I was wrong.
My own data proved it.
KBL 110 km. Kilometres 70–80 looked like a textbook night-time collapse: 12:47/km, a quarter slower than my average. I filed it as a breakdown and nearly built my next training block on it. Then I measured the terrain.
First diagnosis: “the night low”
The race started Friday 17:00 and finished Saturday 11:34 — 18 hours 34 minutes over 110.21 km with 4396 m of climbing. Kilometres 70–80 fell around three in the morning and are the slowest ten of the whole race.
The conclusion writes itself: night, fatigue, pace drop. That conclusion has one flaw — it is convenient. It matched what I felt, so I did not feel like checking it.
What the terrain showed
These ten kilometres hold four climbs totalling +683 m — I spent 77% of the segment going uphill. The hardest piece starts just after kilometre 74: 1.99 km, +322 m, averaging 16.2%. At that gradient you do not run — you hike. It took me 30 minutes.
The same effort, a different moment of the race
To compare two climbs of different steepness you need a common measure. I use grade-adjusted pace: the energy cost of uphill running from Minetti's polynomial (2002) converts real metres into “flat metres”, and I divide the time by that adjusted distance.
| Climb | When | Grade | Pace | Grade-adj. | HR |
|---|---|---|---|---|---|
| km 12,5 | 3 h in | 8,4% | 11:44/km | 7:54/km | 140 |
| km 74,5 | after 15 h | 16,2% | 15:25/km | 7:55/km | 121 |
After fifteen hours, on a climb twice as steep, the grade-adjusted pace was one single second per kilometre slower — at a heart rate 19 beats lower. That does not look like a body that has broken down.
An honest word on this number's limits. At 16.2% the Minetti correction is larger than at 8.4%, so the model treats the steeper climb more generously — which is exactly why I do not claim I was stronger. I claim only that I lost nothing: one second per kilometre sits inside measurement noise, and after a genuine crisis I would expect minutes. A model named out loud can be checked; “I felt slow” cannot.
The second witness: heart rate and power
One metric can lie; two independent ones rarely do. Across the successive big climbs, heart rate stepped down: 146 is the highest reading of the whole race, 140 on the climb at km 12.5, 121 on the climb after kilometre 74, 107 on the last one. Power dropped more gently — from 203 W on the first climb to 177–178 W on the final two.
And the fact that closes the case: the fastest ten-kilometre block of the entire race was the last one — 100–110 km at 8:43/km, with 286 m of climbing. Nobody who truly broke after kilometre 74 finishes like that.
The cost of a wrong diagnosis
Had I stayed with “the night low”, my next block would have gone toward crisis resilience: more night sessions, mental work, second-half fuelling. All sensible — and all beside the point. I would have been fixing something that was not broken, at the expense of what did need work: repeatable strength on short, very steep sections taken on tired legs.
The difference between those two blocks is not a nuance. It is several months of a season.
How to check this yourself
- Take the watch file, not the summary. Every number here comes from raw FIT records — 66,835 points. The session summary reported several hundred metres less ascent, because it filters barometric noise.
- Segment the course by gradient, not into equal kilometres. A 16% kilometre and a flat kilometre are two different physiological events filed under one heading.
- Compare effort, not pace. Pick an adjustment model (mine is Minetti) and state it alongside the result. A number without a model cannot be verified.
- Look for a second witness. Heart rate at steady power, descent pace, the final ten kilometres. If three independent signals agree, the diagnosis holds.
Join UltraFlow
I ran this teardown with the tools inside UltraFlow: gradient-based course segmentation, effort converted to a common measure, and heart rate set against power on the same section. The system does not guess why you slowed down — it shows what was under your feet at that moment.