STEPHEN F. WILLIAMS, Circuit Judge:

The Federal Energy Regulatory Commission certified Transcontinental Gas Pipe Line's Southern Expansion service, subject to a condition altering its proposed rate design; on rehearing, it reversed course and accepted Transco's proposed design; thereafter, it twice modified the order, ultimately ruling that the approved rates should become effective only a year after the start of service. Transcontinental Gas Pipe Line Corp., 51 FERC ¶ 61,173 (1990); 56 FERC ¶ 61,037 (1991); 57 FERC ¶ 61,331 (1991); 62 FERC ¶ 61,211 (1993); 64 FERC ¶ 61,099 (1993). Transco argues that the rates should be made effective as of the start of service. Although we disagree with Transco's contention that the Commission was jurisdictionally barred from its various tinkerings with the effective date, we find its arguments on the merits in substantial part persuasive. We thus reverse the orders below in part and remand for further proceedings consistent with this opinion.

* * *

On September 2, 1988, Transco filed an application with the Commission pursuant to Section 7(c) of the Natural Gas Act, 15 U.S.C. § 717f(c) (1988), requesting a certificate to provide transportation service to 25 existing customers during the peak and shoulder demand months, November through March. To implement the service, Transco proposed construction of 27.22 miles of gas pipeline looping and some related facilities. Transco calls this project "Southern Expansion" service.

The application grew out of extensive negotiations between Transco and the 25 customers, who needed access to additional natural gas during the winter months to accommodate "a steady addition of new temperature-sensitive customers". 51 FERC at 61,467. As in most agreements, each side bargained away some possible advantages in exchange for others it deemed more valuable. Transco, for example, allowed the customers "maximum flexibility and the opportunity to secure the most economically priced gas available." Id. at 61,468. The customers would be allowed to contract with third parties for the purchase of gas and upstream transportation with virtually no limitations.

For their part, the customers agreed that the initial rate for Southern Expansion service would utilize the straight fixed-variable ("SFV") rate design methodology. Under SFV, a pipeline recovers all of its fixed costs through a demand charge and all of its variable costs through a commodity charge. Pipeline customers pay the demand charge each month based on their entitlement to service, regardless of whether or how much they actually use it. They pay the commodity charge only for their actual shipments through the pipeline. Use of the SFV methodology guaranteed Transco recovery of all of its fixed costs, no matter how little its customers used the Southern Expansion service. Because all parties had agreed to SFV methodology, Transco's application incorporated throughput levels projected by the customers on the basis of SFV rates.

On May 14, 1990 FERC issued an order granting Transco certificate authority for the Southern Expansion service, subject to certain conditions and modifications. 51 FERC ¶ 61,173 (1990). Only one modification is relevant here: FERC rejected the application's proposed use of the SFV rates and substituted a modified fixed-variable ("MFV") rate design. Under MFV, a portion of the pipeline's fixed costs — return on equity and related income taxes — is included in the commodity charge, not the demand charge. The Commission explained its rejection only by observing that under the SFV methodology Transco would bear none of the risk of facility underutilization and that it had required MFV on a Transco expansion project three years earlier. Id. at 61,472. It noted, however, that it found MFV appropriate only as an initial rate and that in Transco's next rate filing any party or FERC staff could "propose new rates based on different methodologies, principles, or concepts". Id. at 61,473. FERC also recognized one factor that we think helps to explain the parties' contractual risk allocation: although the customers wanted the insurance provided by additional capacity, many were unsure whether or how much they would use the additional service, particularly in the near term. Id. at 61,470. If one viewed the service as an insurance policy for the customers, SFV would make perfect sense: persons who buy insurance pay the premium even if the risk insured against — here, an excess of demand over otherwise available supplies — fails to materialize.

Before proceeding further, we must explain an important detail of the MFV methodology and its application by the Commission here. Because a portion of the MFV commodity rate is designed to permit recovery of a fixed sum (the pipeline's return on equity plus related taxes), that portion of the rate is determined by dividing the fixed sum by the estimated throughput. Thus, if the fixed cost to be recovered were $1 million, and estimated throughput 2,000,000 units, the unit commodity rate would include 50 cents for that recovery. As a matter of simple arithmetic, the higher the estimated throughput, the lower the unit commodity rate. Here, in determining the commodity rate under the imposed MFV rate design, the Commission used the throughput levels estimated by the customers on the basis of the SFV methodology that they and Transco had agreed on. This was somewhat incongruous, for if there was any elasticity of demand, the customers would use less of the service at a high (MFV) commodity rate than at a low (SFV) rate.

