The EPC contractor says the plant passed its performance test. The owner’s engineer reads the same data and says it failed. Nobody changed a number. They disagree about the test itself. Which days count? Which temperature is the result corrected to? Is the grid outage in week two excluded? Is the answer an average of ratios or a ratio of sums?
Most handover disputes start there. The plant can be well built and still fail a badly written test, and a weak plant can pass a loose one. This guide is part of our solar engineering hub. It is written for developers and asset owners who want the test procedure fixed in the EPC contract before the first module is energised. It builds on our IEC 61724 monitoring explainer, which covers sensor classes. This post covers the acceptance test that sits on top of that monitoring.
Quick answer. A performance ratio (PR) test compares a plant’s measured AC energy with the energy its DC capacity and the measured sunlight should have produced over a set window. A capacity test compares power with a model at fixed conditions. Neither runs itself. The contract must fix the duration, irradiance filter, temperature correction, exclusions and pass threshold, or the result will be argued.
The standards that define these tests are sold by IEC and ASTM. This guide uses their public scope pages, free previews, open NREL and IEA reports, and one public Indian tender. Buy the edition your contract names before you draft numeric limits.
What is the difference between a PR test and a capacity test?
A PR test is an energy test. A capacity test is a power test. They answer different questions, and a contract should say which one it uses and why.
| PR test | Capacity test | Energy evaluation | |
|---|---|---|---|
| What it compares | Measured AC energy against the energy implied by DC nameplate and plane-of-array (POA) irradiation | Measured power against modelled power at fixed reporting conditions | Measured energy against expected energy from actual weather |
| Defined in | IEC 61724-1:2021 (metric) plus the contract’s own procedure | IEC TS 61724-2:2025 or ASTM E2848-13(2023) | IEC TS 61724-3:2016 |
| Typical window | Days to weeks (the NREL sample protocol uses 5 days, the SECI tender below uses 30) | About 3 to 5 days under IEC TS 61724-2 | An extended period, such as a year |
| Inside the boundary | Every loss that lowers AC energy, including downtime | Installed capacity and conversion performance while output is unconstrained | Operating-time energy, with downtime reported as availability |
| Weak point | Biased by temperature and season unless corrected, unfair if inverters clip | Says nothing about annual energy | Needs a long window, which sits awkwardly with handover payment dates |
Sources: IEC TS 61724-2:2025, IEC TS 61724-3:2016, ASTM E2848-13(2023), NREL/TP-5200-57991, NREL/TP-5200-60628 and IEA PVPS T13-25.
Three points from those documents decide how you use the table.
- A capacity test excludes constrained operation by design. Per the IEC webstore, IEC TS 61724-2 evaluates the plant only when output is not limited by the inverters’ AC power limit. ASTM E2848 says its measured capacity should not be used to make claims about energy generation.
- A PR test counts downtime. The IEA PVPS O&M guidelines (2022) describe PR as capturing the combined effect of all losses, including system or network downtime. That is why exclusions matter so much in a PR test (see the exclusions section).
- PR is a poor commitment metric for a plant that clips. The NREL weather-corrected PR protocol says PR should not be the metric if inverters clip at high irradiance, because PR penalises those periods unfairly. High DC to AC ratios make clipping common. IEC TS 61724-2:2025 was revised for this case: its foreword cites the limited times of unconstrained operation caused by clipping and curtailment.
Which test do contracts use? Some Indian utility tenders use a PR test, and the SECI example below is one. A 2024 paper in the IEEE Journal of Photovoltaics (Deline et al., with NREL authors) calls the capacity test especially widespread in the United States. It is typically run at the completion of construction, and ASTM E2848 is the most used method. A contract can also use both, for example a capacity test at completion and a PR or energy test later. Whichever you choose, the rest of this guide applies to the procedure, not the label.
The IEC TS 61724-2:2025 edition (Edition 2.0, published 22 September 2025) replaces the 2016 edition. It adds bifacial systems and a test that can finish in three to five days in favourable conditions, per its foreword. The published ASTM E2848 does not contemplate bifacial systems, per the same 2024 paper. Specify rear-side irradiance treatment in the contract if the plant is bifacial.
