When Do fastdid and Stata csdid Agree? A Replication and Practitioner's Checklist

The Callaway and Sant'Anna (2021) estimator for staggered treatment adoption is now available in several software packages, and applied researchers increasingly rely on the assumption that these implementations, sharing a common theoretical foundation, will return the same answer from the same data. This paper examines that assumption directly by comparing the fastdid package in R with the csdid package in Stata on a placebo design in which the true treatment effect is known to be zero by construction. We draw a broad, treatment-free sample of one hundred thousand Turkish firms, assign a fake staggered adoption year at random, and estimate the overall effect in both packages. The central finding is reassuring rather than alarming: when five specific settings are aligned across the two tools, they agree to six decimal places, and both sit at the placebo zero. The risks we do identify lie elsewhere than in dramatic silent divergence. The most consequential is a trap in the shape of the output itself: one package's compact result contains two rows, a pre-trend diagnostic and the actual treatment effect, and reading the wrong row silently reports a pre-trend as if it were the estimate, with no warning. We began this study expecting to document large hidden discrepancies between the two implementations. The evidence pointed instead to a more useful conclusion, which we state as a short checklist: the tools are trustworthy when settings are aligned, and the practitioner's real task is to read the right quantity and to report which settings were matched.