Power Generation Reliability Benchmarking: KPIs, Peer Groups, and Accreditation Methods

Most generating units record the operating data needed to benchmark reliability in the DCS and PI Historian. The quarterly NERC Generator Availability Data System (GADS) file contains the submitted version. Reliability analysts turn those records into standardized indices and compare them against a valid peer group. Operations managers then direct action toward the measured gap. This guide serves generation operations managers who own outage records and NERC compliance officers who own submissions. It also addresses the IPP or cooperative executive who answers for the capacity revenue that now depends on market-specific accreditation rules.

What is power generation reliability benchmarking

Power generation reliability benchmarking compares a unit’s outage and derating indices, plus its availability indices, against the same indices for a population of similar units, computed the same way. The definitions come from IEEE Std 762-2023. The 2026 NERC GADS Data Reporting Instructions state that the IEEE 762 definitions “were incorporated” into GADS. Two numbers that share a name but use different formulas cannot form a benchmark.

IEEE 762 fixes the asset boundary. IEEE 762-2023 defines a generating unit as all equipment up to the high-voltage terminal of the generator step-up transformer. Everything past that terminal follows different measurement rules:

• Generation reliability: IEEE 762 and NERC GADS measure period hours and equivalent capacity loss, along with net generation. Key indices are AF, EAF, FOR, EFOR, EFORd, and NCF.
• Distribution reliability: IEEE 1366-2022 measures customer interruptions on “distribution systems, substations, circuits, and defined regions.” Key indices are SAIFI, SAIDI, and CAIDI.

GADS operates separate conventional and renewable programs. The renewable programs cover wind and solar. Standalone battery storage sits outside those programs. The 2026 State of Reliability overview states that “standalone energy storage is not currently collected by NERC under a Section 1600 data request,” and that NERC’s Performance Analysis team is still evaluating BESS data collection.

Why reliability benchmarking matters

Capacity accreditation now depends on outage or performance data under market-specific rules. A unit that does not know where it sits against its peers is negotiating its accreditation blind.

• Capacity accreditation: NYISO derates conventional generators by their own outage history. Its accreditation documentation sets UCAP = Adjusted ICAP × (1 − AEFORd), where AEFORd is the average of the two previous like-Capability-Period EFORds. PJM used EFORd from five years of eGADS data through the 2024/2025 Delivery Year. From 2025/2026, its agreement revisions set Accredited UCAP = Effective Nameplate Capacity × ELCC Class Rating × ELCC Resource Performance Adjustment. MISO’s Resource Accreditation White Paper allocates class-level UCAP to individual units using Tier 1 and Tier 2 availability, weighted 20% and 80%.
• Performance penalties: PJM’s Manual 18 revisions define the Non-Performance Charge as Performance Shortfall × a rate built from LDA Net CONE. Winter Storm Elliott produced $1.8 billion in such charges before a FERC-approved settlement in Docket ER23-2975 cut them 31.7% to $1.2 billion, per PJM’s settlement FAQ and Utility Dive reporting. ISO-NE’s Forward Capacity Market Capacity Performance Payment Rate is $9,337/MWh for 2025-2026 and thereafter, up from $5,455/MWh.
• Regulatory exposure: NERC requires GADS reporting under Rules of Procedure Section 1600. The 2026 DRI requires event and performance data at OATI’s webE-GADS within 45 days after the end of every calendar quarter. A miscoded event that reaches NERC creates a data-quality problem the compliance officer owns.
• Peer positioning and corrective action: NERC publishes only fleet averages, so a plant that wants to know its quartile must build the comparison itself. GADS carries 1,700+ cause codes, which provide the granularity needed to rank failure modes for capital review.

For an IPP or co-op executive with a two-person GADS team, the accreditation formulas above are the business case. In NYISO, submitted EFORd directly affects UCAP. PJM and MISO now use different performance-adjusted accreditation methods, so each plant must map its submitted outage record to the rules of its own market.

Core reliability KPIs every plant should track

NERC divides its indices into factors and rates. The difference decides which metric belongs in which conversation. NERC’s GADS training on factors and rates states that factors use period hours and are additive, with WAF + WFOF + WMOF + WPOF = 100%. Rates use service hours and are not additive.

