Opinion for the Court filed by Circuit Judge TATEL.

TATEL, Circuit Judge.

Petitioner, an owner and operator of electric generation facilities in New York City, appeals an order of the Federal Energy Regulatory Commission, arguing that the Commission acted arbitrarily and capriciously in deciding that the New York Independent System Operator (NYISO) had not violated the filed rate doctrine in establishing a pair of formulae for translating “installed capacity” into “unforced capacity.” Because we agree, we vacate and remand.

The New York Independent System Operator is a not-for-profit organization charged with administering New York’s electricity markets. Among other duties, NYISO enforces rules designed to ensure the reliability of the state’s electricity grid. One way NYISO does this is by requiring electricity retailers (referred to as ‘load-serving entities’ or LSEs) to purchase capacity — as opposed to actual electricity— from generators in regularly held auctions. The purchase of capacity differs from the purchase of electricity. Like an option contract, the LSE compensates the generator for the option of buying a specified quantity of power irrespective of whether it ultimately buys the electricity.

NYISO calculates the amount of capacity LSEs must purchase by multiplying each LSE’s expected peak load by one plus a figure called the installed reserve margin, which is set by the New York State Reliability Council (NYSRC), a non-profit corporation established by NYISO. During the summer of 2002, the period at issue here, NYSRC set the installed reserve margin at 18%. For instance, if an LSE expected that the most power it would need — its peak load — would be 1,000 MW, then that LSE would have to acquire access to 1,180 MW of capacity in order to maintain a cushion against shortfall. NYSRC sets the installed reserve margin at a level necessary to prevent system *807 failure from occurring (probabilistically) more than once every ten years.

Until 2001, NYISO denominated the LSEs’ capacity obligation in terms of “installed capacity,” i.e., the amount of power a generating facility can produce under ideal conditions. This approach had a significant shortcoming: it gave generators little incentive to improve their reliability, leaving them free to sell their entire quantity of installed capacity regardless of how often their facilities underwent “forced” outages — that is, unplanned outages such as mechanical breakdowns and the like. NYISO sought to correct this problem by requiring LSEs to purchase “unforced capacity” instead of installed capacity. For those unfortunate souls not steeped in the vernacular of electricity regulation, a generator’s unforced capacity (UCAP) is its installed capacity (ICAP) discounted or “de-rated” by its forced outage rate (or equivalent forced outage rate demand (EFORd)). The forced outage rate equals the historical percentage of the generator’s maximum output lost to forced outages when such output is demanded. The translation of installed into unforced capacity can be represented mathematically as follows:

UCAP = ICAP x (1 — EFORd)

For example, consider a generating facility with installed capacity of 1,000 MW that has historically lost 10% of its output to forced outages. That generator would have unforced capacity of 900 MW. If it could have sold 1,000 MW in the capacity auctions before the switch from installed to unforced capacity, then after the switch it could sell only 900 MW, giving it an incentive to reduce its forced outage rate.

Having reduced the quantity of capacity generators could sell, NYISO also reduced the quantity of capacity LSEs had to purchase in order to reflect that they were purchasing unforced rather than installed capacity. Critical to the issue in this case, NYISO used different methodologies for making the two translations. For translating generators’ installed capacity into unforced capacity, NYISO directed — in revisions to its Services Tariff filed in July 2001 — that the forced outage rate would be calculated from the past twelve months of outage data for each generator. Nothing in the tariffs, however, specified how the LSEs’ capacity obligation would be translated, only that it would be done “in accordance with the ISO procedures.” J.A. 487. NYISO later filled this gap in the “ICAP Manual,” a document that specified various aspects of the new system and that had been prepared as part of a months-long stakeholder process. The ICAP Manual explained that for purposes of translating installed capacity into unforced capacity for LSEs, the forced outage rate would be “based on the data used to determine the Installed Reserve Margin by the NYSRC,” J.A. 485, a formulation understood to mean that NYISO would discount the LSEs’ installed capacity requirement by the average forced outage rates generators had experienced over the past ten years.

