A repeat buyer of Section 8 single-family rentals underwrote every deal by hand, which meant each read cost an evening and no two reads used quite the same assumptions. The volume made the inconsistency expensive before it made the hours expensive.
The tool takes an address and a price. It returns a populated rental proforma, a one-page decision memo, and a provenance record naming the source behind every load-bearing number. The first working version shipped the same day the job was awarded.
The suppression rule
Every figure the tool relies on carries a provenance record holding its value, its source, a confidence level, the number of retrieval attempts, and any reconciliation between competing sources. Rent comes from the published HUD fair market rent schedule for the county and bedroom count. Taxes come from the assessor record. When a figure cannot be sourced, the tool reports the gap and withholds its verdict rather than substituting an estimate.
A read the buyer cannot trace never reaches the buyer.
EXHIBIT 1 · INTAKE TO VERDICT
The package
- A command-line runner and a browser interface over one underwriting engine
- A local database of HUD fair market rents, rebuilt on the annual publication cycle
- Per-deal output folders holding the proforma, the memo, and the provenance record for every read
- A regression suite that runs the whole engine with the network disabled, so a change to the code cannot quietly change a verdict
The transferable point
An automated read is worth having only when it declines to answer. A tool that fills every blank with a plausible default produces an underwriting packet that looks identical whether the inputs were sourced or invented, and the buyer inherits a risk the tool created. Suppression makes the difference visible on the page.