Trail Race Time Predictor

Treat any prediction as a window, not a time. This one reports a range taken from its own calibration: your result becomes an estimated performance index, and the spread of that index is what the finish window is built from - roughly plus or minus 4% in time. Conditions the model cannot see move the real answer further than the model does. Heat, altitude, a night section, mud, a course change and how you fuel all sit outside it.

You need one thing you have already run, and the target course's distance and elevation gain. The tool converts your past result into a performance level, then asks what that level is worth on the new course. Pick your race from the list and the organiser's official figures load automatically, or enter a custom course if your race is not listed.

The most recent one that reflects your current fitness, and the one closest in character to the target. A mountain 50 km predicts a mountain 100 km far better than a flat road marathon does, because our verified results are thinner at some distance and vertical combinations than others - the tool flags it when your result and your target sit in different race categories. A result from six months ago is fine if your training has not changed. A result you were ill or injured for is not.

The working rule across trail running is that 100 m of climbing costs roughly the time of one extra flat kilometre - that is ITRA's km-effort definition, and it is what the performance index is scored on. A 50 km race with 3,000 m of gain is therefore worth about 80 km-effort, so expect a finish nearer an 80 km flat race than a 50 km one. Steep descent costs real time too, through eccentric muscle damage rather than metabolic cost, which is why the km-effort figure shown beside your prediction adds a descent term the official ITRA formula leaves out - and why the two numbers differ.

The classic Riegel exponent of 1.06 was fitted to road results, where distance is the only thing that changes. It has no term for 10,000 m of climbing, for technical footing, or for the fact that pace decays faster in an ultra than a road-derived power law expects. Applied raw to a mountain ultra it will underpredict you, often by hours. This tool uses a power law too - but with an exponent fitted to public trail results rather than road ones, which is why its answer is slower and closer to what the field actually runs.

Yes, and it is the most common thing people do here - but it is the least reliable input the tool accepts. A road marathon is below the 50 km-effort mark where our calibrated model starts, so the tool scales from it with a power law instead and says so, showing a wider range. Expect the real answer to land toward the slow end of that range, and expect the gap to widen the more vertical the target course carries. A longer or hillier reference result gives a better answer.

Yes, and it uses more of your history than one typed-in result can. Connecting Strava lets TrailMath read the last 90 days you have actually run - your real volume, your real climbing rate, your long runs - and estimate your finish from all of it instead of from a single race. It is read-only, nothing is ever posted to Strava, and you can disconnect from Strava's own settings at any time.

No, and that is worth being explicit about. The model is fitted on distance, climb and finish times, so an average level of technicality is baked into the curve, but nothing in it can tell a rocky, rooted singletrack from a fire road at the same gradient. Technical ground can add 10-20% at the same gradient. Treat the prediction as a lower bound on a genuinely technical course, and add your own buffer.

None of the three are in the model, and all three are real. As rough starting points: 10-15% on a hot canyon race if you are not heat-acclimatised, 20-25% at Hardrock altitudes off a sea-level reference, and time lost to footing rather than to the profile on a night race. The Race Conditions Simulator breaks these down factor by factor - run your predicted time through it before committing to a goal.