Keep the basis beside the result
A $/SF value without inclusions and area basis should not enter an estimate database. Store the full benchmark record.
Compare like with like
Filter benchmarks to comparable project type, scale, quality, geography, and date. Use ranges and sample counts rather than false precision.
Use it as a cross-check
Compare the detailed estimate to relevant benchmarks and investigate large differences. Do not force the estimate to an average without understanding the cause.