Capacity Guarantees, PR Guarantees, and the Risk Nobody Signs For

A plant can pass every performance test in its EPC contract and still miss the revenue in the base case by a wide margin, with no claim against anyone. The contractor guarantees that the equipment is present and that it converts sunlight efficiently. It does not guarantee that the sun shows up. That last variable is the largest single driver of annual revenue and it is the one item in the financing case that no counterparty has signed for.

This is the fourth post in the completion-and-delay series, and the first of two on performance. The previous three dealt with a plant that finishes late. This one deals with a plant that finishes on time and simply produces less than the model said. You will get the difference between the three things that can be guaranteed, how performance liquidated damages are actually calculated, the testing gap between provisional and final acceptance, and a worked comparison of covered against uncovered shortfall.

ℹ️ Note: This describes how these mechanisms work in practice. It is not legal, tax or investment advice — testing regimes and drafting vary considerably between contracts and markets.

What Can Actually Be Guaranteed in a Solar EPC?

Three distinct things, commonly conflated in conversation and carefully separated in the contract. Stoel Rives' Law of Solar guide identifies the pair that normally appear as express obligations — "output/capacity guarantees and performance ratios guarantees" — tested over "30-90 days or the first year of operations."

The third thing is the one your model actually runs on.

	Capacity guarantee	Performance ratio guarantee	Annual energy
What it asks	Is the promised equipment installed and working?	Does the plant convert available irradiance at the guaranteed efficiency?	Did the plant produce the MWh in the base case?
Measured against	Module nameplate / DC capacity	Pre-construction yield assessment simulation	The P50 energy estimate
Test window	Short, point-in-time	7–15 days at PAC; long-term at FAC	A full year, repeatedly
Weather	Corrected to reference conditions	Corrected to reference conditions	Not corrected — it is the variable
Who is liable if it fails	Contractor	Contractor	Nobody

That final column is the subject of this post. Capacity and PR are engineering obligations with a counterparty behind them. Annual energy is a forecast, and a forecast has no obligor.

What Does a Capacity Test Prove?

That the hardware you paid for is physically present and functioning at its rated output. Per Solar Best Practices, the capacity test proves "the installed capacity can reach the level promised" by calculating "DC capacity of the plant, based on the peak powers of the installed PV modules."

It is a procurement check more than a performance one. It catches the failures that matter enormously and are, thankfully, rare — modules that do not meet their datasheet, strings wired incorrectly, inverters clipping below specification, sections of the array not energised at all.

What it cannot tell you is anything about a year. A capacity test is a photograph taken under corrected conditions at a single moment. The revenue line in your model is a film.

What Does a Performance Ratio Test Prove?

That the plant converts the irradiance it receives into electricity at the guaranteed ratio. PR is the bridge between resource and output: it divides actual energy produced by the energy theoretically available given measured irradiance and module rating, so it isolates plant efficiency from weather.

The provisional test is short. Solar Best Practices describes a PR test running "7 to 15 days, depending on the contract," with minimum weather conditions to make the result meaningful — "8 days over a 15-day period with irradiance greater than 5kWh/m²/day" and "irradiance over 500W/m² during at least 3 consecutive hours in 8 days." The measured result is validated against "the guaranteed value based on the pre-construction yield assessment simulations."

Because PR is weather-corrected by construction, a high PR in a dark fortnight is a genuine pass. That is the point of the metric — and it is also precisely why passing it tells you nothing about whether the plant will hit its revenue.

ℹ️ Note: PR is deliberately insulated from the resource. Every sensible technical argument for using it as the acceptance metric is simultaneously the reason it cannot protect the revenue line. The metric is doing its job; the problem is that people read the pass as a revenue assurance.

Why Doesn't Either Test Guarantee Your Revenue?

Because the largest input to annual revenue is the resource, and the resource belongs to nobody. Debt was sized against a P50 energy estimate — a central case with genuine variance around it — and the EPC contractor never took that variance.

Follow the chain. The yield assessment produces a P50 annual energy figure. The financial model turns that into revenue and the lender sizes debt against it, typically with its own haircut. The EPC contract then guarantees a performance ratio measured against that same yield assessment's simulation assumptions, not against its energy output. If the plant converts perfectly and the year is simply poor, PR passes, the contractor is discharged, and the revenue is short.

