Skip to content
Igor Sabodakha

Wind Farm Investment Calculator · Methodology

How the wind farm model works

This page documents what the calculator computes, in which order and with which formulas. The wind farm is fictional; every input comes from a public source or is a documented assumption — both are listed on the sources page, checked on 30 Sep 2026. The numbers below come from the same engine that runs in the calculator.

Illustrative calculation — not investment, tax or legal advice. The wind farm is fictional.

The base case at a glance

Wind farm
5 × 6.3 MW = 31.5 MW in Hesse, outside the Südregion
Energy yield
Site quality (Gütefaktor) 68% → 2,439 full-load hours (P50, net) → 76.8 GWh a year
Tender
Award price (Zuschlagswert) 4.79 ct/kWh, average of the round of 1 Aug 2026, announced on 17 Sep 2026 → applicable value (anzulegender Wert, AW) 6.30 ct/kWh
Timeline
Financial close 1 Jan 2027 → 18 months of construction → commercial operation 1 Jul 2028 → end of life 1 Jul 2053 (25 years)
Investment
1,814 €/kW net of VAT, €57.1m; total uses including financing costs, reserves and start-up liquidity €59.3m
Debt
KfW programme 270, 5.35% fixed, 20 years from close with 3 grace years, 68 equal quarterly instalments from 31 Mar 2030 to 31 Dec 2046; €20.6m (35% of uses)
Tax
GmbH & Co. KG (a GmbH as a switch), trade-tax multiplier (Hebesatz) 400%

1. Principles

  • Deterministic and nominal. The same inputs always give the same result. Amounts are nominal euros; unit prices are entered in 2025 or 2026 money and indexed with inflation. Two items are nominal by nature: power futures, and the AW, which is not indexed (see section 4 for its reviews).
  • Dated cash flows. Construction runs monthly with cash flows at month end; operations run by calendar year with cash flows at 31 December. The first and the last year count only the days in operation. The loan runs on a monthly grid with quarterly instalments, added up by calendar year.
  • Returns from dates. IRR and NPV use the exact dates with an Actual/365 day count, like XIRR and XNPV in Excel, so a half first year and monthly construction are weighted correctly.
  • Two views. The project view (unlevered) shows what the wind farm earns; the owners’ view (levered) shows what reaches the equity after debt service and reserves.
  • Checked inputs. Before any calculation — from the input panel, a shared link or a stored session — every input is checked for range and consistency. Inputs that cannot be calculated produce a message instead of a result.

2. Timeline

  • Financial close on 1 Jan 2027 is also the valuation date. After 18 months the farm starts commercial operation on 1 Jul 2028, so the first operating year has six months.
  • The award was announced on 17 Sep 2026. It lapses if the farm is not running 36 months later, on 17 Sep 2029 (§ 36e EEG), and commissioning more than 30 months after the announcement triggers a penalty (§ 55 EEG). The base case commissions after about 21 months; later dates are flagged by the checks.
  • EEG support lasts 20 years from commissioning (§ 25 EEG; for tendered plants it is not extended to the end of the year). It is extended by the negative-price periods counted in the commissioning year and the 19 calendar years after it, rounded up to whole days (§ 51a EEG). With 6.5% of the time at negative prices, support ends on 7 Oct 2049.
  • The operating life is 25 years (30 as a variant); the farm is dismantled at the end, on 1 Jul 2053.
  • The KfW loan runs 20 years from the commitment, which the model sets at financial close. For 3 years it pays interest only; then it is repaid in 68 equal quarterly instalments, the first on 31 Mar 2030 and the last on 31 Dec 2046 (KfW Merkblatt 270). The grace years must last at least until commissioning: the model does not cover instalments during construction.
  • The KfW drawdown period is 12 months from the commitment and can be extended by up to 24 months. An 18-month construction needs such an extension, which the model assumes.