Transco filed a timely request for rehearing, challenging FERC's imposition of MFV for its initial rate. The Commission on July 13, 1990 granted the request solely for purposes of securing more time for its consideration. On November 1, 1990, while the issue was still pending, Transco started service. Not knowing which rate structure FERC would ultimately adopt, customers presumably made decisions about whether (and how much) to use the service at least in part on the basis of their guesses as to whether SFV or MFV would more likely prevail. North Carolina Natural Gas decided against usage in the first season, while others decided in favor. See 64 FERC at 61,907.

On July 5, 1991 FERC granted rehearing on Transco's rate design claim and approved an "initial rate" based on SFV principles. 56 FERC ¶ 61,037 (1991). It found that "as a general matter, agreements among the pipeline and its customers regarding rate design should be given significant weight and implemented, absent a determination by the Commission that it would be unreasonable to do so. . . ." Id. at 61,136. It gave examples of factors that would render acceptance of the agreement unreasonable — mentioning specifically pipeline market power that could deny the customers "a meaningful choice" and an absence of arm's length bargaining — and found neither problem present here. Id.

Accordingly, Transco made a "compliance filing" that (if accepted) would implement the SFV rate "effective November 1, 1990", the start of service. Notice of the filing was duly published in the Federal Register, 56 Fed.Reg. 43,016 (1991), and FERC's Director of the Office of Pipeline and Producer Regulation accepted the filing, thereby establishing November 1, 1990 as the effective date.

On September 13, 1991 North Carolina Natural filed a protest to the compliance filing — a protest that was untimely under Commission rules as to the filing order and even more untimely vis-a-vis the July 5, 1991 decision itself. The petition attacked the November effective date, arguing that its adoption was tantamount to "retroactive ratemaking" that would

unfairly penalize Southern Expansion shippers for economic decisions not to utilize fully Transco's winter seasonal firm transportation service during the winter of 1990-1991 in light of the availability of lower-cost alternatives compared to the marginal cost of Southern Expansion service at the then-filed tariff rates.

57 FERC at 62,056.

On December 6, 1991 FERC issued a third order. 57 FERC ¶ 61,331 (1991). It said that its July 5 order had been silent as to the effective date of the SFV rates and that therefore North Carolina Natural did not have notice that Transco sought to have the change in rate design applied retroactively until Transco made its compliance filing. Id. at 62,057. On the merits, FERC concluded that North Carolina Natural's expectations deserved protection, so that the SFV rate should be made effective only as of its July 5, 1991 rehearing order. "North Carolina Natural could not have anticipated that the Commission would adopt the SFV rate methodology after the Commission employed the MFV methodology for so long." Id. In analyzing what purchasers might have anticipated, however, the Commission discussed neither the law nor Commission policy on approving rates agreed on by the parties, nor the actual expectations of any parties other than North Carolina Natural.

A barrage of responses ensued from Transco and various customers, resulting in a fourth FERC order, moving the effective date to November 1, 1991, the first anniversary of the start of service. 62 FERC ¶ 61,211 (1993). FERC recognized that, because of the seasonal nature of the Southern Expansion service, a mid-year change created at least a theoretical risk of Transco's enjoying a double recovery of its return on equity and related taxes. Id. at 62,513-14. In the winter season Transco had collected all of the commodity charges that could be collected for the first contract year — at an MFV rate that included equity and related taxes; yet with a change to SFV as of July 5, 1991, it would be able to collect for the second part of the year an SFV-based demand charge that also included those costs. Accordingly, FERC changed the effective date for SFV rates once again, to November 1, 1991, the beginning of the second contract year. This, FERC thought, would address all concerns; it specifically stated its assumption that "under a full contract year of MFV rates Transco should recover its first year of costs." Id.

Not surprisingly, Transco was not satisfied and sought rehearing. 64 FERC ¶ 61,099 (1993). It objected that applying a full year of MFV rates would have allowed it to recover its first year of costs only if throughput projections had proved accurate. But because of the incongruity that we have already noted — the use of an SFV-based throughput estimate to compute MFV commodity rates — aggregate customer purchases turned out to be substantially lower than the projections. (Of course Transco's challenge to the MFV methodology created uncertainty as to which commodity rate would ultimately be applied, so each customer's purchase decisions presumably reflected its estimate of the likely outcome of the dispute, or some average of the possible rates, weighted in accordance with their probabilities.) Given the customers' actual decisions on use of the service, Transco said that in fact it underrecovered its costs by $1.2 million. Id. at 61,907. FERC denied the petition, responding to Transco's arguments with a kind of loose estoppel theory, saying that Transco had put itself at risk by accepting the certificate, constructing the facilities, and beginning service under the Commission's then-current mandate of MFV rates.

Transco knew, when it accepted its certificate and decided to go forward with the project under an MFV rate design, prior to the Commission's ruling on its rehearing request, that it faced the risk of underrecovering a portion of its return on equity and related taxes if it did not meet its throughput. . . . [That] was a risk that Transco willingly assumed, we find, when it decided to go forward with the project.

