Just-in-time (JIT) compilers leverage type feedback to optimize dynamic languages by recording runtime observations during interpretation and using them to generate specialized code. However, this recording process introduces significant execution time overhead to program warmup. This paper presents a characterization study of how feedback information is utilized by JIT compilers, providing motivation for future work on reducing recording overhead. We instrument the Ř compiler for the R language to track feedback slot usage throughout compilation, analyzing well-known benchmarks and real-world programs. Our measurements reveal that recording adds overhead to the interpreter up to 1.6×(mean 1.2×), yet at least 59% of non-empty slots are not used by the compiler. A large fraction of these unused slots can be attributed to dead code that the compiler eliminated. Most of the remaining slots are type-stable—meaning the inferred type matches the observed type—making speculation futile. To quantify the potential for improvement, we evaluate two oracle-based configurations that establish upper bounds on achievable reduction: a conservative set approach that preserves optimization quality while improving warmup, and an optimistic set approach that sacrifices some compiled code performance for faster interpretation.