EFOR, EFORd, and FOR

FOR ignores deratings. EFOR counts them. EFORd counts them only when the unit was needed. NERC Appendix F provides the formulas:

• Forced Outage Rate (FOR): FOR = FOH ÷ (FOH + SH) × 100, where FOH is forced outage hours and SH is service hours, according to the 2024 Appendix F. A unit that runs at half load for a week with a tube leak posts the same FOR as one that ran flawlessly.
• Equivalent Forced Outage Rate (EFOR): The 2026 Appendix F, Index No. 24, gives EFOR = (FOH + EFDH) ÷ (FOH + SH + RSH + EFDHRS). EFDH is equivalent forced derating hours, RSH is reserve shutdown hours, and EFDHRS is forced derating hours during reserve shutdowns. A derating enters as the fraction of net maximum capacity lost multiplied by the hours it lasted, so the half-load week above becomes 84 equivalent full-outage hours.
• Equivalent Forced Outage Rate demand (EFORd): Index No. 26 gives EFORd = (FOHd + EFDHd) ÷ (FOHd + SH), where the “d” terms count only forced outage and derating hours during demand periods. NERC pools EFORd using IEEE 762 Annex F Method 2, calculating from reported totals so that small groups avoid divide-by-zero errors.

EFORd is the accreditation metric in NYISO and was PJM’s derating input through 2024/2025. Plant engineering teams use EFOR to size the derating problem because the demand condition in EFORd removes outages that happened while the unit sat in reserve.

Availability factor, EAF, and capacity factor

Availability factor counts hours the unit could have run. Equivalent availability factor discounts those hours for capacity the unit could not deliver. The 2024 Appendix F defines both:

• Availability Factor (AF): AF = AH ÷ PH × 100, where AH is available hours and PH is period hours.
• Equivalent Availability Factor (EAF): EAF = (PH − POH − MOH − FOH − EUDH) ÷ PH × 100, where POH is planned outage hours, MOH is maintenance outage hours, and EUDH is equivalent unplanned derated hours. A unit available 8,000 of 8,760 hours but derated 20% for half of them shows AF of 91.3% and a materially lower EAF.

Capacity factor answers a different question: how much energy the unit produced against what it could have produced at full capacity for the whole period. IEEE 762-2023 separates availability, the fraction of time capable of service, from productivity, or actual versus potential production. Capacity factor is a productivity index. The denominator matters. GADS requires Net Maximum Capacity, which is gross maximum capacity less station service, while gross figures are voluntary. A capacity factor computed against nameplate will read lower than the same unit’s GADS NCF, and the two should not appear on the same chart.

MTBF, MTTR, and planned maintenance percentage

NERC does not publish a fleet MTBF. Its equivalent index is Mean Service Time to Forced Outage. The 2025 Appendix F defines MSTFO = (PH − UOH) ÷ (number of forced outage occurrences), where UOH is unavailable operating hours. NERC’s licensed pc-GAR MT module computes MTBF and mean time to repair as user outputs from the GADS database rather than publishing them as benchmarks.

Mean time to repair for forced events is forced outage hours divided by forced outage occurrences. Two units with identical FOR can have opposite maintenance problems: one trips often and recovers fast, while the other trips rarely and stays down for weeks. Only the MTBF and MTTR pair separates them.

Planned maintenance percentage is not an Appendix F index. It is a maintenance-program ratio built from GADS state codes. Compute it as planned plus maintenance outage hours divided by total outage hours (POH + MOH + FOH). The link to forced outage rate is arithmetic: for a fixed total of outage hours, every hour that moves from the forced column to the planned column lowers FOR. The 2026 DRI sets the boundary through its definition of U3, which requires removal before the end of the next weekend, versus MO, which allows deferral beyond the next weekend but before the next planned outage. Coding a U3 as an MO raises PMP and lowers EFOR without a single wrench being turned.

Top-quartile benchmark targets

NERC publishes fleet averages rather than quartiles. The State of Reliability reports give a capacity-weighted WEFOR by unit type, and those figures provide the authoritative public anchor for “average” performance:

Sources: NERC 2026 State of Reliability overview and 2025 State of Reliability technical assessment.

