Article
Fast is useless if it's wrong. So Riverbend is checked two ways: against itself and against a model I didn't write.
Against itself:
The workbook rebuilds the engine in live Excel formulas, with inputs only on the Assumptions tab. Then:
- CF Delta: every cash-flow line, every year, Excel minus engine. It reads zero.
- Checks tab: balance sheet, cash, debt repaid at maturity, the cash-flow match and the engine's own controls, rolled into one master flag. ALL PASS.
- Green Light Gate: 22 credit and integrity conditions. 22 of 22.
Against an independent model:
NREL's System Advisor Model, run through its own code with the same inputs. On the annuity case with 27% tax, every cash-flow line matches to the rand, and the equity IRR is 21.45% in both. On the sculpted case, the debt agrees within R350 on R505m.
And against a published South African case: Meridian Economics' 2020 levelised cost for solar PV. On the stated assumptions the engine lands at 65.4 c/kWh against 64.0 published.
Every one of these is rebuilt by a script on every run, not quoted from memory.
Next: one run to the full document set.
Riverbend Solar is an illustrative case study: a fictional deal with no real counterparty.