Energy assessment of electrified vehicles requires driving cycles that include altitude change, whereas certification cycles and most synthesis methods describe flat-road speed profiles only. This study synthesizes time–speed–altitude driving cycles from real-world data (145.7 h, 10,491 km). Road grade was obtained from a vehicle-mounted barometric sensor corrected for its speed-dependent pressure drop, and its low-frequency drift was replaced by a digital elevation model (DEM). A distance-based semi-Markov route model of road types (urban/rural, highway and mountain) was combined with a 1 s Markov chain of type-specific acceleration and grade transitions; sign symmetrization and a grade-trend state removed the net-ascent bias. For 4934 synthesized cycles, the route model reduced the root mean square error (RMSE) of the steep-grade share by speed bin from 9.2 to 2.5 percentage points, and sampling the next acceleration instead of the next speed reduced the deviation of the mean positive vehicle-specific power (VSP) from +27% to +3.4%. With terminal decelerations sampled from measured stop approaches, the recoverable energy exceeded that of real-world windows ending with the same kind of stop approach by 3.3% (distance-weighted), and neglecting grade underestimated it by 18%. In day-level bootstrap and leave-one-day-out cross-validation, the traction power and energy deviations stayed within the day-to-day variability. A flat output reproduced the mean positive VSP, positive specific energy, and energy consumption of zero-grade real-world driving within 3%. Finally, a representative cycle and speed-group × VSP-severity cycles were selected.