Learning gains of programs that teach foundational numeracy and literacy are often measured through different types of tests. Two of the most prominent tests are ASER and EGRA/EGMA. Recent studies have found that these tests can draw different conclusions on how much children learned for the same programs. We compared student learning in a Teaching at the Right Level (TaRL) program in Botswana across ASER, EGRA, and EGMA test items to understand how results differ across them. Pre-post learning gains measured in standard deviations are smaller on the EGRA and EGMA tests than on the ASER test. Yet, when we examine multiple alternative ways to relate assessments, the tests perform more similarly. We find stronger correlations across EGRA and ASER than for items within EGRA. When measured in equivalent years of schooling, the EGMA/EGRA-ASER gap decreases, in particular for numeracy. In a randomised A/B test of a literacy programme tweak, the estimated impact was similar on both tests in terms of standard-deviations but much larger on EGRA items when compared to control group gains. We conclude that standard deviations can be misleading, and that ASER and EGRA/EGMA are important complements.