Kasing Ng, CPA ← Home

How the numbers are checked

Every Deal Tax Projection comes out of software I built. Before you lean on one of these reports, the fair question is simple: who checks the software? I spent years as an auditor asking exactly that about other people's models, so here is the honest answer, including the parts I don't check.

21 × 7test deals and key tax figures double-checked in a separate spreadsheet I built by hand

15,000+saved results locked in; if my code moves one by accident, the check catches it

40+automated checks run before any change reaches your report

1. A second, independent calculation

The tax you owe when you sell is the part that has to be right, so I check it twice. Alongside the software, I built a completely separate version in Excel, typing the tax rules in by hand rather than copying the software. Then I ran 21 different deals through both and compared the figures that matter: your gain at sale, the depreciation you have to pay back, the capital-gains tax, the total tax bill when you sell, and the losses you carry forward.

Every one matches, on all 21 deals. Two separate tools, built two different ways, landing on the same number. If the software had a hidden tax error, the hand-built Excel would not agree with it. The check re-runs any time, so a mistake turns up here, not in your report.

2. Saved results that can't move without me knowing

I keep more than 30 test deals on hand, from ground-up builds to stabilized purchases, each with its correct answers on record. A separate check loads 20 of my real deal models the same way you would click through the app, plus one with the lease-up turned off, and compares about 15,600 saved results against what they should be.

If a single result drifts by more than rounding, the check stops and names exactly which one, on which deal. Nothing in your report changes quietly. A number moves only when I mean it to, and I review the change before accepting it.

3. Checks proven to actually catch mistakes

A test that always passes proves nothing. So whenever I add a check, I deliberately break the software in a throwaway copy, for example nudging one month of income by a tenth of a percent, and confirm the check turns red on exactly the deals it should. If it doesn't catch the break, it doesn't count. The alarms get tested before I trust them.

What is not checked, and what this is not

The hand-built double-check covers the tax at sale. How the cash is split between you and your investors is checked a different way, against saved results, not a second calculation.
The year-to-year tax doesn't tie to the penny between the two tools. They differ for two reasons I tracked down and wrote up: a vacancy allowance in the spreadsheet, and a timing difference in when depreciation starts. Known, not hidden.
Testing proves the math does what it's built to do. It can't make your inputs right. A projection is only as good as the rent, cost, loan and sale assumptions behind it, which is why every report lists them in full.
This is not an audit. These are my own quality checks on my own software, not a formal opinion on any deal or financial statement, and not a cost segregation study.

See the output these checks stand behind

The sample Deal Tax Projection is a fictional multifamily development run through the same engine.

See the sample (PDF) → Book a 15-min call

Descriptions of testing refer to the engine as of the date of this page and to the scenarios described; they are not a guarantee of any result. This page is educational and is not tax, legal, or investment advice. Advisory engagements are non-attest: I do not perform audits, reviews, or compilations, and I do not prepare or file tax returns. Kasing Ng, CPA — California License 139270.