3. Energy yield

Capacity          P       = turbines × turbine rating
Site yield        h_site  = site quality × reference yield           (Anlage 2 Nr. 7 EEG)
P50, net          h_P50   = h_site × min(1, availability / 98%) × (1 − other losses)
Output            E_y     = P × h_P50 × f_y × (1 − degradation)^n_y
Output sold       E_sold  = E_y × (1 − share of output at negative prices)
P90               E_P90   = E_P50 × (1 − 1.2816 × σ)
  • The reference yield of the 6 MW class is 3,623 full-load hours at the 100% reference site (Deutsche WindGuard, 2025). At a site quality of 68% the site yield is 2,464 hours.
  • Site quality is an input, not a result: it sets both the output and the correction factor, as the law does. A better site produces more energy but receives a lower AW.
  • By law the site yield already excludes wake losses, up to 2% unavailability, electrical losses and curtailment required by the permit. The model deducts only the shortfall of availability below 98% (97% in the base case) and optional other losses: 2,439 hours.
  • fy is the share of the year in operation, ny the calendar years since the year of commissioning; degradation is 0.2% a year.
  • In periods with negative prices the direct marketer curtails the farm and no premium is paid (§ 51 EEG). 5% of the output falls into such periods in the base case.
  • Grid curtailment under Redispatch 2.0 is compensated (§ 13a EnWG) and is not deducted.
  • P90 uses the standard deviation of the yield: σ = 12.8% for a single year (P90 = 83.6% of P50) and 10.3% for a ten-year average (86.8%). The one-year value tests debt service in a bad year; the ten-year value describes a weaker wind resource.

4. Revenue

Correction factor  KF      = table of § 36h EEG, linear in between
Applicable value   AW      = award price × KF, rounded to 2 decimals in ct/kWh    (§ 36h (5))
Base price         B_y     = power futures 2027–2029, then the long-term price (2026 money) × price index
Market value       JW_y    = B_y × wind capture factor         (annual market value of onshore wind)
Market premium     MP_y    = max(0, AW − JW_y)                 (Anlage 1 Nr. 4 EEG)
Market revenue             = E_y × JW_y × support share of the year
Premium revenue            = E_sold × MP_y × support share of the year
After support              = E_y × JW_y, or E_sold × PPA price, × rest of the year
Site qualityCorrection factor
50% *1.55
60%1.42
70%1.29
80%1.16
90%1.07
100%1.00
110%0.94
120%0.89
130%0.85
140%0.81
150%0.79

* 50% applies only in the Südregion; elsewhere the factor stays at 1.42 below 60%. At 68% the factor is 1.316, so the AW is 4.79 × 1.316 = 6.30364, rounded to 6.30 ct/kWh.

