The full methodology is published.
Every number on a Chart Solar forecast traces back to a model, a source, and an assumption you can challenge. This page is the index; the deep-dive lives in Field Notes.
Hourly physics, not annual averages
We run an 8760 (one value per hour, per year) production model on top of NSRDB irradiance for your address. Same family of math NREL ships in PVWatts; pvlib for the engine. We re-run the year against a 30-year climatology so the variance you see is real, not assumed.
500 Monte Carlo paths on every escalator
Your installer prices a forecast at 'utility raises rates 4%/yr forever.' We don't. We sample 500 escalation paths from the historical CPI-deflated tariff record, then carry each one through the model. The output is a P10–P90 distribution, not a single headline number.
Capital opportunity cost is in the model
The chart shows solar against a high-yield-savings opportunity-cost line. Decision Pack adds mortgage prepayment, S&P 500 (with sequence-of-returns risk), and 30-year-treasury comparisons. The honest question isn't 'does solar pay back' — it's 'does it pay back better than the next-best thing the same capital could do.'
What we don't know is documented
Roof shading without a site visit is partial. Snow loss varies by microclimate. NEM 3 export rates are recent enough that the historical regression is thin. Each report calls these out by name (KNOWN / PARTIAL / UNKNOWN), with a confidence band on the headline.
The full per-source attribution (NSRDB year, pvlib version, tariff dataset, climatology window) lands with Phase 1b — driven by the engine itself rather than written by hand. For now, this overview plus the Field Notes archive is the canonical reference.