WEFOR is capacity-weighted and EFORd is demand-conditioned, so a unit’s EFORd cannot be placed on this table without adjustment. Top-quartile cutoffs by technology exist, but only inside NERC’s licensed pc-GAR software, which supports “distributions” and “benchmarking” outputs that NERC does not post publicly. The most recent public five-year Generating Unit Statistical Brochure covers 2018–2022, posted August 29, 2023. The brochure landing page currently holds no downloadable numeric values.

Consultants and vendors cite figures such as an MTBF above 800 hours or a planned maintenance percentage of 85% or higher. None of the NERC GADS, IEEE 762, EPRI, or ORAP documents reviewed for this guide states either threshold. Treat them as rules of thumb. They do not qualify as standards or industry targets for a board deck.

Where published targets do exist, they are program-specific:

• Nuclear: WANO’s 2025 targets set a 36-month Forced Loss Rate of 2.0% for the industry and 5.0% for an individual unit. IAEA PRIS reports a world Unplanned Capability Loss Factor of 3.9% in 2023, 3.2% in 2024, and 3.1% in 2025 across roughly 400 reactors. Neither metric equals GADS EFORd.
• F-class gas turbines: A 2010 Strategic Power Systems paper drawn from ORAP reports a 2009 F-class fleet (n > 200) availability factor of 92.2% and forced outage factor of 1.2%.
• Small gas turbines: A DOE generation study of 121 units over 2000–2002 found MTBFO of 3,604.62 hours for 20–100 MW gas turbines (n=9) at 93.53% availability, and 5,317.73 hours for steam turbines under 25 MW (n=25).

The practical target is the 25th-percentile EFORd of a peer group you define yourself in pc-GAR, computed under Appendix F Method 2. That calculation uses the selected peer population and the same GADS methodology. A borrowed 800-hour MTBF does not.

How to categorize lost generation

The correct lost-generation categories depend on the asset class. Conventional units separate forced outages, planned or maintenance outages, deratings, and reserve shutdown. Wind and solar also track resource-unavailable conditions or curtailment under their own GADS rules. The 2026 DRI defines the conventional state codes:

Deratings are the category plants most often lose. The GADS event record requires Net Available Capacity for each event, and that field converts a partial reduction into EFDH for forced deratings or EUDH for unplanned deratings. A plant that logs full outages diligently but records deratings as operator notes in the shift log will submit an EFOR that is closer to its FOR than reality.

Wind and solar carry additional categories that must stay out of the forced bucket. The 2025 GADS Wind DRI records Resource-Unavailable Turbine-Hours for “environmental conditions that are outside the operating specifications of the wind turbine.” The 2025 GADS Solar DRI states that “all curtailments are considered reserve shut down hours.” Both DRIs trigger a reportable event only at a loss of at least 20 MW of plant total installed capacity, with MWHLoss as a required field.

Once the plant team assigns every event to the correct category and calculates lost MWh, the reliability analyst sorts the events. Multiply each event’s equivalent outage hours by net maximum capacity, group the results by cause code, then rank them in descending order. The plant team uses that ranked list to allocate its corrective-action budget. Run the same sort within the derating category alone; a chronic 10% derate from a fouled condenser can outrank every trip on the unit.

Peer group selection for valid benchmarks

A peer group is valid when the units in it fail for comparable reasons at comparable rates. Three normalization axes do most of the work:

• Unit type and fuel: NERC’s GADS brochures group by unit type and size class. ORAP groups by OEM and model, then adds frame class, generation, application, and duty cycle. A combined-cycle block and a peaking combustion turbine share a fuel and almost nothing else.
• Unit size: GADS size classes exist because component populations and maintenance intervals scale with rating.
• Operating profile: The World Energy Council study excluded “units whose utilisation factor is less than 40%” as a comparability filter. A unit that cycles daily should not be measured against baseload peers.

On group size, the standards are silent by design. IEEE Std 762-2023 states: “No minimum sample size is specified in this document.” NERC’s 2025 Appendix F notes that “by allowing sums, smaller groups of units can be used to calculate EFORd without experiencing the divide by zero problem,” with no floor on unit count.

