Differential privacy enables data practitioners to analyze sensitive data without transgressing individual privacy – even when combining multiple pieces of information. Researchers have designed privacy mechanisms that degrade privacy to a guaranteed level when users compose sequential queries. More advanced composition theorem allows users to query a database quadratically more times than basic privacy mechanisms; however, it does not allow users to select privacy parameters based on the outcomes of past computations (e.g. when a user determines that they are almost done with their analysis and can expend the remainder of their privacy budget). Previously, Rogers et al. introduced “fully adaptive composition” to adaptively select the privacy parameters of the algorithms. However, their implementation was impractical and required strong assumptions about the algorithms being composed. Here, Whitehouse et al. leverage advances in time-uniform martingale concentration to present privacy “filters” and “odometers” that match the tightness of advanced composition.