The valuation model is a factor model, a concept well known in finance. The difference is that it is a conditional neural network, not an ordinary linear regression model.
As in all factor models, it relies on market factors which we call benchmarks. These are liquid and observable financial instruments or other reference data that provide signals on the price of carbon.
For instance, EUAs (European Union Allowances) are one of the benchmarks, which clearly provides signals on the price of carbon in at least one of the global compliance markets: the EU ETS.
Another benchmark is the daily carbon dioxide concentration in the atmosphere. The rationale is that if carbon dioxide concentration increases, it means that emissions are increasing and more demand for carbon offsets should happen in line with the climate change mitigation goals.