A yield model that survived exactly one real tape

Say an analyst spends about six weeks building a spreadsheet for performing notes: remaining term, rate, purchase price as a percent of balance, yield to maturity, and a column stressing collateral value down 20 percent to see what recovery looks like. Clean, and easy to be proud of. Then a real tape shows up. Nineteen loans. Half the fields the model was built around are not on it. No origination date on eleven of them, so seasoning cannot be computed without asking. Two have a payment amount that does not match the rate and balance, which turns out to be because they were modified and nobody put the mod on the tape. Property values are listed as "BPO" with a number and no date, and one of those dates, once tracked down, is from 2019. The model wanted inputs the market does not hand over. The real work is the six emails asking for origination dates and mod documents and a current value opinion, the eleven day wait, and then finding out four of the nineteen have already sold. The lesson for anyone still in the analysis stage: the spreadsheet is the easy half, and it gets done first because it feels like progress, when the actual work is chasing the documents the tape never included.