How is a temperature-corrected PR calculated?
Raw PR divides measured energy by the energy that nameplate power and measured irradiation imply:
PR = E_AC / ( P_STC x H_POA / G_STC )E_AC is AC energy at the contract meter, P_STC is DC nameplate at standard test conditions (STC), H_POA is plane-of-array irradiation over the period and G_STC is 1,000 W/m². See our performance ratio glossary entry for the plain definition.
Raw PR moves with the weather. Module output falls as modules get hotter, so a hot week gives a lower PR and a cool week a higher one for the same plant. The NREL weather-corrected PR report (Dierauf et al., April 2013) reports seasonal PR swings of up to about 10%. It warns that an uncorrected PR can give a false pass in winter and a false fail in summer. In its single-axis tracker simulation, a weather file 3 °C warmer lowered PR by 0.9 percentage points.
The correction divides PR by a factor built from the module temperature coefficient of power (gamma, negative):
PR_TC = PR / ( 1 + gamma x ( T_mod - T_ref ) )The IEA PVPS O&M guidelines present this form and note that it is the form in IEC 61724-1. They also state that when a period is aggregated, the mean module temperature should be irradiance weighted. The standard applies the correction interval by interval. The period-average form above is an approximation that keeps the arithmetic readable.
The formula is the easy part. The contract has to settle two inputs, the reference temperature and where the module or cell temperature comes from. Public documents make different choices:
| Source | Reference temperature | Temperature input |
|---|---|---|
| IEC 61724-1 form, as summarised by IEA PVPS T13-25 | An agreed value, often the annual weighted module temperature from a standard model, or 25 °C | Measured module temperature |
| NREL weather-corrected PR | Irradiance-weighted annual average cell temperature from the project weather file | Cell temperature computed from measured weather using the same heat-transfer model as the annual average |
| SECI 300 MW tender, Annexure-A | 25 °C | Average of the module temperature sensors, per 15-minute block |
NREL argues against correcting to 25 °C. Modules more often run near 45 °C, so a 25 °C correction usually raises PR, can overstate real performance and changes the PR value the contract states. NREL also argues for computing cell temperature from weather data instead of measuring it. A direct measurement can hide design or construction faults that cause high operating temperatures, and so give a false pass. These are NREL’s recommendations. A tender can choose differently, as SECI did, if it says so.
Worked example: same contract, two test weeks
All numbers below are illustrative assumptions, not project data. Two 25 MWp plants have the same contracted PR of 81.0%, defined on an annual-equivalent basis. The reference module temperature is 46 °C, taken as the irradiance-weighted annual value from the project weather file. Gamma is -0.35% per °C. Each plant is tested for five days with H_POA of 32.0 kWh/m².
| Plant A (hot week) | Plant B (cool week) | |
|---|---|---|
| Measured AC energy | 636,000 kWh | 656,800 kWh |
| Raw PR = E_AC / (25,000 kWp x 32.0 kWh/m² / 1 kW/m²) | 0.795 | 0.821 |
| Irradiance-weighted module temperature | 52 °C | 38 °C |
| Correction factor, 1 + gamma x (T_mod - 46) | 0.979 | 1.028 |
| Corrected PR to the 46 °C reference | 0.812 | 0.799 |
| Result against 81.0% | Pass | Shortfall of 1.1 points |
| Raw PR against 81.0% | Fail | Pass |
| Corrected to 25 °C instead (factor 0.9055 and 0.9545) | 0.878, pass | 0.860, pass |
The raw reading gets both plants wrong. The correction to the project’s own reference temperature reverses it: Plant A passes and Plant B has a real shortfall that cool weather was hiding. Correcting to 25 °C instead makes both plants pass, and Plant B passes with five points of margin it does not have. The two corrections compare the test against different yardsticks.
One more number belongs in the contract. NREL’s sample protocol recommends 95% as a default tolerance on the contracted PR, to allow for measurement uncertainty, and says the parties can change it. On an 81.0% commitment, 95% is 76.95%. Plant B’s 79.9% would pass inside that tolerance. A tolerance is a commercial term. Write it as an explicit figure and know what it is worth in PR points.