The 30-unit minimum that appears in benchmarking guides comes from introductory statistics rather than NERC, IEEE, EPRI, or ORAP. It is the central-limit-theorem heuristic for treating sample means as roughly normal, and it fails for rare events. A peer-reviewed critique finds that for a Poisson variable with a low rate such as λ = 0.02, “the sample size n may need to be in the hundreds or even thousands before the convergence to the normal distribution becomes noticeable.” Forced outages are exactly that kind of variable. Agresti recommends at least 15 observed outcomes of each type before trusting a proportion interval. Hesterberg calls n ≥ 30 “a relic of the pre-computer era.”

Count expected forced outage events instead of units. Thirty nuclear units with a 2% forced outage rate produce far fewer events per year than ten peaking turbines, and the confidence interval on the nuclear group’s EFOR will be correspondingly wider. Newly registered solar plants face the opposite problem: the 2025 GADS Solar DRI phased in 20 MW plants in 2025, and NERC’s Category 2 registration FAQ defers their first GADS submission to May 15, 2026. Any solar peer group built this year therefore rests on a thin record.

Industry databases, standards, and organizations

Each source below answers a different benchmarking question. The table maps those roles before the detail:

NERC GADS

GADS is the only database in which participation is mandatory, which is why it defines the North American peer universe. The 2026 DRI states that “as of January 1, 2013, GADS became a mandatory industry program for conventional generating units 20 MW and larger.” It covers ten unit types: fossil steam, nuclear, gas turbines and jet engines, internal combustion engines, hydro and pumped storage, combined cycle blocks, cogeneration blocks, multi-boiler/multi-turbine units, geothermal, and miscellaneous conventional units. EPRI’s data guidance identifies GADS as the primary source for historical forced-outage data on more than 5,000 North American units.

The programs have different thresholds and record structures:

• Conventional: GADS uses event records in 07 format and performance records in 05 format, plus design records. Owners must submit nine mandatory design fields before any event or performance data. Detailed design data became mandatory January 1, 2024.
• Wind: The GADS Wind page sets the threshold at plants of 75 MW or greater with a commercial date of January 1, 2005 or later; mandatory reporting began January 1, 2018. The 2026 Wind DRI, version 4.9, requires configuration and monthly performance records, plus event records. Turbine-hour categories include Forced, Maintenance, Planned, and Resource Unavailable.
• Solar: The 2025 Solar DRI requires reporting for plants of 20 MW or greater with commercial operation on or after January 1, 2010. The program phased in plants at 100 MW for 2024 and 20 MW for 2025.

All programs share the 45-day post-quarter deadline. What comes back is limited: NERC’s public outputs are the State of Reliability WEFOR averages and the aging brochures, while quartile distributions require pc-GAR. NERC also controls use of the data. In a February 5, 2026 decision, it denied SPP’s market monitor ongoing access to unit-specific GADS records, stating the “GADS Section 1600 data request is not collected for market monitoring purposes.”

The reporting workload is where many plants lose accuracy. PowerGADS from Integ Enterprise Consulting captures unit events directly from the DCS, SCADA, and PI Historian. It runs the Appendix F performance calculations automatically and produces GADS submissions with a full audit trail for conventional, wind, and solar units. Integ reports that PowerGADS handles GADS reporting for approximately 70% of U.S. generating units. Go-live takes two weeks, and training takes under one hour per user.

EPRI, IEEE 762, ORAP, and other frameworks

IEEE 762 supplies the definitions. EPRI supplies renewable and component-level benchmarks, while ORAP supplies the deepest gas turbine record. Each has a distinct edge.

The IEEE-SA Standards Board approved IEEE Std 762-2023, “IEEE Standard Definitions for Use in Reporting Electric Generating Unit Reliability, Availability, and Productivity,” on March 30, 2023. IEEE published it on August 15, 2023, and the standard supersedes the 2006 edition. It covers conventional and variable energy resource generators. It also adds EFORd equations and energy-weighted group indexes. Other additions include outside-management-control definitions and pooling methods. A P762 revision project, approved December 6, 2023, will extend the standard to standalone and hybrid energy storage.

