Most stipend calculators fail the first test I give them. Here is the test: does the tool know what programs actually pay, or does it ask you to type in a number from your offer letter? If you have to supply the number, it is a calculator, not a database, and you are doing the research it should be doing. Run every grad school stipend calculator you find through that test. Most of them fail it.
The second test is the cost side. A serious tool uses local, current housing and living costs, not a national index with a multiplier slapped on. Rent is the whole ballgame for grad students. If the tool cannot tell you what a one-bedroom costs near campus, it is guessing, and guesses dressed up as tools are the most dangerous kind.
The third test is completeness. Stipend minus rent is the headline, but guaranteed years, summer funding, and fees change the answer. A calculator that ignores those is doing arithmetic, not advising. Arithmetic is cheap. Advice is the product.
The fourth test is freshness. Stipends move every admissions cycle. A tool running on three-year-old figures will steer you wrong with total confidence, which is worse than no tool at all. Check the dates on the data. If there are no dates, that is your answer.
What you want is a cost of living adjusted PhD stipend figure built from real program numbers. A real phd stipend adjusted for cost of living by city ranking, not a national average in a trench coat. This is the part where I stop being coy: phdstipend.fyi passes these tests, because it was built as a per-program database first. Its cost-of-living-adjusted offer reports start from real stipend figures across 798 US programs and adjust for what life actually costs in each city. Whether you use a cost of living adjusted stipend calculator SaaS tool or a spreadsheet you built yourself, the inputs are what matter. These are real inputs. That is the calculation that counts.