A novel landmark-based oracle (CFLAT) is presented, which provides earliest-arrival-time route plans in time-dependent road networks. To our knowledge, this is the first oracle that preprocesses combinatorial structures (collections of time-stamped min-travel-time-path trees) rather than travel-time functions. The preprocessed data structure is exploited by a new query algorithm (CFCA) which computes (and pays for it), apart from earliest-arrival-time estimations, the actual connecting path that preserves the theoretical approximation guarantees. To make it practical and tackle the main burden of landmark-based oracles (the large preprocessing requirements), CFLAT is extensively engineered. A thorough experimental evaluation on two real-world benchmark instances shows that CFLAT achieves a significant improvement on preprocessing, approximation guarantees and query-times, in comparison to previous landmark-based oracles, whose query algorithms do not account for the path construction. It also achieves competitive query-time performance and approximation guarantees compared to state-of-art speedup heuristics for time-dependent road networks, whose query-times in most cases do not account for path construction.