Because forced outages had declined significantly over the ten years leading up to 2002, measuring forced outage rates over two different time periods produced substantially different results: the average generator’s 12-month forced outage rate was roughly 5%, while the 10-year outage rate used by the LSEs was roughly 10%. Using a higher forced outage rate for LSEs than for generators effectively reduced the amount of capacity LSEs had to purchase, regardless of whether capacity is measured in terms of installed or unforced capacity. To see why, imagine a market that, for ease of illustration, has LSEs obligated collectively to buy 10,000 MW of installed capacity and generators collec *808 tively possessing 10,000 MW of installed capacity. In this simple scenario, supply and demand are in balance. If, however, installed capacity is translated into unforced capacity using a 10% outage rate for buyers and a 5% outage rate for sellers, the balance shifts. Under such a scenario, LSEs must purchase 9,000 MW of unforced capacity, but generators can sell 9,500 MW of unforced capacity, creating a capacity surplus of 500 MW or about 5% of total supply.

In other words, using a lower forced outage rate for generators than for LSEs decreases the quantity of capacity sold in the market. This decrease can be measured by translating the quantity of unforced capacity sold in the above hypothetical back into installed capacity. Rearranging the equation produced above yields a formula for doing so:

(1) UCAP = ICAP x (1 — EFORd)

(2) ICAP = UCAP

(1 — EFORd)

The generators in this example sold 9,000 MW of unforced capacity and were de-rated using a 5% forced outage rate. Plugging those numbers into the above equation yields 9,474 MW of installed capacity sold, substantially less than the 10,-000 MW sold before the application of differential forced outage rates.

Following the summer of 2002, the NYSRC and NYISO staffs produced calculations using the same logic as above showing that the use of different translation methodologies for generators and LSEs had reduced the quantity of installed capacity sold from 35,960 MW, the amount required by the NYSRC formula, to 34,189 MW, a 1,771 MW reduction. As a result, the NYSRC and NYISO staffs concluded, the actual statewide installed reserve margin had fallen from the NYSRC-specified level of 18% to 12.2%. NYISO’s Business Issues Committee later wrote that “the unforeseen and unintended consequence of the current translation methodology is a shortfall of Installed Capacity and therefore an inability to comply fully with the New York State Reliability Council’s Reliability Rules.” J.A. 115. Accordingly, in the fall of 2002, NYISO switched to a system in. which both generators and LSEs measure forced outage rates over 12-month periods.

Petitioner Keyspan-Ravenswood (Rav-enswood) generates electric power and sells capacity in New York City. In 2004, Ravenswood filed a complaint with the Commission claiming that NYISO had violated Section 205 of the Federal Power Act (FPA), which requires public utilities to file with the Commission all “practices” and “regulations” affecting their “rates and charges” and mandates that public utilities may only depart from such filed rates “after sixty days’ notice to the Commission and to the public.” 16 U.S.C. § 824d(c), (d). Ravenswood argued that in three separate filings NYISO had obligated itself to enforce NYSRC’s installed capacity requirements, yet by using a lower forced outage rate for generators than for LSEs in the summer of 2002, NYISO had fallen well short of NYSRC’s installed capacity requirement, thus violating the FPA. Ravenswood presented economic evidence purporting to show that, by both reducing the quantity of capacity sold and, through the law of supply and demand, reducing the market price of capacity, NY-ISO had caused it to lose $23.3 million in sales, which it asked the Commission to order in refunds.

Rejecting Ravenswood’s complaint, the Commission stated that “the rates charged by NYISO ... conformed with the Commission’s applicable orders governing NY-ISO’s ICAP and UCAP requirements, and were consistent with NYISO’s then-effec *809 tive tariffs, rate schedules and manuals.” Keyspan-Ravenswood, LLC v. N.Y. Indep. Sys. Operator, 110 F.E.R.C. ¶ 61,116 at 61,471 (2005) (“Complaint Order”). The Commission explained that it had “approved” the methodologies employed by the Commission in an earlier set of orders referred to as the “UCAP Orders,” which rejected Ravenswood’s challenge to NYI-SO’s 2001 filing that effected the overall change from installed capacity to unforced capacity. Id. at 61,475 (citing N.Y. Indep. Sys. Operator, 96 F.E.R.C. ¶ 61,251 (2001), order on reh’g, 98 F.E.R.C. ¶ 61,180 (2002), vacated sub nom. KeySpan-Ravenswood, LLC v. FERC, 348 F.3d 1053 (D.C.Cir.2003), on remand sub nom. N.Y. Indep. Sys. Operator, 108 F.E.R.C. ¶ 61,309 (2004)). The Commission also appeared to rely on the fact that the ICAP Manual required NYISO to use different forced outage rates for LSEs and for generators. Complaint Order at 61,475.