EPRI’s current reliability tools are program-specific, and the names have changed:

• SUPER (Solar Utilization, Performance, and Reliability): The October 2025 product benchmarks fleet KPIs including capacity factor, performance ratio, inverter downtime, curtailment, and performance loss rate. Access costs $18,000 annually for Program P207 funders and $28,000 for non-funders.
• WinNER (Wind Network for Enhanced Reliability): WinNER v4.0 was published December 3, 2024. The database held 7,358 turbines and 16 GW as of late 2021.
• BEST (Benchmarking Energy Storage Technology) Reliability: Published September 19, 2025, this is the only current program addressing the storage gap NERC has not filled.
• Preventive Maintenance Basis Database: PMBD v7.5, published October 15, 2025, covers maintenance tasks and failure modes for 150+ components. EPRI’s own documentation states that it “provides no data on absolute failure rates.” It is a maintenance-basis tool rather than a benchmarking database.

Legacy EPRI names such as the Plant Reliability Database and NPRDS/EPIX do not appear as current products. EPRI’s most recent ORAP document is a 1999 interim report; ORAP itself belongs to Strategic Power Systems.

The industry created ORAP in the mid-1970s. Strategic Power Systems has operated it since the mid-1980s. A 2015 Gas Turbine World article states that data from more than 2,600 gas and steam turbines streams to SPS monthly, and “no plant data can be added to the ORAP database until it passes both a Manual Data Validation (MDV) and an Automated Data Validation (ADV) process.” Subscribers receive quarterly reports that compare their units against fleets of similar turbines and cover mean time data, component trends, and inspection intervals. A 2022 ORNL report confirms coverage of gas, steam, wind, and reciprocating engine assets. OEM agreements and NDAs protect the data. The classification also differs from GADS: Power Engineering notes that ORAP subsumes some NERC U3 events into “scheduled maintenance,” which understates forced outage hours relative to GADS.

IEEE 1366-2022 marks the far boundary. The IEEE indices reference defines SAIFI = customers interrupted ÷ total customers served, SAIDI = customer minutes of interruption ÷ total customers served, and CAIDI = SAIDI ÷ SAIFI. The Missouri PSC documentation gives its Major Event Day threshold as T_MED = e^(α + 2.5β), using five years of log-transformed daily SAIDI. LBNL notes that the method yields an expectation of 2.3 major event days per year. Utilities report these indices to EIA Form 861. None describes a generating unit, but the common mistakes section below explains how generation analysts can use the 2.5 beta concept when separating extreme events from baseline performance.

Technology-specific benchmarks

The WEFOR table above spans a tenfold range from nuclear to simple-cycle combustion turbines, and no single “good EFOR” exists across fuels. The metrics themselves also change with technology.

• Gas turbines and combined cycle: Simple-cycle combustion turbines post the highest WEFOR of any conventional class, partly because rates use service hours and peakers accumulate few of them. Combined cycle units sit far lower. ORAP’s OEM-and-frame peer grouping matters most here because a heavy-frame F-class fleet and an aeroderivative fleet fail for different reasons.
• Nuclear: Nuclear records the lowest WEFOR of any class. Operators also report to WANO and IAEA under Forced Loss Rate and Unplanned Capability Loss Factor, both measured on different windows and definitions than EFORd. WANO’s 2024 indicator revision moved from Method 4 to Method 10 and replaced Unit Capability Factor with Unit Capability Rate, so nuclear trend lines need a break marker at Q1 2024.
• Coal: WEFOR rose from 11.2% to 14.1% between the 2023 and 2024 operating years. That was the second-largest absolute increase in the table, behind Gas – Steam Turbine’s rise from 19.2% to 22.4%. Plant teams should compare EFOR with FOR directly to quantify how much derating contributes at each coal unit rather than assume a fleet-wide causal pattern.
• Wind: Conventional EFOR does not apply because low-wind hours are RUTH rather than forced hours. NERC’s 2024 State of Reliability overview reported a wind weighted resource forced-outage rate of 18.9% for 2023, up from 18.1% in 2022, and cautioned that it is “not an exact comparison to the WEFOR” for conventional units. Capacity factor provides the productivity benchmark used here. EIA reports fleet capacity factors of 33.2% in 2023 and 34.3% in 2024, with a 2025 preliminary value of 34.2%. LBNL’s wind report puts 2023 at 33.5% and projects built in 2022 at 38.2%. Its executive summary finds that “by year 20, the median wind project has a capacity factor that is roughly 70% that of year 2.”
• Solar: EIA reports utility-scale PV capacity factors of 23.2% for 2023 and 2024 and 24.4% preliminary for 2025. LBNL’s solar report shows plant-level net AC capacity factors from 6% to 36% with a median of 24%. Its report slides put degradation at 0.9%/year for newer plants and 1.47%/year for older ones. For availability, NREL’s PV report gives steady-state system availability of P50 = 0.991, P80 = 0.968, and P90 = 0.947, with a median performance loss rate of −0.75%/year across 4,915 inverters.