How long should a solar performance ratio test run, and which data count?
The standards leave you room, and tenders fill it differently. Compare four public references:
| Source | Window | Valid-day and data rules |
|---|---|---|
| NREL sample protocol (2013) | 5 days, extended until 5 valid days are recorded | At least 3 days with POA above 600 W/m² for 3 continuous hours and a daily total above 3,000 Wh/m². No liquidated damages for a weather delay. 15-minute records from a scan rate of at least one minute |
| SECI 300 MW tender (FY 2023-24) | 7-day trial run, then 30 consecutive days, starting within 60 days of commissioning | Blocks with POA below 200 W/m² dropped. A day counts only with at least 24 valid 15-minute blocks, otherwise the test extends |
| IEC TS 61724-2:2025 | About 3 to 5 days in favourable conditions | The contents list filters for POA irradiance, inverter utilization, partial shading, angle of incidence, invalid readings, stuck values, missing data, equipment not functioning, trackers not tracking and unstable irradiance. Numeric limits sit in the paywalled text. Annex A is a template for documenting test elections |
| ASTM E2848-13(2023) | Set by the user before measurement | The parties must agree the irradiance measurement plane before an acceptance test. NREL authors describe a target period of one to two weeks in their capacity method |
Sources: NREL/TP-5200-57991, the SECI tender (tender SECI/C&P/OP/11/013/2023-24), IEC TS 61724-2:2025 free preview, ASTM E2848 and Waters et al., NREL, 2019.
The SECI document is a public tender for a 300 MW (AC) balance-of-system package in Andhra Pradesh. It is one buyer’s terms, not a template for a private C&I plant. Its value here is that it is complete and readable.
Decide five things and write them down:
- Start deadline and duration. How many days after commissioning the test must start, and how many valid days it needs.
- Valid-day rule. The minimum irradiance and the minimum number of valid data blocks for a day to count.
- Irradiance floor. The SECI floor is 200 W/m² and the NREL sample criterion is 600 W/m² for a day to qualify. They do different jobs, and the contract should state which job it means.
- Interval. Raw one-minute data and the interval used for the calculation. Name the one that the pass decision uses.
- Extension rule. What happens if the weather never delivers enough valid days. NREL’s sample protocol extends the test with no liquidated damages and accepts the plant if the five strongest days meet the contracted PR. A contract that is silent here turns a cloudy fortnight into a dispute.
Which sensors and calibration records does the test need?
The test is only as good as its weakest instrument. The questions below should be answered in writing before the test starts.
- Which irradiance sensor? NREL’s energy evaluation report (Kurtz et al., 2013) notes that a commitment derived from thermopile-validated historical weather data should logically be tested with thermopiles. It also notes that matched reference cells respond to angle and spectrum more like the modules do. It says all parties should agree on the instruments. The NREL sample PR protocol uses two matched reference cells with a target uncertainty of 3%, and records a calibrated pyranometer for reference only. The SECI annexure uses the average of all pyranometers. Our IEC 61724 explainer covers the sensor classes.
- Which plane? ASTM E2848 requires the parties to agree the plane of irradiance measurement before an acceptance test. For a tracker or a bifacial plant, say whether the reading is front-side POA, rear-side or a combination.
- Which meter, and where? SECI reads energy at the plant-end ABT meter and states the PR at the 220 kV side of the pooling substation. Losses between the inverter and the meter either count or they do not. The contract should name the meter and its voltage level.
- Calibration and condition. The SECI annexure requires valid calibration certificates for pyranometers and temperature sensors, a tilt check before the test and visual cleanliness checks during it. The NREL report adds a data check: night-time irradiance that is not zero points to a zero-calibration error.
- Timestamps. IEC TS 61724-2 includes a synchronization requirement in its contents. Write the time source into the procedure so that energy and irradiance records line up.
What can be excluded from the test?