  • Why the two revenue parts use different volumes. The annual market value averages the price over all wind output, including hours with negative prices. A curtailed farm only gives up zero or negative prices, so output × market value is a cautious estimate of its market revenue. The premium, by law, is zero in negative-price periods, so it applies only to the output sold.
  • Annual market value. For awards after 1 January 2023 the premium is based on the annual, not the monthly, market value. The one simplification: the farm’s own capture price equals the market value of all onshore wind.
  • Base-case prices. Futures of 128.37 / 98.88 / 82.74 €/MWh for 2027–2029, then 75 €/MWh in 2026 money plus inflation, times a capture factor of 0.80: a market value of 66.2 €/MWh in 2029 and 72.1 €/MWh in 2035. The long-term price is the model’s most influential assumption; the tornado shows its weight.
  • The EEG as a floor. The AW (63.0 €/MWh) stays below the expected market value in every year, so the base case receives no premium. In the downside case, with prices 20% lower, the premium is paid in 15 years — the floor then works as insurance.
  • Site-quality reviews. The AW is fixed in nominal terms, but § 36h (2) EEG re-determines it from the start of years 6, 11 and 16 from the site yield of the five years before, and settles the payments of those years if the site quality moved by more than 2 percentage points. The base case assumes that the assessed site quality of 68% holds for the whole period, so the AW stays at 6.30 ct/kWh. The multi-year stresses apply the review: with the ten-year P90 yield the site quality is 59.0%, the AW becomes 6.80 ct/kWh from 1 Jul 2033, and the first five years are settled at that value. Interest on paybacks to the grid operator is not modelled.
  • Payment timing. The grid operator pays monthly advances, due on the 15th of the following month. § 26 EEG allows them to be based on the previous year’s market value, which the model assumes. The rest of the premium and the § 6 refund come with the final settlement, assumed on 31 March of the next year. The settlement lag is an input, but the model runs in calendar years: a lag of 1 to 12 months puts the final settlement into the next year, 13 to 24 months into the year after — that is the delay stress — and the month within the year does not change the annual cash flows. So in the first year the market value falls below the AW, the advances are still zero and the whole premium arrives the year after; the lender’s case contains the same lag. December’s advance, paid on 15 January, is open at a year end only while the support period runs in December. Market sales are collected after 30 days of revenue, counted on the actual operating days of a part year.
  • Two-sided premium (switch). A simplified stress, not the calculation of the EEG 2027 draft, which adds further rules such as quarter-hour adjustments and a minimum amount the operator keeps: the premium turns negative when the market value exceeds the AW. It is off by default — a 2026 award falls under EEG 2023, and whether the draft will apply to it is not settled. Advances are never negative; a payback is settled with the final settlement.
  • After support the farm sells at the market value or, as a switch, under a PPA at a fixed price in 2026 money. The PPA is assumed to have a negative-price clause: the farm is curtailed in negative-price periods and paid only for the output sold.

5. Operating costs

Fixed items        = €/kW a year (2025 money) × capacity × price index × f_y
Land lease         = max(10% of revenue, €70,000 per turbine in 2026 money)
Direct marketing   = 0.20 ct/kWh × output sold            (2026 money, indexed)
Municipal payment  = 0.2 ct/kWh × output                    (§ 6 EEG)
Guarantee fee      = 1.25% a year × decommissioning security
Generator grid fee = €/kW a year                              (0 in the base case)
€/kW a year, 2025 moneyYears 1–1011–2021+
Maintenance (full service)141617.6
Technical and commercial management455
Insurance111
Other666
  • The direct-marketing fee of 0.20 ct/kWh is an assumption. The statutory deduction of 0.228 ct/kWh for plants after support (Netztransparenz, 2026) is a reference point, not a market offer for a new farm.
  • The municipal payment is voluntary under § 6 EEG. The grid operator refunds up to 0.2 ct/kWh in the next year’s final settlement for quantities that received a premium — in the model, in years with a positive premium; any payment above 0.2 ct/kWh is a plain cost. After support it continues as a plain cost (switch).
  • The decommissioning security follows the Hessian rule — hub height × €1,000 per turbine — and is provided by a bank guarantee.
  • Decommissioning costs 50 €/kW in 2026 money at the end of life. The cash is set aside in equal instalments over the last 5 years.
  • Inflation follows the Bundesbank projection for 2026–2028 (2.9% / 2.7% / 1.9%) and then the ECB target of 2%.

6. Investment and funding

Capex item (net of VAT)€/kWPayment profile
Turbine incl. transport and installation1,16310% at close, 70% on delivery (4–2 months before COD), 20% at commissioning
Foundation8840 / 40 / 20% over the thirds of construction
Roads and crane pads11940 / 40 / 20% over the thirds of construction
Grid connection13340 / 40 / 20% over the thirds of construction
Development and permits163at financial close
Compensation measures35at financial close
Other60evenly over construction
Contingency (3%)53evenly over construction
Total1,814€57.1m
  • Costs are Deutsche WindGuard’s 2025 levels for projects commissioned in 2025–2028, without escalation to financial close; turbine prices were stable in 2026.
  • VAT of 19% is paid with each invoice and refunded 2 months later. A VAT bridge loan at the senior rate plus 0.85 percentage points funds the gap; its interest, including the months after commissioning, is part of the uses.
  • Financing costs during construction: an upfront fee of 1% of the loan at close; a commitment fee of 0.15% a month on the undrawn amount from month 7 (the KfW rule); interest on drawn debt. KfW charges only interest in the grace years, so this interest is paid in cash and funded like capex rather than added to the loan.
  • The debt service reserve account (DSRA) is funded at commissioning with 3 months of the debt service of the first full repayment year (2030): €573k.
  • Start-up liquidity of €498k funds the receivables of the first operating year, so the first months’ sales need no extra cash.
  • Debt and equity are drawn pro rata each month (equity first as a switch).
Uses    = capex + upfront fee + commitment fee + interest during construction
          + VAT-loan interest + initial DSRA + start-up liquidity
Sources = senior debt + equity                                   check: sources − uses = 0