Component-level data exists for renewables where plant-level EFOR does not. EPRI’s gearbox study reports failure rates of 0.9%/year in years 0–4 and 1.6%/year in years 5–10. Use those component rates in maintenance planning. Availability and capacity accreditation analyses require plant-level metrics.

Probabilistic methods and system adequacy

A unit’s EFORd is an input to system adequacy analysis. The system’s loss-of-load risk is the output. Analysts use probabilistic methods to convert each unit’s forced outage distribution into a fleet-wide probability that load exceeds available capacity, which is how a plant metric becomes a grid metric.

The core quantities, in order of construction:

1. Capacity Outage Probability Table (COPT): Each unit’s forced outage rate defines a two-state distribution, or a multi-state distribution when deratings apply. The Booth-Baleriaux convolution combines them into F(x), the cumulative distribution of available capacity.
2. Loss of Load Probability (LOLP): For any hour or day, LOLP = 1 − F(x) at the load level. SPP’s LOLE document defines it as the probability that at least one shortfall occurs in the period.
3. Loss of Load Expectation (LOLE): LOLE (days/year) = Σ p_d over 365 daily peaks. PJM Manual 20A defines it as the number of days per year expected to have a loss-of-load event “regardless of the duration or magnitude.”
4. Loss of Load Hours (LOLH): NERC’s ProbA technical guideline computes LOLH from every hourly load and warns that “LOLE of 0.1 days per year is not equivalent to a LOLH of 2.4 hours per year.”
5. Expected Unserved Energy (EUE): Analysts measure EUE in MWh across all hours. NERC’s 2026 ProbA instructions normalize it as (EUE MWh ÷ Net Energy for Load MWh) × 1,000,000, in parts per million.

Monte Carlo simulation replaces the analytical convolution with repeated random draws of unit states and load. It can also incorporate renewable output. In that form, a 2019 research paper defines EUE = (Σ E_i) ÷ N_total, the average unserved energy across N simulated years. Monte Carlo can represent weather-correlated renewable output and storage state of charge without relying on the independence assumption behind the basic COPT.

The industry adopted the 0.1 days/year criterion as a planning convention without deriving it mathematically. The IEEE PES history of LOLP credits C.W. Watchorn with first proposing one day in ten years. Billinton and Chu, cited by SPP, describe the industry coalescing around it through the 1960s. NERC does not mandate it. FERC’s BAL-502-RFC-02 order found that NERC’s continent-wide standards “do not address assessment of resource adequacy in the planning horizon” and called one-in-ten “one common approach” rather than a requirement. Modeling, Data, and Analysis standard MOD-032-1 governs the modeling data that feeds planning cases. The adequacy targets themselves are regional:

Reserve margin is the deterministic shorthand for the same risk. NERC’s 2025 LTRA Data Form Instructions define Anticipated Reserve Margin as (Anticipated Resources − Net Internal Demand) ÷ Net Internal Demand, with the Reference Margin Level set by the regulator or ISO/RTO. If an area supplies none, the 2026 instructions apply a default of 15% for predominantly thermal systems and 10% for hydro systems. They also require that “all variable resources (wind, solar, hydro) must be derated.” NERC’s January 2026 Long-Term Reliability Assessment flags an area as high risk when annual LOLH exceeds 2.4 hours/year or normalized EUE exceeds 0.002%, which is 20 ppm.

Effective Load Carrying Capability is the bridge from a variable resource to a reserve margin. Garver introduced ELCC in 1966. PJM defines it as “a measure of the additional load that the system can supply with a particular generator of interest, with no net change in reliability,” expressed as a percent of nameplate. NERC’s September 2025 staff report calls ELCC “a widely used approach” for wind, solar, and storage. It also notes that outputs are “highly dependent on the assumptions made,” with no formal NERC guideline yet issued. The class ratings now set what a megawatt is worth:

Sources: PJM 2027/28 ratings and 2028/29 ratings, under the marginal ELCC method FERC approved in Docket ER24-99 on January 30, 2024.