Ravenswood petitioned for rehearing, arguing that the UCAP Orders were irrelevant. Those orders, Ravenswood claimed, concerned another aspect of NYI-SO conversion from installed to unforced capacity and had nothing to do with NYI-SO’s decision to use different forced outage rates for LSEs and generators. Responding to the Commission’s reliance on the ICAP Manual, Ravenswood argued that the manual could not possibly cure a violation of the filed rate doctrine because NYISO had never filed it with the Commission.

The Commission denied rehearing, basing its decision on slightly different reasons than it had given in the Complaint Order. Keyspan-Ravenswood v. N.Y. Indep. Sys. Operator, 111 F.E.R.C. ¶ 61,336 (2005) (“Rehearing Order”). No longer insisting that the UCAP Orders had “approved” NYISO’s translation methodology, the Commission stated only that those orders “did not prescribe the use of Ravens-wood’s recommended particular methodology.” Rehearing Order at 62,488. The Commission also de-emphasized the importance of the ICAP Manual, explaining that it had not “solely rel[ied] on the ICAP Manual as a basis for its conclusion,” but rather that “the Commission primarily relied on the fact that what NYISO did was not only consistent with the ICAP Manual, but also was consistent with Commission orders and was not inconsistent with any requirement in NYISO’s tariffs.” Id. at 62,487. Finally, introducing an alternative ground for its decision, the Commission asserted that even had NYISO violated its tariff obligations, the Commission would have ordered no refunds because Ravens-wood failed to prove its injury with sufficient “clarity.” Id. at 62,488.

Once again, Ravenswood petitioned for rehearing, challenging the Commission’s new rationale that the company had failed to prove its injury. Without reaching the merits, the Commission denied rehearing, resting on the proposition, unchallenged here, that “[t]he Commission does not allow rehearing of an order denying rehearing.” Key span-Ravenswood, LLC v. N.Y. Indep. Sys. Operator, 112 F.E.R.C. ¶ 61,-153 at 61,885 (2005).

Ravenswood petitions for review, claiming that the Commission failed to confront its argument that NYISO’s use of two different translation methodologies violated NYISO’s filed commitment to enforce NYSRC’s installed capacity requirements. We review the Commission’s decisions under the arbitrary and capricious standard, affirming if the Commission has articulated a “rational connection between the facts found and the choice made.” Motor Vehicle Mfrs. Ass’n v. State Farm Mut. Auto. Ins. Co., 463 U.S. 29, 43, 103 S.Ct. 2856, 77 L.Ed.2d 443 (1983) (internal quotation marks omitted).

*810 II.

As a public utility regulated under the FPA, NYISO may only change its rates or “practices ... affecting such rates” by first filing those rates with the Commission. 16 U.S.C. § 824d(c). The Commission’s regulations require that “[e]very public utility shall file with the Commission ... full and complete rate schedules ... clearly and specifically setting forth all rates and charges ... [and the] practices, rules and regulations affecting such rates and charges.” 18 C.F.R. § 35.1(a).

The Commission does not dispute that, despite the fact that NYISO switched the requirement it imposed on LSEs and generators from installed to unforced capacity, NYISO nonetheless had a filed obligation to enforce NYSRC’s installed capacity rules which required 35,960 MW of installed capacity in New York State during the summer of 2002. Complaint Order at 61,-473 (“The three Commission-approved rate schedules applicable to NYISO required NYISO to enforce ICAP requirements for 2002 for both statewide and In-City generation.”); see also Agreement Between the New York Independent System Operator and New York State Reliability Council § 3.4 at Pet’r’s Br. Attach, at C4 (“The ISO shall require LSEs ... to maintain appropriate levels of Installed Capacity consistent with the Reliability Rules.”). Nor does the Commission dispute the analysis, explained supra, that the use of different forced outage rates for generators and for LSEs effectively reduced the quantity of installed capacity purchased during the summer of 2002 from 35,960 MW to 34,189 MW. Given these two undisputed facts, and given that NYISO never filed its translation methodology with the Commission, we have no trouble concluding that the Commission acted arbitrarily and capriciously in ruling that NYISO had not violated the filed rate doctrine. The Commission and the intervenors advance several additional arguments, all unpersuasive.