In the base case the uses of €59.3m are capex €57.1m, financing costs €1.1m, the initial DSRA €0.6m and start-up liquidity €0.5m; they are funded by debt of €20.6m and equity of €38.7m.

7. Debt sizing

The model’s lender’s case counts only the guaranteed floor during support: at most the AW per kWh sold. This is a conservative convention of the model, not a statement of what every lender does. The premium and the § 6 refund arrive with the same timing as in the operating case, including the lag of § 26; after support the case uses base prices (a switch sizes on base prices throughout). Cash flow available for debt service has one definition for sizing, covenant and lock-up:

CFADS_y   = EBITDA_y − change in working capital_y − taxes_y
Debt service per euro of loan   a_y   (quarterly instalments and monthly interest, summed by year)
Loan from the DSCR targets      D_DSCR = min over y of  min( CFADS_y[lender, P50] / (1.20 × a_y),
                                                           CFADS_y[lender, P90 1-year] / (1.00 × a_y) )
Loan                            D      = min( D_DSCR, 80% × uses )
  • Circularity. Taxes depend on the interest, and the uses depend on the loan through fees, interest during construction and the DSRA. The model iterates until the loan changes by less than €0.50, then recalculates both lender cases with the final loan and checks both targets again.
  • Variants. Annuity and sculpted repayment are generic commercial profiles, not KfW 270. Annuity: equal quarterly payments. Sculpted: each year’s debt service is the lower of CFADS ÷ target for P50 and P90; the loan is the amount those payments repay exactly by maturity — found in closed form, since balances are linear in the loan — and smaller if a weak year would otherwise need a negative instalment.
  • Base case. The binding constraint is the P90 DSCR on the EEG floor: debt of €20.6m, 35% of uses; the lowest lender DSCRs are 1.27x at P50 and 1.00x at P90. Community wind farms financed at lower rates and higher award prices show 79–90% debt in their prospectuses; at 5.35% and a floor of 6.30 ct/kWh the debt capacity is far lower. In September 2026 a group of 18 banks warned of rising equity requirements for wind projects.
  • Sizing target and covenant. The one-year P90 target of 1.00x sits below the covenant of 1.05x. The target applies to the lender’s case at floor prices; the covenant is tested on the operating cash flows, where the one-year P90 stress in the base case still shows 1.14x. Neither threshold has a published market benchmark; both are assumptions and inputs.
  • No loan. If the lender’s case has a year without cash for debt service — CFADS at or below zero — no amount of equal instalments fits, and the model sizes no loan. A sculpted loan is repaid before such a year where the earlier years allow it. Loans below €1 count as none, and years with less than €1 of debt service have no DSCR.

