Operating solar diligence: P50/P90, losses and actual generation
Reconcile solar P50/P90, weather, meter data and incremental losses, then connect recoverable output to repair funding and acquisition cash flow.
Jurisdiction: Ukraine. Sources reviewed . Evergreen framework. Next scheduled editorial review: . Material rule changes require earlier review.
An operating solar plant should be priced against a reconciled explanation of its output. Start with the installed configuration, a consistent measurement boundary and a weather-adjusted model. Then separate resource variation, equipment performance, external restrictions and data uncertainty. Last year's MWh alone cannot establish sustainable generation or the cost of restoring it.
This framework concerns technical diligence for an acquisition of a Ukrainian plant. The numerical cases are invented teaching examples, with no site coordinates or private operating records. Engineering findings must ultimately connect to the acquisition price, completion budget and lender's cash-flow definition.
Establish one configuration and one data boundary
Record DC module capacity in MWp, AC inverter capacity in MW and permitted export separately. Identify replacements, repowering, shared transformers, auxiliary demand and any battery. A model for the original configuration cannot be compared directly with a later expanded site. State the model version and the dates on which equipment and limits changed.
Reconcile commercial-meter export with inverter/SCADA readings and settlement records. These sources can measure different points or use different corrections. Preserve raw interval data, time zone, interval duration, units and quality flags; distinguish measured values, estimates, missing observations and genuine zeros. Align daylight-saving transitions and meter replacements before aggregating annual energy.
Ask an independent engineer to connect the following evidence rather than review each file in isolation:
Table — scroll horizontally if needed
| Evidence | What to reconcile | What can distort the conclusion |
|---|---|---|
| As-built inventory and single-line diagram | DC/AC capacities and actual export boundary | Comparing different configurations |
| Meter and SCADA exports | Timestamp, units, interval length and calibration/change records | Treating missing values as zero generation |
| Irradiance and weather series | Plane-of-array resource, sensor condition and reference dataset | Blaming equipment for a less sunny year |
| Model and loss assumptions | Version, baseline losses and temperature treatment | Deducting the same loss twice |
| Fault, work-order and restriction logs | Start/end, affected capacity and event evidence | Assigning one interval to several loss categories |
| Invoices, settlements and receipts | Exported energy versus sale and collection | Calling unpaid revenue a technical production loss |
Keep a coverage report showing the share of intervals that are measured, estimated or unusable. A complete-looking annual total can contain material estimation. An information gap affects confidence; it does not automatically prove that equipment failed.
Read P50 and P90 as conditional distributions
P50 and P90 describe exceedance probabilities for a defined energy estimate and horizon. P90 is not 90% of P50. The NREL uncertainty report distinguishes resource, modelling and reliability uncertainty. A P-value is meaningful only with its assumptions and uncertainty method attached.
Specify whether the assessment concerns one future year or the average over several years. Weather variability and persistent modelling bias do not necessarily reduce in the same way over a longer horizon. Obtain the loss assumptions, correlations and treatment of uncertainty rather than accept a spreadsheet haircut labelled "P90".
For a deliberately simplified normal distribution, suppose P50 equals 10,500 MWh and total annual standard deviation is 5%, or 525 MWh. The illustrative lower tenth percentile is 10,500 − 1.2815516 × 525 = 9,827.19 MWh. Normality and the 5% input are assumptions, not findings about a Ukrainian plant.
If that 5% comprises independent annual weather variability of 4% and a persistent uncertainty component of 3%, the ten-year-average standard deviation in this teaching model is sqrt(4%² / 10 + 3%²) = 3.2558%. Its P90 average is 10,061.89 MWh per year. Hold central yield constant here solely to isolate uncertainty; an operating forecast also needs degradation and changing losses. Dividing the whole 5% by sqrt(10) would incorrectly average away the persistent component. Real assessments need appropriate distributions and correlations; this calculation is not a bankable yield study.
Reconcile the weather effect before diagnosing equipment
Performance ratio can help investigate the gap. Sandia's PVPMC definition relates final AC yield to reference yield and notes temperature sensitivity. Define the DC reference capacity, irradiation and AC boundary consistently; compare like conditions before drawing conclusions from a changing PR.
Assume an invented 10 MWp DC / 8 MW AC plant has baseline annual export of 10,500 MWh at plane-of-array irradiation of 1,400 kWh/m². Its simplified PR is 10,500 / (10 × 1,400) = 75.00%, using reference irradiance of 1 kW/m². The observed year's irradiation is 1,260 kWh/m². For this example only, hold every other weather effect constant and scale output linearly: expected export becomes 9,450 MWh. An engineering replay would also examine temperature and the actual operating model.
Table — scroll horizontally if needed
| Energy bridge | MWh | Meaning in this invented case |
|---|---|---|
| Baseline at reference-year resource | 10,500 | Defined configuration and baseline loss assumptions |
| Resource effect | -1,050 | Lower irradiation under the simplified scaling |
| Expected at observed resource | 9,450 | Starting point for incremental loss attribution |
| Additional grid restriction | -150 | Separately evidenced external event |
| Additional equipment unavailability | -180 | Supported by fault/work-order records |
| Additional soiling loss | -120 | Increment above the baseline allowance |
| Actual commercial-meter export | 9,000 | Reconciled annual output |
The total shortfall from the reference model is 1,500 MWh, but the residual after resource adjustment is 450 MWh, or 4.76% of 9,450 MWh. Actual PR is 71.43% at the same DC reference. These ratios identify questions; they do not by themselves establish the three causes in the table. Those allocations are assumed here and require independent event evidence in a real case. In particular, additional soiling must not be deducted on top of an identical loss already present in the model.
Translate recoverable output into cash and funding
Separate repairable loss from restrictions expected to remain. Suppose the 180 MWh equipment loss and 120 MWh additional soiling can be eliminated, while the 150 MWh grid restriction continues. At the same weather conditions, output would rise from 9,000 to 9,300 MWh. Do not add a second resource recovery to this scenario.
Assume a net realised contribution of EUR 75/MWh after energy-linked deductions, annual fixed costs/tax/reserve deductions of EUR 250,000 and annual debt service of EUR 350,000. These are teaching inputs. Simplified cash flow available for debt service is initially 9,000 × 75 − 250,000 = EUR 425,000: DSCR 1.21×. Recovering 300 MWh while adding EUR 5,000 of annual maintenance gives EUR 442,500 and 1.26×.
If the repair also costs EUR 120,000 in cash, funding its implementation matters. Under an illustrative covenant definition deducting that expenditure in the repair year, available cash falls to EUR 322,500, DSCR 0.92×, leaving a EUR 27,500 debt-service shortfall. A lender may use a different definition or dedicated funded account. Align the actual covenant, reserves and repair financing; an improved steady-year ratio does not finance the intervention itself.
Put the unresolved findings into the acquisition terms
Request an engineer's reconciliation with data-quality limits, causes, confidence ranges, repair scope, cost, timing and residual output risk. Ask which warranties or service claims are assignable and collectible, rather than treating estimated lost MWh as compensation. Run the price model with the unresolved downside and the actual expected cash profile.
The principal red flags are incompatible meter boundaries, an unexplained model change, P90 without a horizon or method, missing intervals treated as zero, overlapping loss deductions and a repair uplift without its cost or downtime. Allocate each open issue to a document, responsible party and decision before the binding capital commitment. A Passport can organise those findings; it does not certify the yield.
Read the operating solar acquisition guide, curtailment and valuation framework and solar valuation guide. Discuss your mandate with the production evidence and repair funding your team requires before underwriting output.