This is not a drafting failure. Nobody will sell you a guarantee on the weather at a price worth paying. But it does mean the risk register has a large entry with no counterparty, and it should be modelled as such rather than quietly absorbed into a base case.

The practical consequence is that performance LDs and resource variance are different orders of problem. One is a claim. The other is an equity outcome.

Why Is the Yield Assessment the Document That Matters Most?

Because both the guarantee and the debt trace back to it, and they draw different things from it. The PR guarantee is validated against the assessment's simulation — Solar Best Practices puts it as the guaranteed value "based on the pre-construction yield assessment simulations" — while the financial model takes the assessment's P50 energy and turns it into revenue. One document, two consumers, no reconciliation between them.

That structure has an uncomfortable property: an error in the assessment is invisible to both tests.

Suppose the assessment is optimistic. The P50 energy is too high, so debt is sized against revenue the plant will not produce and equity underwrites a return it will not earn. But the PR guarantee is measured against that same optimistic simulation, so the contractor's obligation is calibrated to the same wrong baseline. The plant can convert exactly as the simulation predicted, pass its PR test cleanly, and still miss the model — and no test anywhere in the package flags it, because every test agreed with every other test.

A conservative assessment produces the mirror image: an easy PR guarantee, a contractor discharged on a low bar, and headroom that shows up as outperformance nobody priced.

Two practical checks follow, and they are cheap.

Confirm the guarantee and the model reference the same assessment. Yield assessments get revised through development — new met data, a layout change, revised soiling or availability assumptions. The EPC was often negotiated against an earlier revision than the one the financial model eventually used. Matching the revision number takes minutes and is almost never done.

Read the independent engineer's review as a commercial document, not a technical one. The IE's adjustments to the assessment move the P50, and the P50 moves the debt. That review is where the number the whole structure rests on is actually set.

Three Risks, Three Documents, No Overlap

Worth setting out explicitly, because the gaps between them are where the unguaranteed revenue lives.

Risk	Covered by	Not covered by
Conversion — does the plant turn resource into electricity as designed?	The PR guarantee	Resource, grid
Resource — will the sun shine or the wind blow?	The yield assessment, as a probability	Conversion, grid
Grid — will the network accept the output?	Nothing in either document	—

The third row is the one that catches projects. A performance ratio test and an energy yield assessment between them cover the plant and the weather, and neither has anything to say about whether the network will take the energy. Curtailment, basis and capture rate are separate exposures with their own distributions, and they are addressed in the market structure series rather than in any document the EPC contractor signs.

The related trap is inside the second row. A P90 is not a scenario, it is a point on a distribution, and the distribution depends on the period it is measured over. Interannual weather variation averages out across a long period while systematic uncertainty — the resource dataset, the long-term adjustment, the conversion model — does not. On plausible assumptions a wind project's one-year P90 is about 89% of P50 while its fifteen-year P90 is about 92%.

That matters here for a specific reason. A performance guarantee is tested in a single year, usually the first operating year, and a single year carries the full interannual variability. A debt facility is sized over fifteen, where most of that variability has averaged away. So the same assessment supports two legitimately different numbers, and using one where the other belongs is a three-point error in either direction.

The practical instruction is to be precise about which question is being asked:

• Testing a performance guarantee? One-year distribution.
• Sizing the debt? Term distribution.
• Setting an annual coverage covenant? One-year distribution again.

All three numbers come from the same yield assessment, and the assessment almost always contains them. They rarely make it into the financial model, which typically carries a single cell labelled "P90" with no period attached.

How Are Performance Liquidated Damages Actually Calculated?

By capitalising the shortfall — converting a percentage of lost output into a single payment that compensates for the revenue the project will not earn over the life of the asset.

Solar Best Practices sets out the two-stage approach. At intermediate acceptance, "LDs are based on the annual production shortfall and the electricity selling price of the PV plant." At final acceptance, they "are also calibrated to reflect the loss of revenues that are expected for the full project lifetime," calculated as the "Net Present Value of future revenues shortfall."