8. Taxes

EBT           = EBITDA − depreciation − interest − increase of the decommissioning provision
Add-back      = 25% × max(0, interest + 50% × land lease − €200,000)           § 8 Nr. 1 GewStG
Trade income  = EBT + add-back − loss carry-forward (€1m in full, 60% above)     § 10a GewStG
                rounded down to €100, less €24,500 for a partnership              § 11 GewStG
Trade tax     = 3.5% × trade income × multiplier (Hebesatz)
Corporate tax = rate_y × (EBT − loss carry-forward: €1m in full, above it
                70% to 2027, 60% from 2028)            GmbH only; § 23 KStG, § 10d EStG
                rate_y: 15% to 2027; 14 / 13 / 12 / 11% in 2028–2031; 10% from 2032
Solidarity    = 5.5% × corporate tax
  • Depreciation. The base is capex plus the capitalised financing costs. All wind-farm assets are depreciated straight-line over 16 years (the German depreciation table; BFH IV R 46/09); the first year counts from the month of commissioning. Declining-balance depreciation (3 × straight-line, at most 30%: 18.75%) is a switch that applies only to assets completed between 1 July 2025 and 31 December 2027 — not to the base case — and changes to straight-line once that is higher. Any book value left when the farm is dismantled is written off in the final year.
  • Decommissioning provision. For tax, the obligation is accrued pro rata over the operating life at the prices of each balance-sheet date and discounted at 5.5% — except when less than twelve months remain (§ 6 (1) Nr. 3a e) EStG). Its increase reduces taxable profit. The cash reserve is separate.
  • Taxes on profit are not deductible (trade tax: § 4 (5b) EStG; corporate tax and solidarity surcharge: § 10 no. 2 KStG), so the tax calculation itself has no circularity. Taxes are paid in the year they arise.
  • Legal form. On its profit, a GmbH & Co. KG pays only trade tax; its partners are taxed on their shares of the profit, which is not modelled. The equity IRR of the KG is therefore before the partners’ taxes and is not directly comparable with the GmbH, which also pays corporate tax and the solidarity surcharge.
  • Trade tax goes to the municipality of the turbines: for a wind farm, § 29 (1) Nr. 2 GewStG splits the tax base 9/10 by installed capacity and 1/10 by payroll, so with one site and no staff of its own the whole base goes there and one multiplier applies. The checks flag a multiplier below the legal minimum of 280% from 2027, and interest of €3m or more a year, at which the exemption limit of the interest barrier (§ 4h EStG) no longer applies; the model does not calculate the barrier itself.

9. Cash waterfall

CFADS           = EBITDA − change in working capital − taxes
− interest − principal
  shortfall     covered first by cash held back under the lock-up, then by the DSRA
± DSRA          top up to 3 months of next year’s debt service, release the excess, draw on a shortfall
− reserve       equal instalments for decommissioning in the last 5 years
  lock-up       if DSCR < 1.10 during the loan term, the year’s cash stays in the company
                until the DSCR is back above the threshold
= cash available to equity
  • In the final year the DSRA, any cash held back and the reserve are released and decommissioning is paid. If the final year’s cash falls short, the case counts as not funded like any other shortfall: payments by the owners are not modelled.
  • Funding. If cash still falls short during the life, the model carries the shortfall as negative cash and marks the case as not funded: the equity IRR and NPV are then shown as “n.m.” (not meaningful). Payments by the owners to cure a shortfall are not modelled.
  • Limits on distributions. The last line is cash available to equity before legal limits on distributions. A GmbH may not pay out assets needed to keep its registered capital (§ 30 GmbHG); a limited partner of a KG whose withdrawals push its capital account below the registered contribution is liable to creditors again up to that amount (§ 172 (4) HGB). In the base case the model’s balance sheet shows negative book equity in 2043–2050 — a sign that such limits could bite, not a legal test. Shareholder loans, whose repayment is not a distribution, are not modelled.

