Summary
A reviewer is testing whether your number can be reproduced, not whether it sounds reasonable. Saying which method you used isn't the same as supporting a rate. The record that answers a revision request is the one you wrote while the analysis was still in front of you. That's a note at the time, or an afternoon of remembering three weeks later.
What a revision request usually is
Most of them aren't arguments about value. They are requests for the reasoning behind a number, from somebody who wasn't there when you did the work and can't see anything you did not write down.
Put that way, the request is answerable. The problem is timing. It arrives weeks after the file closed. The MLS search that produced the rate was never saved, and the sales you rejected are not listed anywhere. So the reasoning has to be rebuilt from memory, and then made to match a number you already reported.
Naming a method isn't the same as showing a rate
This is where most support language falls short. Compare these two.
"The living area adjustment was derived through paired sales analysis of comparable sales in the subject's market area."
That sentence appears in a great many reports and answers none of the reviewer's questions. Which sales? What did each pair indicate? What range? Why this figure out of that range? It describes an activity. Support has to produce a number.
"Seven pairs from the subject's subdivision, time-adjusted to the effective date, indicated $38 to $71 per square foot of living area. The four closest in age and condition indicated $44 to $58. A grouped-median test across the full 34-sale set indicated $51. $52 was selected as consistent with both, and applying it narrowed the adjusted range of the comparables from 14% to 6%."
The second is barely longer. It names the data, the method, the range, the second check, the selection and the test. A reviewer reading it has nothing left to ask.
The four questions, in order
Every adjustment gets tested the same way, whatever it measures. Answer in this order and the exchange usually ends.
- Which sales. The set, with your filters stated, including what you left out and why.
- Which method. What you measured, and what backed it up independently.
- What range. The spread the data supports, not only the point you chose from it.
- Why this figure. The reason it sits where it does within that range.
Add a fifth when you have it: what test it passed. The range check asks whether the adjustment brought your comparable indications closer together. It's quick, it's objective, and reviewers find it convincing because it shows a result instead of making an argument.
Write the record while the analysis is open
All of the above is easy while you're doing the work and expensive afterward. The inputs are on screen, the rejected sales are still in the list, and the reason you picked fifty-two over fifty-eight is still in your head. Three weeks later you have to recover every one of those, and the recovery is slower and less accurate than the original.
So treat the support exhibit as something the analysis produces, not a document you write at the end. Anything that captures the data set, the indications and the selection as it happens will do. The format matters far less than the timing.
What belongs in the workfile
- The data set as it was, not a description of it. Re-run a market search later and it returns different results, and you won't be able to see the difference.
- The sales you excluded and why. Exclusions are where selection bias hides. Stating them is the cheapest answer to the accusation.
- The indications, not just the conclusion. The range is what shows the number was measured rather than chosen.
- The dates. Which date each comparable was time-adjusted from, and at what rate. See contract date, not closing date.
- The zeros. A deliberate no-adjustment needs its sentence, or it reads as an oversight. See sometimes the right adjustment is no adjustment.
The file has to agree with itself
A reviewer comparing your grid to your narrative is checking that the two match. Rates that appear in the addendum but not on the grid. A market conditions discussion describing a rising market next to a negative adjustment. A site value in the cost approach that doesn't match the one behind the lot-size line. Each of these is easy to catch, and each costs more credibility than the underlying mistake deserves.
Deriving each rate once and carrying it everywhere it appears takes care of most of this. It's worth one deliberate pass before the report goes out.
Common questions
How much support is enough?
Enough that another appraiser with your data could reproduce your number without talking to you. That standard is useful because you can test it, and it is usually shorter than people fear. A paragraph and a table per adjustment.
Does more documentation increase my exposure?
Documented reasoning that turns out to be arguable is a professional disagreement. An undocumented number is much harder to defend, because there is nothing to point at except your recollection. Stating the range and the limits of the data protects you.
What if the client only wants the form?
The workfile requirement doesn't depend on what the client wants to receive. The support can live in the file whether or not you deliver it, and a revision request will ask for the file.