The Commission’s argument that “nothing in the NYSRC’s Reliability Rules ... [or] the tariffs themselves expressly spell out the appropriate methodology” misses the point. Rehearing Order at 62,487. Ravenswood has never argued that NYISO must adopt any particular method for translating installed capacity into unforced capacity, only that NYISO may not adopt a method that causes it to violate its filed commitment to enforce NYSRC’s installed capacity requirements.

The Commission also insists that it properly relied on the ICAP Manual because the Manual had been “incorporated by reference in the Service Tariff.” Resp’t’s Br. 21; see also Complaint Order at 61,474. In support, the Commission points to a passage in the 2001 Services Tariff stating that NYISO shall translate the installed capacity requirement into an unforced capacity requirement “in accordance with the ISO Procedures.” J.A. 487. We fail to see, and the Commission fails to explain, how the unelaborated reference to “ISO Procedures” can be understood as “clear[] and specificf]” notice that NYISO intended to follow the translation methodology in the yet-to-be-adopted ICAP Manual. Moreover, as Ravenswood points out, the ICAP Manual nowhere states expressly that the forced outage rate used by LSEs will be measured over ten years. Rather, the Manual cryptically says that the translation would be “based on the data used to determine the Installed Reserve Margin by the NYSRC,” J.A. 485, thus placing still more distance between the vague reference to “ISO Procedures” and the regulatory obligation that NYISO “clearly and specifically” set forth all “practices, rules and regulations affect *811 ing [its] rates and charges.” 18 C.F.R. § 35.1(a).

Next, the Commission suggests that NYISO had no need to file its method for translating installed capacity into unforced capacity because requiring such detail in a filing goes beyond the “rule of reason.” In City of Cleveland v. FERC, 773 F.2d 1368, 1376 (D.C.Cir.1985), upon which the Commission relies, we held that utilities must file “only those practices that affect rates and service significantly, that are realistically susceptible of specification, and that are not so generally understood in any contractual arrangement as to render recitation superfluous.” A cursory review of the facts of this case shows why City of Cleveland’s exception to the filing requirement is inapplicable. Analysis by NYI-SO’s own staff demonstrates that by reducing the installed reserve margin by nearly a third from 18% to 12.2% NYISO’s translation methodology significantly affected its compliance with the Reliability Rules. Just as clearly the translation methodology is “susceptible of specification” — the rule that forced outage rates be measured over a one-year period for generators and a ten-year period for LSEs is, to say the least, easily reduced to writing.

Appearing as intervenors, NYISO and a coalition of LSEs argue that the Commission approved NYISO’s translation methodology in the UCAP Orders, a rationale that the Commission appeared to advance in the Complaint Order. As the Commission concedes in its brief, however, the issue of how NYISO should translate quantities of installed capacity into quantities of unforced capacity was never raised in those orders. Rather, those orders concerned a different aspect of the conversion from installed to unforced capacity. In the capacity market, NYISO enforces a price cap designed to prevent abuse of market power among capacity sellers. Because the capacity sellers had, by definition, less unforced capacity to sell than they had installed capacity, NYISO needed to raise the price cap to account for that reduction. Concerning only the method for adjusting that price cap, the UCAP Orders have no direct bearing on the issue in this case, i.e., the translation of installed capacity quantities into unforced capacity quantities. See generally The UCAP Orders, supra.

The Commission and the interve-nors also suggest that we excuse NYISO’s violation of the filed rate doctrine because Ravenswood participated in the stakeholder process that developed the ICAP Manual. Ravenswood’s participation in the development of the ICAP Manual, the character of which is not entirely clear from the record, may have relevance to the Commission’s refund inquiry. See Towns of Concord v. FERC, 955 F.2d 67, 75 (D.C.Cir.1992) (describing refunds as “a form of equitable relief’). But the Commission provides no reason to believe that Ravenswood’s participation in such a stakeholder process relieved NYISO of its statutory obligation to file significant changes to its rates and “practices ... affecting such rates.” In any event, the Commission clearly stated in the Compliance Order that it reached its decision without regard to the argument that Rav-enswood “sidestepp[ed] NYISO’s stakeholder process.” Complaint Order at 61,-475.