10. Results

MeasureDefinitionBase case
Equity IRRXIRR of the owners’ monthly contributions during construction and the annual cash available to equity; n.m. when the company runs out of cash3.09%
Project IRRXIRR of capex (net of VAT and financing costs), start-up liquidity and EBITDA − change in working capital − decommissioning; after tax, taxes are recomputed without interest4.04% pre-tax, 3.67% after tax
Equity NPVXNPV of the owners’ flows at the cost of equity (8%), discounted to financial close−€16.1m
LCOEIn the style of Fraunhofer ISE: (PV capex + PV opex net of the § 6 refund + PV decommissioning) ÷ PV output sold; real 2026 money at a real WACC of 3.9% (nominal: 5.8%), without taxes and financing costs. Unlike ISE it uses output sold and a lease tied to revenue, so it moves with the power price7.25 ct/kWh (nominal 9.22)
DSCRCFADS ÷ (interest + principal) in each year of the loan; the minimum over all years with debt service, the average over the repayment years1.44x min (2030), 2.20x average
LLCRPV of CFADS over the loan term at the loan rate ÷ loan, at COD — the year of maturity only until the maturity date; reserves not counted2.06x
PaybackYears from commissioning until the owners’ cumulative cash flow turns positive18.8 years
GearingLoan ÷ total uses34.8%

11. Scenarios, sensitivity and bid calculator

The base case sizes the loan. The stress scenarios keep that loan and every instalment — the lender’s view after financial close. Applying a P90 output to every year is a stress, not a 90% probability for the whole life: the one-year value tests debt service, the ten-year value is the better guide for returns.

ScenarioChanges against the base caseEquity IRRMin DSCREquity NPV
BaseP50 output, base prices and costs; sizes the loan3.09%1.44x−€16.1m
P90 1-yrLender’s stress: one-year P90 output in every year (83.6% of P50)0.61%1.14x−€23.1m
P90 10-yrTen-year P90 output in every year (86.8%), with the § 36h review1.16%1.20x−€21.5m
DownsideTen-year P90 output, market prices −20%, fixed opex and grid fee +10%, capex +5% paid by the owners, with the § 36h review−2.12%1.18x−€28.8m

Tornado. Each driver is moved to its low and high value with the loan re-sized — the view before financial close — and the drivers are sorted by the swing of the chosen measure. A run in which the company runs out of cash, or a calculation check fails, shows no equity IRR. The negative-price driver also moves the capture factor with (1 − share): output that moves into negative-price periods earns about nothing, so the market value of wind falls with it.

DriverLowHigh
Site quality60%76%
Long-term power price60 €/MWh90 €/MWh
Wind capture factor0.720.86
Award price4.49 ct6.06 ct
Capex−10%+10%
Fixed opex−10%+10%
Interest rate4.40%6.05%
Inflation−0.5 pp+0.5 pp
Negative prices (market value adjusted)3%9%
Land lease6%14%
Trade-tax multiplier320%450%
Operating life25 yrs30 yrs

Bid calculator. It finds the lowest award price at which the equity IRR reaches a target, re-sizing the loan at every step, and reports three things: the price for the return alone; the lowest price that is also financeable — fully funded, within the covenant and with an award that has not lapsed under § 36e; and whether that price is within the tender ceiling of the inputs. The IRR is not monotonic in the award price: a higher floor allows more debt at 5.35%, which can lower the equity return. So the search scans a 0.25 ct grid for the first qualifying price and then bisects within that step. Bids carry two decimals, so the answer is the lowest such price that qualifies, re-run at that price. In the base case, an equity IRR of 8% needs 7.41 ct/kWh (AW 9.75 ct/kWh) — above the ceiling of 7.25 ct/kWh.

12. Checks and validity

Every recalculation runs 25 checks in five groups. The calculator summarises them in one line under the key figures: whether the case is fully funded, meets its covenant and stays within the model’s scope.

