A proposal is only as good as its generation number. If it is too high, the customer feels cheated in the first summer; if it is too low, you lose the order to someone more optimistic. So we test PitchSun's estimate the way an engineer would: against independent, published tools, on Indian sites, with the same inputs.
The short answer
- The maths is right. Set up with the same methods, pvlib (the open-source PV library from Sandia and NREL) and PitchSun agree to 0.01% at every site, and hour by hour.
- The physics is close to PVsyst's. Against pvlib set up the way PVsyst models a system, PitchSun's designer simulation is +0.40% (range +0.36 to +0.43%).
- NREL SAM agrees too. Sunlight on the panels matches SAM within 0.1%. Yield is −0.38% against SAM's own fit of the same module datasheet.
- Quotes stay on the safe side. A quick quote (PVGIS with a flat 16% loss) is about 7% below the detailed simulation.
Results by city
Yearly energy in kWh per kWp. Same inputs for every tool: PVGIS 5.3 typical-year weather, 15° tilt facing south, free-standing rack, no shade, soiling 4%, LID 1.5%, mismatch 2%, DC wiring 1%, inverter 97.5%, AC wiring 1%, and the same module (Jinko JKM410M-72HL-V).
| City | PitchSun | pvlib, PVsyst-style | NREL SAM | Quick quote (PVGIS) |
|---|---|---|---|---|
| Delhi | 1659 | 1653 | 1664 | 1528 |
| Jaipur | 1654 | 1648 | 1661 | 1550 |
| Lucknow | 1638 | 1632 | 1642 | 1495 |
| Ahmedabad | 1626 | 1619 | 1634 | 1509 |
| Bhopal | 1583 | 1576 | 1588 | 1468 |
| Kolkata | 1500 | 1493 | 1496 | 1358 |
| Mumbai | 1577 | 1571 | 1593 | 1472 |
| Pune | 1545 | 1539 | 1547 | 1463 |
| Hyderabad | 1605 | 1598 | 1614 | 1462 |
| Bengaluru | 1581 | 1575 | 1584 | 1482 |
| Chennai | 1615 | 1608 | 1631 | 1459 |
| Kochi | 1389 | 1384 | 1391 | 1304 |
NREL SAM here uses its own fit of the module datasheet, the counterpart of what PitchSun does. With the CEC library's parameters for the same module, SAM comes out about 2.4% lower. Respected tools disagree with each other by around 2.5%, mostly over how a module behaves in heat and weak light. PitchSun sits inside that spread.
How the test works
- Same weather. For each city, the PVGIS 5.3 typical year: 8,760 hours of sunlight, temperature and wind.
- Same system. About 10 kWp at 15° facing south, with the same losses, module and inverter in every tool.
- Code check. pvlib set up with PitchSun's own methods (sun position, Perez sky model, glass reflection, cell temperature, losses, inverter). Any gap here would be a bug. The gap is 0.01%.
- Independent check. pvlib set up the way PVsyst works (a one-diode model of the module, Fresnel glass reflection, PVsyst's cell temperature), and NREL SAM with its own models.
The designer's hourly run fits a one-diode model to each module's datasheet, the same family of model PVsyst and SAM use. Weak-light and hot-day losses come from the module itself, not from a fixed curve. If you import a module's PVsyst .PAN file into your price list, PitchSun uses PVsyst's own parameters for that module instead.
What comes next
- PVsyst itself. We are running the same 12 cities in PVsyst and will add its column to the table above, whatever it shows.
- Real rooftops. Every night PitchSun compares monitored plants' actual output with their proposals, corrected for how sunny each month really was. We will publish that figure once at least 10 plants have 30 plant-months of data between them.