are equally applicable in both realms.

If we think instead that the Commission took the position that Hathon and Motley were not experts, lacked expertise, or were somehow not properly designated as experts, that interpretation likewise collapses in the face of the record. We have said that a court excluding expert testimony must “articulate with reasonable specificity the reasons why it believes the testimony is insufficiently reliable to qualify for admission,” because otherwise the lack of such explication makes it difficult (or impossible) for us meaningfully to review the court’s decision. Mihailovich v. Laatsch, 359 F.3d 892, 919 (7th Cir.2004). Here, the Commission’s opinion suggests less an evaluation of the witnesses’ testimony and more “an utter disregard for uncontroverted sworn testimony” and other evidence presented by the Secretary. Jet Star, Inc. v. NLRB, 209 F.3d 671, 676 (7th Cir.2000) (quoting Carry Cos. of Ill., Inc. v. NLRB, 30 F.3d 922, 928 (7th Cir.1994)). Especially in administrative adjudication, there is no magical set of procedures for designating someone as an expert witness. As we observed in United States v. Williams, 81 F.3d 1434, 1442 (7th Cir.1996), the “difference between an expert witness and an ordinary witness is that the former is allowed to offer an opinion, while the latter is confined to testifying from personal knowledge.” The test is whether the witness has “specialized knowledge that the lay person cannot be expected to possess” and reasonably applies that knowledge to the relevant facts. United States v. Conn, 297 F.3d 548, 554-55 (7th Cir.2002). In this case, both the Secretary and Gunite recognized Hathon’s and Motley’s expertise, as Gunite admitted that their testimony would be probative on the issue of feasible engineering controls.

Gunite’s argument on appeal is primarily limited to a reminder that the court should defer to the Commission’s decision when that decision is supported by substantial evidence, but its brief is remarkably short on specific evidence that might have supported this particular ruling. If the Commission had made a reasoned decision supported by substantial evidence in the record, we would defer to that decision, even if we ourselves might have preferred another outcome. Such deference extends to credibility determinations about witnesses, including expert witnesses. See Sierra Resources, 213 F.3d at 992. Nonetheless, deference has its limits. In order for us to uphold an administrative commission’s decision, there must be “an accurate and logical bridge between the evidence and the result.” J.C. Penney, 123 F.3d at 995. That same bridge is necessary between the evidence in the record and the credibility determination about expert testimony. The Commission’s opinion in this case does not meet those standards. No matter how we look at it, the opinion is not supported by substantial evidence.

*560 Finally, we address two additional arguments that Gunite briefly advances. The first is that the employees’ use of individual respirators solved the problem so well that Gunite should not be liable. But the hierarchy of controls, as we noted above, requires feasible administrative or engineering controls; it permits individualized protective equipment such as respirators only when such systemic administrative or engineering controls are not feasible. See 29 C.F.R. §§ 1910.1000(e), 1910.134; American Iron & Steel Inst., 182 F.3d at 1269. Nothing in this case provides any basis for rethinking the hierarehy-of-controls policy, which is now more than three decades old. The fundamental reason the regulation privileges systemic over individualized controls is that the latter can be undermined by workers’ failure to use them or their misuse of them. Gunite’s foundry provides a textbook example, as evidence in the record indicates that some employees chose not to use the individualized equipment because it was uncomfortable, while others misused it, thereby diminishing their own protection. The Commission itself noted that Gunite’s insurer found that its “management admitted it was not enforcing its policy requiring respiratory protection in areas where respirator use was deemed necessary,” using that admission to justify a finding of willfulness on another item not before this court.

Second, Gunite contends that at the time of the citations, it had already planned and even begun to implement the engineering controls that the Secretary agrees likely would solve the problem for the mold line employees. Thus, Gunite argues, it should not be liable. Indeed, the Secretary took the planned engineering controls into account when she classified the citation based on the overexposure of the mold station workers as serious rather than willful. See United States v. Ladish Malting Co., 135 F.3d 484, 490 (7th Cir.1998) (discussing the four levels of violations of occupational safety regulations, which control levels of penalty). Nevertheless, the planned fixes are not an escape from liability. The problem with airborne silica in the Gunite foundry was hardly new in 1998. Even the Commission found that Gunite did not act in good faith in addressing the problem, noting that Gunite “did little in the way of controlling dust problems through the use of housekeeping measures,” failed to maintain its ventilation and ducts, and moved at a “glacial pace” in solving the engineering problems. The problem of excess silica in the air resulting from the shaking of the conveyor belts was clear in 1989; the fact that Gunite was still in the process of addressing the problem a decade later cannot and should not absolve the foundry of liability. Future solutions simply cannot erase past violations, any more than they can retroactively protect the foundry’s employees’ lungs.

For these reasons, we Revekse the Commission’s decision and ReMand the case with instructions to affirm the four contested citations.