Calculation

  • Sources of funds equal uses
  • Balance sheet balances at COD
  • Loan repaid at maturity
  • Loan matures within the operating life
  • Balance sheet balances at every year end
  • Debt sizing converged
  • The loan meets both DSCR targets in the lender’s case
  • Gearing within the maximum
  • Equity IRR exists and is unique

Funding

  • The company never runs out of cash
  • Decommissioning is paid without payments by the owners

Covenant

  • DSCR at or above the covenant in every year
  • No year with distributions held back

Inputs

  • Award price within the tender ceiling
  • Award announced before financial close
  • Trade-tax multiplier at or above the legal minimum
  • Declining-balance depreciation is allowed
  • Loan term and grace years match a KfW 270 variant
  • Drawdown within 12 months (KfW allows an extension of up to 24)

Model scope

  • Commissioning before the award lapses (§ 36e EEG)
  • Commissioning within 30 months of the award (§ 55 EEG)
  • Loan matures before the EEG support ends
  • Interest below the interest-barrier threshold (§ 4h EStG)
  • Book equity stays positive (limits on distributions)
  • One-sided premium (the two-sided option is a simplified stress)

13. Verification

  • Automated tests cover the financial maths (against Microsoft’s published XIRR and XNPV examples), the § 36h table and rounding, the loan calendar, the premium timing, taxes and depreciation, the model and its scenarios, the bid calculator and the Excel export.
  • A seeded sweep of 500 random input sets within the usual ranges runs as a test: with the one-sided premium, the base scenario never breaches its covenant, never runs out of cash and passes every calculation check. A wider sweep of 1,000 sets also moves total capex and opex, the futures, inflation, the two-sided switch and the repayment profile: every scenario passes every calculation check, and no year carries debt-service dust.
  • The Excel workbook is a second implementation of the model in spreadsheet formulas. Its copy without stored results was recalculated by Microsoft Excel alone. In 40 variants — every switch (legal form, repayment profile, lender’s basis, two-sided stress, PPA, equity first, declining-balance depreciation, Südregion, dates and lags) and the edge cases of the reviews (no loan, no cash, negative returns, a lapsed award, inputs outside the usual range) — all of its roughly 26,000 formula cells agree with the engine: money to the cent, rates and ratios to 10⁻⁷. Every pair of switch values meets in at least one variant. Two edits made in the file behave as described: a new cost of equity matches the engine in every cell; a new award price shows no error and asks for a re-solve. Two negative controls prove the comparison can fail: a trade-tax multiplier of 410 % instead of 400 %, and a corrupted trade-tax formula, each move the tax lines and the equity IRR but not revenue, the loan or capex. The last run is archived with the Excel version in the repository (verification/excel-run.json); a test fails as soon as the workbook changes, until Excel checks it again.
  • On 1 October 2026 the base-case workbook was also opened in Google Sheets: its results match the website’s, no formula shows an error, the switches accept only their listed values, and the two edits made in the file behave as in Excel.
  • The workbook takes three results from the website as pasted values: the loan, the total uses and, for a sculpted loan, the principal per year. They are the fixed points of the model’s circular links. Its Checks sheet recomputes the loan from the DSCR targets and the gearing cap and says when the pasted values no longer fit.
  • On 30 September 2026, before the reviews, a spreadsheet built from the written specification — not from the code — recomputed the base case of the first engine in Microsoft Excel. All 20 key figures and twelve annual lines agreed to the cent.
  • The PDF report of the base case is printed from the calculator’s own charts and numbers. A test fails when the published PDF is older than the code it was made from.
  • The source code, including the tests, is public on GitHub.

14. Limitations

  • Operations are annual; output is spread evenly over the year (winter is in fact windier).
  • The farm’s capture price equals the market value of all onshore wind.
  • The taxes of a KG’s partners on their shares of the profit are not modelled; taxes are paid in the year they arise.
  • One senior loan: no tranches, shareholder loans, refinancing or cash sweep; instalments during construction are not supported.
  • Not modelled: the § 55 penalty for late commissioning, interest on § 36h paybacks, the interest barrier, and payments by the owners to cure a shortfall.
  • Uncompensated grid curtailment (a draft of the grid package) and generator grid fees (the regulator’s AgNes process) are not in the base case; the fee is available as an input.
  • In the Excel workbook the loan, the total uses and the sculpted principal are solved on the website; after a change that affects them, re-solve there and download again. The tornado and the bid calculator are included as values.
  • The wind farm is fictional and the results are illustrative.