How the adjustment works

Grade-Adjusted Pace (GAP) answers one question: if you'd spent the same energy running on flat ground, how fast would that pace have been? It uses the metabolic cost-of-running curve measured by Minetti et al. (2002), where the energy per metre at gradient i is C(i) = 155.4·i⁵ − 30.4·i⁴ − 43.3·i³ + 46.3·i² + 19.5·i + 3.6 J/kg/m. Each course is expanded into a high-resolution profile of ~480 points and the model integrates that cost over every one of those small segments — so the constant micro-undulation you see on a Garmin trace is actually priced in, not averaged away. Dividing raw pace by the resulting effort factor gives your GAP.

The Boston paradox

Here's the counter-intuitive part most calculators get wrong: on road gradients the metabolic penalty is small, and Boston is net downhill — so purely by oxygen cost it's about 1% easier than flat, not harder. Yet almost everyone runs Boston slower than a flat course. The reason isn't in your lungs, it's in your legs: the steep early descents cause eccentric muscle damage that wrecks your quads, and the Newton Hills arrive at 26–34 km, exactly when you're already cooked. Metabolic GAP can't see that, so the difficulty score above is a separate measure that folds in total climb, net-downhill damage, and where the hills fall.

Bottom line: GAP tells you what your effort was worth in oxygen terms; difficulty tells you why the clock still punished you.

How the profiles are built

Each profile combines a macro shape (where the real climbs and descents fall) with realistic high-frequency micro-undulation, tuned so the profile's total ascent equals the surveyed figure for that race. This deliberately avoids raw-GPS inflation: a watch might read 400–600 m on Boston because of sensor noise, but the calibrated profile sums to the true ≈248 m while still showing constant small ups and downs. The visible min/max reflect real course elevations; exact micro-placement is representative, not a survey.

Elevation data

Ascent figures compiled from findmymarathon, RunDida, worldmarathoner and official race material. Boston ≈248 m gain (net −140 m); NYC ≈294; Berlin ≈73; Chicago ≈57; London ≈127; Tokyo ≈60; Valencia ≈60; Sydney ≈317; Paris ≈180; Cape Town ≈260 (sources range 205–380 m; high point ~53 m near km 33; no official figure published); Warsaw 93 m (official 47th-edition profile, net-zero loop); Poznań ≈130 m (estimated — official sheet gives min 55 / max 95 m but no ascent sum). Descent is derived from gain and net elevation change.

Ascent figures compiled from official race material and findmymarathon / RunDida / worldmarathoner. Marathon majors use sourced gain figures; the two Warsaw shorter races use official/Garmin data (ING Biegnij Warszawo from the 2026 route profile; Bieg Powstania Warszawskiego 10 km shaped from a recorded run). Most other 1-mile to half-marathon courses are schematic — shaped to reflect each course's character (flat, rolling, bridge climbs, net-downhill) with an approximate gain, not a survey. GAP is a physiological estimate, not a race predictor; heat, wind, altitude, fitness and pacing move the clock more than terrain does.