Exclusions protect both parties. An owner should not reject a plant because the grid tripped, and a contractor should not pay for a day the utility shut the plant down. The risk is vague wording. Public sources give a starting list:
| Exclusion | What the sources say | Drafting point |
|---|---|---|
| Grid outage | IEA PVPS: periods of network downtime are typically excluded. SECI: 15-minute blocks of grid interruption are removed, recorded jointly by the contractor and SECI and verified from SCADA | Name the evidence, such as a SCADA alarm plus a joint log, and the block size |
| Curtailment | IEA PVPS: the typical approach is to exclude curtailed periods and calculate a non-curtailed PR, with careful analysis if curtailment covers more than 50% of the time. IEC TS 61724-2:2025 was revised for curtailment | State what counts as proof of an instruction, and how much valid time must remain |
| Owner or utility shutdown | SECI: a shutdown explicitly demanded by the owner, DISCOM, STU or CTU is removed | Same evidence rule as above |
| Force majeure | SECI removes force majeure conditions | Define it in the contract, because the test annexure will not |
| Low irradiance | SECI removes blocks with POA below 200 W/m² | Pick a floor and say why |
| Clipping | NREL: PR is unfair to a plant that clips at high irradiance. IEC TS 61724-2 lists an inverter utilization filter | For high DC to AC ratios, filter or prefer a capacity test |
| Missing or bad data | SECI interpolates missing one-minute records and flags bad 15-minute blocks with a reason. NREL: the contractor supplies raw data before any manipulation and highlights gaps | Say who may flag data and who must agree |
IEA PVPS adds one rule that applies to every row: as IEC 61724-1 requires, record every data exclusion.
Equipment failure deserves its own sentence. The SECI annexure says that during the test, equipment failure or interruption of any kind, except SCADA communication failures, will not be accountable. It adds that the test may resume once the system is rectified. It also requires a full 30-day repeat after an outage of key equipment for more than 7 days. The listed equipment is power and inverter duty transformers, power conditioning units, HT switchgear, the SCADA and data logger combination, the tilted pyranometer and other weather sensors. The phrase “not accountable” can be read two ways: the affected intervals leave the average, or the contractor is not charged for them while they stay in. A clause that needs interpretation is a clause that gets disputed. Say which one you mean.
What should the EPC contract say about the test?
This checklist turns the sections above into clauses. It is a drafting aid, not legal advice. Have contract counsel review the wording, and confirm project-specific requirements with your lender, utility and DISCOM.
| # | Clause | Write this down | Why |
|---|---|---|---|
| 1 | Standard and edition | Name the exact edition, for example IEC 61724-1:2021, IEC TS 61724-2:2025 or ASTM E2848-13(2023), and say what governs if the annexure differs | Editions change. IEC lists a 2026 stability date for IEC TS 61724-3:2016 and IEC 61724-1:2021 |
| 2 | Metric and formula | Print the equation with every symbol defined | NREL’s lesson: a party unfamiliar with the process should be able to repeat the evaluation with virtually zero deviation |
| 3 | Reference temperature and temperature source | State T_ref and whether module or cell temperature is measured or computed, and by which model | NREL, IEA PVPS and SECI each choose differently |
| 4 | Baseline model and weather file | Attach the simulation file version, the weather file, the wind-speed treatment and the loss assumptions as an appendix | NREL: agree the project weather file before setting a PR. Anderson et al. (2021): PVsyst’s wind coefficient defaults to zero |
| 5 | Measurement boundary | Meter, voltage level, sensors and their locations | NREL: define what is inside and outside the test boundary. SECI: ABT meter, 220 kV side |
| 6 | Sensors | Type, number, matched cells or pyranometers, calibration certificate dates, tilt check and cleaning interval | SECI and NREL both require calibrated instruments |
| 7 | Preconditions | Commissioning tests complete, punch list rectified, SCADA and weather station functional, and the procedure agreed in writing before the test | SECI Annexure-A 2.1.1 and 2.1.4.1. NREL requires an agreed procedure before the test |
| 8 | Window | Start deadline after commissioning, duration and any trial run | SECI: 60-day start, 7-day trial, 30 days |