Stoel Rives describes the alternatives the contract may take instead. There may be "a cure period if the guaranteed value is not achieved after testing" — the contractor gets a window to fix the plant before damages attach. Failing that, "either delay liquidated damages or performance liquidated damages may attach," or "the contract may require a one-time 'plant buy-down' payment" compensating the owner for accepting an underperforming asset.

The buy-down is the structure to understand, because it is the one that ends the relationship. A buy-down settles the shortfall in cash, the owner accepts the plant as built, and the contractor's performance obligation is discharged. The number must therefore carry the entire remaining life of the underperformance — which makes the discount rate and the assumed life the two most consequential assumptions in the calculation, and they are rarely negotiated with the attention the LD rate receives.

Contractors seek, as Stoel Rives notes, "individual and aggregate caps on liquidated damages, usually at an agreed-on percentage of the contract price." As covered in the second post in this series, whether performance LDs share a cap with delay LDs determines what is left when a project is both late and short.

What Happens Between Provisional and Final Acceptance?

Substantial completion is certified on a short test, and the long-term performance obligation is tested much later — often long after the contractor has been paid, demobilised and released from site.

Solar Best Practices puts final acceptance testing "between PAC and FAC" across a standard "24 months" warranty period, with methodology that "should be based on long-term PR tests" rather than the limited provisional window. It also notes an important adjustment: "the guaranteed PR (and therefore the guaranteed energy) takes into account any event causing non-production due to periods of plant downtime."

Three consequences follow.

Substantial completion is granted on thin evidence. A 7–15 day test stops the delay LD clock, releases retainage and transfers care, custody and control. The obligation that actually protects twenty years of revenue is tested two years later.

Recovery depends on surviving security. By final acceptance, retainage has usually been released and the performance bond may have stepped down. What remains is whatever the contract preserved for the warranty period — which is why the security package covered in post 2 matters as much for performance as it does for delay.

Downtime adjustments cut both ways. Excluding downtime from the guaranteed PR is technically correct — the contractor should not be liable for an outage caused by the grid. It also means a plant with poor availability can still pass its PR test while producing materially less energy than the model assumed. Availability is a separate guarantee, and it is the subject of the next post in this series.

How Do You Model Performance Shortfall in Excel?

Size the covered shortfall and the uncovered shortfall side by side. The comparison is the output that matters, and almost no model produces it.

The inputs

All assumptions, labelled as such:

P50 annual energy                 250,000 MWh
PPA price                         $45.00 /MWh
Guaranteed PR                     80.0%
Measured PR                       77.5%
Asset life for buy-down           20 years
Discount rate                     7.0%
EPC contract price                $90,000,000
Performance LD cap                10% of contract price
P90 energy (resource downside)    8% below P50

The covered shortfall — a PR miss

PR_Shortfall_Relative = (80.0% − 77.5%) / 80.0%          = 3.125%
Energy_Shortfall      = 250,000 × 3.125%                  = 7,813 MWh
Annual_Revenue_Loss   = 7,813 × 45                        = $351,563

Annuity_Factor        = (1 − 1.07^−20) / 0.07             = 10.594
Buy_Down_NPV          = 351,563 × 10.594                  = $3,724,000

Performance_LD_Cap    = 90,000,000 × 10%                  = $9,000,000

Covered, comfortably. The claim is $3.72m against a $9m ceiling, and there is a contractor on the other side of it.

The uncovered shortfall — a poor resource year

Resource_Shortfall    = 250,000 × 8%                      = 20,000 MWh
Annual_Revenue_Loss   = 20,000 × 45                       = $900,000
NPV_Equivalent        = 900,000 × 10.594                  = $9,535,000

Recoverable_From      = nobody

Two and a half times the PR claim, and no counterparty. The plant passed every test.

The comparison that should be on the cover page

Covered_Exposure      = MIN(Buy_Down_NPV, Performance_LD_Cap)   = $3,724,000
Uncovered_Exposure    = NPV_Equivalent                          = $9,535,000
Coverage_Share        = 3,724,000 / (3,724,000 + 9,535,000)     = 28%

Twenty-eight percent of the combined performance exposure has a contract behind it. Sponsors routinely spend weeks negotiating the cap on the first number and no time at all quantifying the second.