Id. at 61,908. This appeal followed.

* * *

Transco argues as a threshold matter that the November 1, 1990 initial SFV rate should be reinstated because FERC's orders deferring the rate's effective date stemmed from a procedurally defective request for rehearing by North Carolina Natural. Because the North Carolina Natural request for relief from the compliance filing was not timely, whether it was construed as an attack on the filing itself or as a request for rehearing of the July 5 order, Transco characterizes it as a "collateral attack", which FERC normally rejects as such without reaching the merits. See El Paso Natural Gas Co., 34 FERC ¶ 61,411 at 61,973, order on reh'g, 36 FERC ¶ 61,022 (1986).

The Commission offers two procedural justifications for addressing North Carolina Natural's protest on the merits. First, it says that the July 5, 1991 order did not establish November 1, 1991 as the effective date of the SFV rate, so that North Carolina Natural was not barred from posing the challenge when the compliance order made it clear. See 57 FERC at 62,057. Second, it argues that it was free to amend its July 5, 1991 order throughout the period that timely petitions for rehearing were pending.

We are quite skeptical of the first theory. The July 5, 1991 order carried at least the natural interpretation that SFV rates would apply to the service from the outset, since it declared that the Commission was adopting SFV for the service's "initial rate", 56 FERC at 61,136, with no suggestion that the rate was not "initial" in every sense of the word. Thus, had the Commission later regarded the issue as decided, and rejected untimely attacks as "collateral", we would likely have upheld its position. In ANR Pipeline Co. v. FERC, 988 F.2d 1229 (D.C. Cir. 1993), for example, we sustained its rejection of an attack on later clarifying orders where "a reader schooled in gas pipeline regulation would have perceived a very substantial risk that [the Commission's earlier order] meant what it later proved to mean." Id. at 1234. Under those circumstances, in effect, a party holding back from demanding clarification does so at its own risk.

Under our precedents, however, the Commission was procedurally free to amend its order at the time it did. Section 19(a) of the Natural Gas Act provides that

Until the record in a proceeding shall have been filed in a court of appeals . . . the Commission may at any time, upon reasonable notice and in such manner as it shall deem proper, modify or set aside, in whole or in part, any finding or order made or issued by it under the provisions of this chapter.

15 U.S.C. § 717r(a) (1988). In Tennessee Gas Pipeline Co. v. FERC, 871 F.2d 1099 (D.C. Cir. 1989), we upheld the Commission's power to modify an order in response to an untimely "Motion for Clarification" filed at a time when timely petitions for rehearing were pending. In so doing we construed § 19(a) to allow the Commission to modify its orders — at anyone's behest — so long as the record had not yet been filed with a court of appeals and time to appeal had not yet expired. Id. at 1108. Accordingly, Transco's procedural attack on the Commission's shift in the effective date fails.

* * *

We thus turn to the merits. We must uphold FERC's orders unless they are "arbitrary, capricious, an abuse of discretion, or otherwise not in accordance with law." 5 U.S.C. § 706(2)(A) (1988). In assessing the orders, we consider the orders in light of the statute, evaluating whether FERC "considered the relevant factors and drew `a rational connection between the facts found and the choice made.'" Associated Gas Distributors v. FERC, 824 F.2d 981, 1016 (D.C. Cir. 1987) (citations omitted). The Commission must, of course, reveal the reasoning that underlies its conclusion. Greater Boston Television Corp. v. FCC, 444 F.2d 841, 851-52 (D.C. Cir. 1970).

We initially clear out of the way two possible red herrings. First, there is no dispute that the correct rate methodology for Southern Expansion service is SFV. The pipeline and the relevant customers — including North Carolina Natural itself — all agreed to SFV rates in advance, and the Commission nowhere suggests that its initial rejection of those rates was correct. Second, although the term "retroactive" is bandied about a good deal in this controversy, none of the possible outcomes involves the sort of "retroactive ratemaking" that is forbidden under the rule against retroactive ratemaking, a corollary of the filed rate doctrine. Transco's petition for rehearing of the initial MFV order put customers on notice that the SFV methodology might ultimately prevail, and such notice "changes what would be purely retroactive ratemaking into a functionally prospective process by placing the relevant audience on notice at the outset that the rates being promulgated are provisional only and subject to later revision." Public Utilities Comm'n of California v. FERC, 988 F.2d 154, 164 (D.C. Cir. 1993) (citations omitted). This permissible and relatively benign sort of retroactivity is in fact enshrined in § 4(e) of the Natural Gas Act, 15 U.S.C. § 717c(e), which explicitly allows FERC to make refunds when a filed rate increase has taken effect but the Commission later determines that a portion of the increase is not justified. See United Gas Pipe Line Co. v. Mobile Gas Service Corp., 350 U.S. 332, 341, 76 S.Ct. 373, 379, 100 L.Ed. 373 (1956) (§ 4(e) gives FERC power "to make its order retroactive, by means of the refund procedure, to the date the change became effective"). See also Towns of Concord, Norwood Wellesley v. FERC, 955 F.2d 67, 72 (D.C. Cir. 1992) (under provision of Federal Power Act paralleling § 4(e) of Natural Gas Act, the "Commission `may,' not must, order utilities to refund portions of newly-filed rates or charges later `found not justified.'") Given that Southern Expansion service started with the issue unresolved, the Commission could have applied any of the contesting rate methodologies to the period between the start of service and the final decision without violating the rule against retroactive ratemaking, even though the rate would have been retroactive in the broad sense of the word.