| 9 | Valid days and filters | Irradiance floor, minimum valid blocks, interval and the extension rule for poor weather | NREL and SECI, see the table above |
| 10 | Exclusions | Each exclusion with its evidence source | IEA PVPS and SECI |
| 11 | Averaging method | Ratio of sums or average of interval ratios | See the next section |
| 12 | Pass criterion and tolerance | The threshold, the tolerance as a number and its value in PR points | NREL’s 95% default is 4.05 points on an 81.0% commitment |
| 13 | Availability | A separate availability metric and threshold | NREL’s pass condition requires the availability commitment to be met. IEC TS 61724-3 reports availability as its own metric |
| 14 | Roles and records | Who runs the test, who witnesses it, raw data to every party before any manipulation, and a deviation log signed by all | NREL protocol |
| 15 | Failure path | Number of retests, when a full repeat is required, the damages formula and cap, and whether an as-is acceptance exists | SECI, see the next section |
| 16 | Decision on disputes | Who decides a data disagreement and within what time | Mechanism is a matter for counsel |
If nobody on your side has written a procedure like this, the review should happen before the test, not after the report. Heaven Designs offers MW-scale project management consultancy. Its page describes project management for utility-scale solar from feasibility through commissioning, and lists commissioning and handover among its scope areas as system testing, performance verification and project handover. The page presents a management, coordination and quality-oversight service. Scope and turnaround are agreed at scoping. Our guide to the owner’s engineer role explains where a test-procedure review fits in a project.
Why do PR test results get disputed, and what happens on a retest?
Most disputes trace to a choice the contract never made. Five recur in the public documents.
- Season. NREL found that uncorrected PR swings with temperature. A contract that gives a PR but no season or correction hands the result to the weather.
- Reference temperature. The worked example above shows how 25 °C and an annual reference flip a result. The documents above use both.
- Averaging. NREL’s equation divides the sum of measured AC energy by the sum of temperature-corrected DC energy over all eligible intervals. The SECI annexure takes each day’s PR as the average of the 15-minute block PRs that fall within a 99% confidence interval, then averages the daily PRs over the 30 days. Both are written down. They do not give the same number.
- Exclusions. Each ambiguous exclusion is an argument about which blocks stay in.
- Sensors. Two irradiance sensors that disagree by a few percent move PR by the same amount.
A small illustration of point 3, with assumed numbers. Take two 15-minute blocks in one day. Block 1 has POA of 250 W/m² and a block PR of 0.74. Block 2 has POA of 900 W/m² and a block PR of 0.82. The simple average of the two is 0.780. Weighting by the reference energy, which is proportional to irradiance for equal time steps, gives (0.74 x 250 + 0.82 x 900) / 1,150 = 0.803. The same data differ by 2.3 points depending on the method. Neither is wrong, and the contract must pick one.
What happens if the plant fails?
The SECI annexure shows one full failure path. The contractor gets two more chances after corrective measures. The test repeats for 30 days after a long outage of key equipment. Liquidated damages are 1% of the contract value for every 0.01 of PR shortfall. If the shortfall exceeds 0.05, the plant is accepted as-is and the operational acceptance milestone payment is not made. These are SECI’s commercial terms in one tender. Another contract may set different ones, and our guide to liquidated damages, PI insurance and performance bonds covers how such terms are structured.
For weather, the NREL sample protocol takes a different route. It extends the test until enough valid days exist and states that the delay triggers no liquidated damages.
The signed test records also reach beyond the contract. Our guide to solar drawdown certificates summarises the IREDA terms for the lender’s engineer. Under them the engineer remarks on site performance and efficiency tests and certifies compliance with the contract’s performance terms. A test nobody can reproduce is a weak record for either audience.
How do ASTM E2848 and IEC TS 61724-2 capacity tests work?
Both compare measured power with modelled power under fixed reporting conditions, and both live or die on the model.