ℹ️ Note: The resource figure is not a defect and it is not a claim — it is distribution risk, which is what reserve accounts, DSCR headroom and P90-based sizing exist to absorb. The error is leaving it out of the exposure summary because it has no counterparty to invoice.

To run the full version — PR shortfall against your own yield assessment, a buy-down discounted on the actual debt tenor, and the resource band modelled as a distribution rather than a single downside — prompt Dezzmond with your test results and base case and it will build both sides of the comparison.

What Do Lenders Actually Check?

Lenders separate what the contractor owes from what the resource might do, because only one of those improves with negotiation.

• Is the guaranteed PR measured against the same yield assessment the model uses? A mismatch between the assessment behind the guarantee and the one behind the base case is a silent basis difference.
• Is there a minimum PR below which the plant can be rejected? A guarantee with no floor means an arbitrarily bad plant is still deliverable at a price.
• How is the buy-down discounted, and over what life? These two assumptions move the settlement more than the PR shortfall does.
• Do performance and delay LDs share a cap? A late-and-short project draws both from the same pot.
• What security survives to final acceptance? Retainage is usually gone; the question is what replaces it.
• How does the base case sit against P90? The uncovered exposure, sized honestly.
• Is downtime excluded from the guaranteed PR? If so, availability needs its own guarantee — and its own damages.

Frequently Asked Questions

What is the difference between a capacity test and a PR test?

A capacity test confirms the installed equipment reaches its promised rating, calculated from the peak powers of the installed modules. A PR test measures how efficiently the plant converts the irradiance it actually received, validated against the pre-construction yield assessment. One checks the hardware, the other checks the conversion.

Does passing a PR test mean the plant will hit its revenue forecast?

No. PR is weather-corrected by design, so it isolates plant efficiency from the resource. A plant can record an excellent PR in a poor solar year and produce materially less energy than the base case assumed.

What is a plant buy-down?

A one-time payment settling underperformance in cash, after which the owner accepts the plant as built and the contractor's performance obligation ends. It must capitalise the entire remaining life of the shortfall, so the discount rate and assumed asset life drive the number.

Who bears the risk that the resource is worse than forecast?

Equity, then debt. No EPC contractor guarantees the weather. That risk is managed through P90-based debt sizing, DSCR headroom and reserve accounts rather than through a claim.

Does a P90 apply to a performance test?

Only the one-year P90 does. A guarantee is tested in a single year and carries full interannual variability; a debt facility is sized over fifteen, where most of that variability has averaged out. On a wind project the two differ by about three percentage points.

What covers curtailment risk?

Neither the performance guarantee nor the yield assessment. Conversion is the contractor's, resource is a probability, and whether the network accepts the output is a third risk with no counterparty in either document.

When is the real performance obligation tested?

Typically across the warranty period — commonly around 24 months — using long-term PR testing, rather than at the short provisional test that certifies substantial completion. That timing gap matters because much of the contractor's security has been released by then.

Closing: Two Guarantees and One Forecast

The performance regime in an EPC contract is precise about what it covers and silent about what it does not. Capacity confirms the equipment. PR confirms the conversion. Both are real obligations, both are measurable, and both have a contractor standing behind them with a capped but genuine liability.

The annual energy number — the one the debt was sized on, the one the equity return depends on, the one everybody quotes — is a forecast standing between those two guarantees, protected by neither. In the worked example above it carries more than twice the exposure of the contractual claim.

None of that is an argument for negotiating the performance guarantees less hard. It is an argument for putting both numbers on the same page, so the effort goes where the money is rather than where the counterparty is.

The next post takes availability: the guarantee that sits in the O&M agreement rather than the EPC, why downtime excluded from a PR test has to be caught somewhere, and how the bonus and penalty band around an availability target actually pays out.

Sources: Stoel Rives — The Law of Solar: Design, Engineering and Construction Agreements · Solar Best Practices — System Commissioning · SgurrEnergy — Preparing a Bankable EPC Contract