MISO’s 2025-2026 LOLE study gives wind ELCC of 20.8% in summer and 29.0% in winter. It gives solar 44.5% in summer and 25.5% in winter, with a transition to Direct Loss of Load accreditation planned for Planning Year 2028/29. SPP’s 2026 ELCC study puts summer wind at 22.3% and summer solar at 62.60%, applied against a 15% base PRM from June 1, 2026.

For the plant, the mechanism depends on the market. NYISO directly applies AEFORd to conventional UCAP. PJM applies an ELCC class rating and a resource performance adjustment, while MISO uses its own availability-weighted accreditation framework. The GADS team should map each event code to the accreditation method that governs the unit before estimating the capacity effect.

Common mistakes

Most benchmarking errors originate upstream of the analysis, in how teams record events and select comparison units. The recurring ones:

• Miscoded events: A U3 logged as an MO, or a forced derating logged as a note rather than an event with Net Available Capacity, moves hours between the forced and planned columns. The 1,700+ GADS cause codes make correct classification a training problem, and the 45-day submission window means errors reach NERC before an annual review catches them.
• Invalid peer groups: Comparing a daily-cycling combined cycle against baseload peers, or a 20 MW solar plant against a population that first reports in May 2026, produces a gap that reflects the group rather than the plant. IEEE 762 specifies no minimum sample size, so the analyst has to justify the group on expected event count.
• Ignoring deratings: A plant that benchmarks on FOR rather than EFOR, or that captures full outages from the DCS but deratings from operator memory, systematically understates equivalent forced hours. For each unit, compare FOR with EFOR to quantify the effect before assigning a quartile.
• Manual entry across disconnected systems: The same event keyed separately into the ISO outage system and the internal outage log, then entered again into the CMMS work order and GADS file, produces four start times and four causes for one trip. Every re-keying creates a chance to miscode, and reconciliation happens after the quarter closes.
• Comparing incompatible metrics: NERC WEFOR is capacity-weighted, while EFORd is demand-conditioned. GADS Wind uses a weighted resource forced-outage rate. WANO uses a 36-month Forced Loss Rate, and ORAP reclassifies some U3 hours as scheduled maintenance. A chart that places these metrics on one axis without adjustment is wrong before the first data point is plotted.
• Storm outliers distorting comparisons: Winter Storm Elliott generated $1.8 billion in PJM non-performance charges in a single event. A unit’s three-year EFORd with Elliott included and excluded are two different numbers. IEEE 1366’s 2.5 beta method separates extreme days from baseline distribution reliability. Generation benchmarking has no codified equivalent, so report both figures and label them.

Each mistake also creates a compliance or decision risk. A miscoded event creates self-report exposure. An invalid peer group supports a bad capital decision, while manual re-entry leaves an incomplete audit trail.

Getting started

Start with the gap rather than the tool. The plant team can use the sequence below to move from a single benchmark comparison to a corrected root cause. At each step, the team creates an artifact needed for the next step.

1. Compute your indices under Appendix F. Run FOR, EFOR, EFORd, AF, EAF, and MSTFO for each unit for the last three years, using the 2026 Appendix F formulas and Method 2 pooling. Anything else will not reconcile to pc-GAR.
2. Build the peer group and quantify the gap. Select units by type and size class, then account for duty cycle. Confirm that the group produces enough forced outage events to support the interval. Record the delta between your EFORd and the group’s 25th percentile in equivalent hours and in MWh.
3. Categorize and rank lost generation. Sort every event into the categories required for its asset class, multiply the applicable outage or derating hours by net maximum capacity, and Pareto-rank the results by cause code.
4. Correct root causes in ranked order. Work the top cause codes first, then re-run the indices at the next quarter close to confirm the gap moved.