ASTM E2848 fits a multiple linear regression of AC power to POA irradiance, ambient temperature and wind speed over the test period. It then evaluates measured and modelled power at the reporting conditions and takes the ratio (Waters et al., NREL, 2019). IEC TS 61724-2 follows different principles. As those authors and a 2024 NREL-authored paper describe the first edition, it applies a deterministic correction for temperature and irradiance and compares measured and expected power directly at the test conditions. The contents of the 2025 edition list three performance model options for the expected power.
The traps are in the inputs:
- Reporting wind speed. Anderson, Kemnitz and Boyd (NREL, 2021) found that moving the reporting wind speed from 0 to 5 m/s shifted the capacity test result by more than 4 percentage points on average, when the model had a fitted wind coefficient. For three of the five models they compared, each extra 1 m/s of reporting wind speed raised the capacity ratio by about 0.8%. They note that ASTM E2939 sets the reporting conditions using the average wind speed after filtering, but that some industry tests assume 1 m/s.
- Model wind coefficient. They also note that PVsyst’s wind coefficient in its cell temperature model is zero by default. The modelled power then barely responds to wind while the measured power does. The baseline model and the test must treat wind the same way.
- Bifacial gain. Waters et al. found that capacity error for bifacial systems fell to the monofacial level once rear irradiance was included, measured or modelled.
A capacity test confirms what was built under defined conditions. It does not forecast energy, and ASTM E2848 says so. Pair it with a PR or energy commitment if annual energy matters to the loan or the PPA.
What should you do before the test date?
- Get the procedure annexure in writing and agree it before the contractor schedules the test.
- Check the sensors against clauses 5 and 6 in the checklist and ask for calibration certificates now.
- Run the procedure on a week of existing SCADA data. A dry run finds the exclusion and averaging disputes while they are still cheap.
- Confirm the baseline model and weather file are the ones the contract names.
- Agree who witnesses, who signs the daily record and when raw data reaches every party.
If you want help reviewing a test procedure before the date, contact the Heaven Designs team with the draft annexure and the PPA or EPC clauses it must match. Our MW-scale project management consultancy page describes the wider scope. For the inspection and electrical tests that come first, see our commissioning checklist and the IEC 62446 acceptance testing guide. For typical PR values, see our PR benchmarks.
Review date: this post uses IEC TS 61724-2:2025, IEC TS 61724-3:2016, IEC 61724-1:2021 and ASTM E2848-13(2023) as listed on 31 October 2026. We will recheck the editions and the SECI test annexure by 30 April 2027.
FAQ
Is a PR test or a capacity test better for handover?
Neither is better in general. A capacity test is short and tests the plant while output is unconstrained, so it suits projects with high DC to AC ratios or curtailment risk. A PR test tests energy over a longer window and counts downtime, so it suits contracts tied to energy outcomes. The contract should name one and explain which risk it covers.
How many days does a PR test take?
It depends on the contract. The NREL sample protocol uses 5 valid days. The SECI tender uses a 7-day trial run and then 30 consecutive days. IEC TS 61724-2:2025 describes a capacity test of about 3 to 5 days under favourable conditions. Fix the duration and the valid-day rule in the EPC contract.
Should the PR test be corrected to 25 °C?
Only if the commitment was set on the same basis. NREL recommends correcting to the annual average cell temperature from the project weather file, because 25 °C usually raises PR and can overstate performance. SECI’s annexure uses 25 °C. Whatever you choose, state the reference temperature and the temperature source in the contract.
Can curtailment and grid outages be excluded?
Usually yes, if the contract says how. IEA PVPS guidance says network downtime is typically excluded and that the typical approach to curtailment is to exclude the curtailed periods. The SECI annexure removes grid interruption blocks verified from SCADA. Name the evidence and keep a record of every exclusion.
Does IEC TS 61724-2 replace the PR test?
No. IEC TS 61724-2 defines a power performance index or capacity test. It does not define the energy-based PR test used in many Indian tenders. IEC TS 61724-3 covers the energy evaluation method. The contract chooses which one applies.
Where can I read the standards?
IEC and ASTM sell the full texts. The IEC webstore and the ASTM page give scope descriptions. The NREL reports cited here are open on OSTI.