This takes us, then, to the heart of the Commission's stated rationale. First, it believed that adopting an effective date of November 1, 1990 would defeat the legitimate expectations of those customers — such as North Carolina Natural — that limited their use of the service because of the MFV rates on file between the commencement of service and July 5, 1991. But the Commission never explained what legitimated those expectations. The Commission never even considered the expectations of customers that used the service heavily, even though they may have done so in reliance on an expectation that ultimately FERC would — as it did — honor the agreement between Transco and its customers.

Indeed, we find no firm ground on which North Carolina Natural could confidently indulge an expectation that the Commission's ultimate resolution of the SFV/MFV issue would apply only prospectively. The only two FERC decisions called to our attention that seem to bear on the issue, Black Marlin, 48 FERC ¶ 61,024 (1989), and KN Energy, Inc., 50 FERC ¶ 61,290 (1990), point the other way. Both assert that the Commission has authority to issue a § 7 certificate when it has not yet resolved the validity of proposed rates, accepting those rates conditionally — subject to refund in the event it later finds them unreasonable. Both decisions exercise that authority. See also Public Service Comm'n of New York v. Federal Power Comm'n, 329 F.2d 242, 249 (D.C. Cir. 1964) (upholding Commission power to award refunds when it has issued a temporary certificate conditionally accepting proposed rates, but remanding for further consideration of refund issue where Commission had denied refunds solely because order issuing temporary certificate had not mentioned refund possibility). The norm seemingly represented by these FERC decisions, which came out shortly before FERC issued the certificate here, is that where service starts under § 7 before final determination of the rates, the rate finally determined will be applied retroactively to the start of service. That norm cuts powerfully for finding the rates effective as of November 1, 1990, and against any expectations to the contrary. The norm makes a good deal of sense, as it means that the "right rate", i.e., whatever rate the Commission lawfully determines to be right, is applied throughout the period despite the Commission's initial uncertainty and delay. The expectations of those who act in anticipation of the right rate are protected, and they would seem presumptively the most deserving.

While the final Commission decision did not explicitly deny that some equities lay with Transco, it evidently found them neutralized by Transco's having initiated service before it was under any compulsion to do so. We can detect no real force in this analysis. In the first place, it does not provide any basis for limiting Transco in the first year of service to a set of rates that were not only internally inconsistent — framed as MFV but based on throughput estimates derived from the parties' unanimous agreement on SFV — but also incapable of allowing full cost recovery unless some deus ex machina appeared on the scene to increase throughput beyond what anyone might reasonably have anticipated.

Because of the Commission's initial decision, the vulnerability of its analysis (confirmed by the Commission's own recantation, but inherent in the decision itself), and FERC's delay in deciding the rehearing, every party's action or inaction involved some risk. North Carolina Natural, like each of the other customers, adopted a strategy that it regarded as an optimal accommodation of the competing risks. So, too, did Transco. Had Transco elected complete inaction, for example, it would certainly have delayed the start of recovery of its investment and perhaps have lost forever the return on that investment for the period of its paralysis. The Commission offers no reason why North Carolina Natural's bets should be protected, but Transco's not. The answer can hardly be that Transco was "active" and North Carolina Natural "passive". The action that Transco undertook was no more than what the customers had sought and Transco had agreed to provide — an insurance policy that would guarantee the customers' ability to meet growing peak-period demand.

Because there may be some legitimate reason for the Commission to apply MFV rather than SFV rate design for all or part of the initial year of service — though none has yet been disclosed to us — we remand the case to the Commission for further proceedings consistent with this opinion. See Public Service Comm'n of New York v. Federal Power Comm'n, 329 F.2d at 248-50 (similar remand). If the exercise of reasoned decision-making leads the Commission to restore November 1, 1990 as the effective date for SFV principles, that will be the end of it. If reasoned decisionmaking leads the Commission to adhere to November 1, 1991 as the effective date, then the Commission will still have to remove the incoherence in the MFV rates, substituting appropriate throughput estimates for the ones it actually used.

So ordered.