III.

Ravenswood also challenges the Commission’s alternative ground for its decision, articulated for the first time in the Rehearing Order, that even had NYISO violated the filed rate doctrine, it would deny Ravenswood the refunds it seeks. The Commission’s explanation consisted, in its entirety, of the following:

*812 [H]ad NYISO actually used the ICAP to UCAP translation supported by Ravens-wood, it still remains unclear what prices would actually have been paid by LSEs because of the nature of the translation and the auction. This lack of clarity, and no instances of reliability problems arising from capacity shortages during the Summer 2002 Capability Period, leads the Commission to conclude that, even if we agreed with Rav-enswood that NYISO’s actions violated its tariffs, Ravenswood still would not have met its burden to show that it was entitled to any refunds, let alone the $23.3 million in refunds that it requested.

Rehearing Order at 62,488. This brief passage contains three shortcomings that leave us unable to “discern a reasoned path to the decision the Commission reached.” East Tex. Elec. Coop., Inc. v. FERC, 218 F.3d 750 (D.C.Cir.2000) (internal quotation marks and alterations omitted). First, the Commission refers only to its uncertainty regarding what the price of capacity would have been had NYISO used uniform methods for translating installed capacity into unforced capacity. But Rav-enswood also claims to be injured by the reduction in quantity of capacity sold, a figure precisely calculated by NYISO’s own staff. Specifically, in its economic analysis presented to the Commission, Ravenswood claimed that $8.1 million of its losses arose from reduction in the price of capacity and that the remaining $15.2 million arose purely as a result of lost sales, irrespective of changes in price. By focusing exclusively on uncertainty relating to price, the Commission’s explanation for denying the entire $23.3 million in refunds remains incomplete.

Second, as to the uncertainty regarding the effect of NYISO’s translation methodology on the price of capacity, the Commission offered no reasons for rejecting Ravenswood’s extensive economic analysis. We will defer to the Commission’s judgment in technical matters within its expertise, but only when the Commission has in fact exercised its judgment. See, e.g., Pub. Citizen Health Research Group v. Tyson, 796 F.2d 1479, 1505 (D.C.Cir.1986) (“While we acknowledge our deference to the agency’s expertise in most cases, we cannot defer when the agency simply has not exercised its expertise.”). The Commission’s conclusory statement that it “remains unclear what prices would actually have been paid by LSEs” fails to clear this diminutive hurdle.

Finally, the Commission’s fleeting reference to the absence of reliability problems during the summer of 2002 requires further explanátion. Although this clause comes in the middle of a sentence concluding that Ravenswood failed to meet its burden to demonstrate its entitlement to refunds, the clause appears to speak to a different concern. Namely, the Commission seemed to be implying a policy judgment that violations of the Reliability Rules should be remedied only when reliability problems actually arise. It is true that the Commission has considerable discretion to deny refunds for reasons of either policy or equity. See Towns of Concord, 955 F.2d at 75-76. But it is also true that “when deciding whether to order refunds, FERC must provide a reasonable explanation for its decision: it must show that it has considered relevant factors and struck a reasonable accommodation among them.” Consol. Edison Co. v. FERC, 347 F.3d 964, 972 (D.C.Cir.2003) (internal quotation marks and alteration omitted). Here, the Commission’s brief statement raises some fairly obvious concerns. To name just one, Ravenswood argued both in its initial complaint and on appeal that an important reason for the absence of relia *813 bility problems is that NYISO ordered it to produce electricity at several points during that summer — a fact that appears undisputed in the record. Thus, to rely upon the absence of reliability problems as a reason for denying Ravenswood refunds seems — without further explanation — inequitable if not Kafkaesque. The Commission will have the opportunity to address this and other questions on remand.

IV.

The petition for review is granted and this case is remanded to the Commission for further proceedings consistent with this opinion.

So ordered.