Race Split Planner

Pace model

Splits are built by converting each mile into a "flat-equivalent" distance using a grade-dependent cost factor, then pacing evenly against that adjusted total rather than the raw mileage. That is an even-effort strategy, not an even-pace one. The cost factor is Strava's empirical Grade Adjusted Pace model (Robb, 2017, An Improved GAP Model, Strava Engineering), fit to heart-rate-equivalent pace across ~6 million real runs rather than to treadmill oxygen cost. Strava's fitted curve is proprietary, so both branches are reconstructed from its published figures: downhill, a parabola through the published minimum (0.88 at about -9% grade) and back to flat cost by -18%, held there for steeper descents; uphill, the published worked example (5:00/km at +5% grade → GAP 4:27/km) gives +2.5% cost per 1% of grade, used as a linear slope. Strava also reports its uphill penalty is pace-dependent, which a single slope does not capture. Grades past +/-45% are clamped, since real GPX grades beyond that are almost always sensor noise on a short segment rather than terrain.

Terrain technicality

Trail and hybrid race types apply an extra downhill cost multiplier (1.15x and 1.075x), since technical footing (roots, rocks, loose surface) keeps runners from converting a descent's gravity assist into speed the way pavement does. This multiplier is a heuristic, not a fitted or cited figure. Treat trail/hybrid splits as an approximation rather than precisely calibrated.

Grade computation

Grade is rise over horizontal (2D) distance, not the diagonal (3D) distance actually run. That is the standard %grade convention; using slope distance instead systematically understates grade, worse as terrain steepens. Because the cost model describes sustained effort (heart rate lags terrain by tens of seconds), raw adjacent-trackpoint deltas are aggregated into ~100 m chunks before computing grade, damping point-to-point GPS/elevation sensor noise that would otherwise be read as real terrain change.