Step 3 is where manual data collection breaks. POWER Magazine documented a fleet where collecting, analyzing, and reporting process data took approximately 40 hours per week per plant before automation. Automated capture pulls the event from the source. OPC UA, released as IEC 62541 in October 2011, provides the standard path from DCS and SCADA into the historian. A PI Connector can map OPC UA variables directly to PI points and PI Asset Framework elements. Critical Infrastructure Protection standard CIP-005-8 permits only needed routable protocol communications through the Electronic Security Perimeter. Department of Energy guidance recommends a historian in an internal DMZ with data flowing one direction out of the operational network. Reliability analytics should read from the historian in the DMZ without touching the control network.

A consolidated event record removes repeated manual entry and reduces the data-collection burden. The available research does not establish a measured event-to-GADS-submission latency, so plants should measure that interval directly before and after automation. If your team is still manually entering unit events and running GADS calculations by hand, PowerGADS connects directly to your control systems and automates the full reporting workflow. Most customers go live in two weeks. PowerAnalytics, the engine underneath PowerGADS, connects natively to PI Historians and DCS/SCADA. It also connects to ISO/RTO feeds and includes 50+ out-of-the-box solutions, so the Pareto ranking in step 3 and the peer comparison in step 2 run against the same validated event record that produced the GADS file. To see how this works against your specific fleet configuration and ISO connections, schedule a demo with the Integ team.

FAQ

What is the difference between FOR, EFOR, and EFORd?

FOR counts only full forced outage hours against service hours, so a unit running derated posts the same FOR as one at full load. EFOR adds equivalent forced derated hours, converting partial capacity losses into full-outage-hour equivalents. The 2026 Appendix F denominator also includes reserve shutdown hours. EFORd restricts both terms to periods when the unit was in demand, which is why NYISO uses it for UCAP derating and PJM used it for accreditation through the 2024/2025 Delivery Year.

How does GADS peer comparison work?

Every NERC-registered generator owner submits event and performance records, along with design records, to GADS quarterly. NERC aggregates them by unit type and size class. Public outputs are limited to capacity-weighted averages in the State of Reliability and the brochures through data year 2022. Percentile distributions and custom peer groups require NERC’s licensed pc-GAR software, where you select the population and compute the same Appendix F indices against it.

How large does a peer group need to be?

No standard sets a number. IEEE 762-2023 states that “no minimum sample size is specified in this document,” and NERC Appendix F explicitly supports small-group EFORd calculations by summing hours. The 30-unit figure is a general statistics rule of thumb that breaks down for low-frequency events. For forced outage rates, judge adequacy by the expected count of events in the group. Expect nuclear or new solar groups to need many more units than peaking gas groups.

What are the top-quartile thresholds for EFOR and MTBF?

NERC does not publish quartile cutoffs, only averages such as a 2024 operating-year WEFOR of 5.7% for combined cycle and 2.2% for nuclear. The commonly quoted MTBF of 800 hours and planned maintenance percentage of 85% do not appear in any NERC, IEEE, EPRI, or ORAP document. Define top quartile as the 25th percentile of your own pc-GAR peer group’s EFORd.

How should deratings be treated?

Record every derating as a GADS event with Net Available Capacity, which converts it into equivalent forced derated hours (EFDH) or equivalent unplanned derated hours (EUDH). Deratings enter EFOR, EFORd, and EAF but not FOR or AF. For wind and solar, classify resource-unavailable hours and curtailment under the applicable RUTH or reserve-shutdown rules rather than as forced deratings.

Which standards bodies and databases apply?

IEEE 762-2023 defines the indices. NERC GADS is the mandatory database for conventional units of 20 MW or more, wind plants of 75 MW or more, and solar plants of 20 MW or more. EPRI’s SUPER, WinNER, and BEST programs benchmark solar, wind, and storage for members. ORAP, run by Strategic Power Systems, provides subscription-based component-level gas turbine benchmarking. IEEE 1366-2022 governs distribution indices and does not apply to generating units.

How does Monte Carlo simulation connect plant metrics to system adequacy?

Each unit’s forced outage rate defines a probability of being unavailable. Monte Carlo simulation draws thousands of possible years using unit states and loads, plus renewable output. It counts the hours and megawatt-hours where load exceeds available capacity, then averages them into LOLE, LOLH, and EUE. Planners measure the reserve margin or ELCC needed to hold LOLE at 0.1 days per year, so changes in fleet outage performance affect the capacity the system must procure.