What a credit committee actually checks
A lender’s job is not to admire the 3D render. Debt service is contractual and sunshine is not, so the analyst’s entire reading of your study is a hunt for the place where you were generous to yourself. They check where the weather data came from and how many years it covers. They check whether the headline yield is a P50 or something braver. They check whether losses are itemised line by line or buried in a single round “system efficiency”. They check whether the uncertainty behind the P90 was built up from named components or back-solved to make the debt fit. And they check whether the assumptions — degradation, availability, soiling regime, tariff path — are stated where they can be challenged.
Most studies fail this reading quietly. Not because the engineering is wrong, but because the document gives the analyst nothing to verify: a yield with no provenance, a derate with no decomposition, a payback with no stated discount rate. A study like that cannot be argued with — which means it cannot be trusted, which means it gets re-done by someone the lender pays by the hour.
The methodology anchor: IEC 61724-1
IEC 61724-1 is the IEC’s standard for photovoltaic system performance. It defines the metrics — reference yield, specific yield, performance ratio — the data requirements behind them, and the discipline of stating exactly how each number was produced. A feasibility study that adopts those definitions buys itself something valuable: accountability after the fact. The performance ratio promised on page four is computed the same way the operating plant will be judged after commissioning, so the promise and the verdict speak the same language.
It also gives you a shared vocabulary with the lender’s independent engineer. When both sides mean the same thing by “performance ratio”, the review becomes a comparison of assumptions rather than an argument about definitions — and comparisons of assumptions are winnable.
P50, P90 and what lenders actually price
These are exceedance probabilities. The P50 is the annual yield you expect to exceed in half of all years — the central estimate, the coin flip. The P90 is the yield you expect to exceed in nine years out of ten — the cautious floor. Lenders size debt on the P90 (and sometimes a one-year P99 for the worst single year) for a simple reason: the loan repays in bad years too.
Concretely: take a plant with a P50 of 1,800 MWh per year and a total yield uncertainty of σ = 6%. Under a normal-distribution assumption, P90 = P50 × (1 − 1.282σ) ≈ 1,800 × 0.923 ≈ 1,661 MWh. Debt service cover is computed on the 1,661, not the 1,800. The 139 MWh between them is the price of your uncertainty — which means that reducing σ honestly, with better data and a tighter model, literally increases the debt the same plant can carry.
Documented uncertainty is the real differentiator
Anyone can print “P90” — the arithmetic is one line. The bankable question is where the σ came from. A defensible study builds it from named components: the uncertainty of the weather dataset (source, period, spatial method), interannual variability, the modelling uncertainty of the simulation chain, and the uncertainty in soiling and long-term degradation. It states each component’s value, combines them root-sum-square, and shows the result. An analyst who disagrees with one component can substitute their own number and re-run the cascade — disagreement becomes a recalculation instead of a rejection.
The same goes for the assumptions register. Degradation per year, availability, soiling by season, tariff escalation, discount rate: stated, in one place, with sources. The studies that get accepted are not the ones with the highest yield. They are the ones where every challenge has somewhere to land.
The loss cascade, line by line
Between plane-of-array irradiance and AC energy at the meter sits a chain of losses, and each link needs its own number and its own reason. Shading — as an hourly matrix across the year, not a flat percentage, because a 9 a.m. winter shadow and a noon summer shadow are not the same loss. Soiling — as a site regime, because a Cairo rooftop in dust season is not a coastal site. Temperature — a cell-temperature model times the module’s coefficient, hour by hour. Mismatch — including bypass-diode behaviour under partial shade (Deline 2013), which is non-linear and routinely underestimated by flat derates. Then DC wiring, inverter conversion, clipping at the chosen DC/AC ratio, AC and transformer losses, availability.
The cascade must reconcile: multiplied through, the lines should land on the stated performance ratio, because that multiplication is exactly what the reviewing engineer will do. This is why we simulate all 8,760 hours of a typical meteorological year with a full hourly shading matrix in the design studio — a flat-derate model cannot produce a cascade that survives the multiplication test.
Why a watermark on a quick estimate is a feature
Sales needs a number in minutes; bankable engineering takes longer. The wrong resolution is to dress the quick number up as the studied one. Our CRM produces a two-second preliminary estimate and watermarks it “preliminary — engineering review pending” — and that watermark is doing structural work. It lets the sales conversation move at sales speed without spending the engineering study’s credibility, and it means that when the full 10-section study arrives, nobody in the room confuses it with arithmetic done on a phone call. A document that admits what it is protects the document that has to be believed.
Methodology, not certification
Last, the claim discipline — because the word “bankable” is routinely inflated. A study can follow bankable methodology: IEC 61724-1 metric definitions, P50/P90 exceedance with a documented uncertainty build-up, an itemised loss cascade, a stated assumptions register. That is what ours does, and the engine behind it carries its own discipline: it grew from a Cairo engineering workbook used on 2.5+ GW of Benban-financed projects, and every release is regression-gated to ±0.5% of that validated baseline.
Third-party certification of the methodology is a different and stronger claim — and it is on the roadmap, not claimed today. Be suspicious of any tool that is vague about that distinction. What makes a feasibility study bankable, in the end, is not a badge. It is that every number in it can be checked by someone whose job is to check it.