Miles, Minutes or Load Points: Which Unit Should You Track?

Your training plan speaks one language, your watch speaks another, and Strava invents a third. Runna hands you “Tuesday: 8 km easy.” Garmin tells you your 7-day Training Load is 612 and sits in the “optimal” band. TrainingPeaks says the same week was 340 TSS and your CTL climbed 4.2 points. Three numbers, one set of legs, and no obvious way to tell whether the plan is building you or breaking you.

This is the part of running training load explained that most articles skip: the unit you choose isn’t a neutral measuring stick. Each one has a bias baked in, and that bias decides which weeks look like progress and which look like failure. Pick the wrong unit and you’ll shorten hill runs to protect your mileage, panic during a heat wave, and add intensity to a down week because the graph dipped.

For most recreational runners following an AI-generated plan, the right primary unit is duration. Minutes. Time on feet. Here’s the case.

What each unit is actually measuring

Distance measures the ground you covered, and nothing else. It’s the default in almost every AI plan, because it’s the unit runners think in and the unit race distances use. It’s also the only one on this list that gets easier to accumulate as conditions get harder to run in, which is a strange property for a fatigue metric.

Duration measures how long your body was under load. Not how fast, not how high, not how hot. Just time.

sRPE (session RPE) is the oldest sports-science method here and still one of the best validated. You rate the whole session 1 to 10 on perceived effort, roughly 30 minutes after finishing, then multiply by the minutes. A 60-minute easy run at RPE 3 is 180 arbitrary units. A 50-minute interval session at RPE 8 is 400 AU. Cheap, no hardware, and it captures the one thing no sensor sees: that you slept five hours and had a bad day at work.

Garmin Training Load is a rolling 7-day sum of per-session EPOC estimates from Firstbeat’s model, driven mostly by heart rate. On a Forerunner 265, Fenix 7 or Epix, it feeds Acute Load, Load Ratio and Load Focus (anaerobic / high aerobic / low aerobic).

TrainingPeaks rTSS normalises your run against your threshold pace: 100 points equals one hour at threshold. It uses Normalized Graded Pace, so hills are grade-adjusted. Set your threshold pace wrong and every number downstream is wrong.

Strava Relative Effort weights time in heart rate zones. No HR strap, and it guesses from pace.

Where distance falls apart

Take a concrete Tuesday and Sunday.

In a distance-based log these are the same run. Twenty kilometres for the week, tick. In reality the Sunday run cost you 8 minutes and 20 seconds more of muscular work, roughly 1,400 extra ground contacts at 170 spm, and a good deal more eccentric loading on the quads coming back down. Distance says identical. Duration says +14%, which is much closer to the truth.

The failure compounds when a plan prescribes distance and you’re running somewhere hilly. To hit “8 km easy” on a trail you run for 55 minutes instead of 48, and your actual weekly training time is 15% higher than the plan thinks. Nobody notices until something in the shin does.

Where heart-rate-driven load scores fall apart

Heat is the clean example. Same runner, same 60-minute easy loop, same effort, three weeks apart.

11°C, dew point 6°C    → avg HR 141   Strava Relative Effort 48   Garmin Load 95
29°C, dew point 21°C   → avg HR 156   Strava Relative Effort 79   Garmin Load 140

The July run generated 47% more Garmin load and 65% more Relative Effort. You did not train 65% harder. You thermoregulated. Cardiac drift in the heat is real physiological strain, so the score isn’t lying exactly, but it’s answering a different question than the one you asked. You wanted to know whether your plan is progressing you. The score told you it was hot.

This matters because the inflation is sticky. A hot August pumps up your 42-day chronic load, so when September cools down, your acute-to-chronic ratio reads comfortably low right at the moment you start adding real intensity. The metric that was supposed to catch overreaching has quietly moved the goalposts.

Heart-rate load also drifts with caffeine, illness, dehydration and a chest strap that slipped. Duration doesn’t care about any of that.

Where load points punish the down week

Build a realistic peak week and its recovery week, and score both.

Peak week (320 min, 5h20):

Down week (245 min, 4h05): all easy, strides only.

Duration says you did 77% of peak. rTSS says 62%. Garmin will say something similar and may drop your Training Status to “Unproductive” or “Recovery” while it’s at it.

Both numbers are arithmetically correct. But 62% feels like a hole, and the predictable human response is to put a tempo back in on Thursday, which defeats the entire purpose of the week. Intensity-weighted scores are designed to answer “how much fitness stimulus did I get,” and a recovery week is deliberately low on stimulus. Using them to police your consistency is using the wrong tool.

Duration’s version of the same week reads as 4 hours instead of 5h20. Lower, obviously, but proportionate and non-alarming. That’s not a rounding difference in how you feel about it; it changes what you do on Thursday.

The practical setup

Make weekly minutes your progression number. Convert whatever your plan prescribes into time at your actual paces, then track two things.

Weekly duration, with a 10% ceiling on increases. If last week was 300 minutes, next week’s cap is 330.

The 7-day to 28-day ratio. Take this week’s minutes, divide by the average of the last four weeks.

Week 1: 280    Week 2: 300    Week 3: 320    Week 4: 240 (down)
28-day average = 285
This week: 350  →  350 / 285 = 1.23

Under 1.3 is ordinary. Above 1.5 is where the injury-risk literature starts getting nervous, and there’s a fuller treatment of the acute-to-chronic evidence and its caveats in our guide to training load and injury risk. Below 0.8 for several weeks running means you’re detraining, not tapering.

Duration has one genuine blind spot: 60 minutes of jogging and 60 minutes of 400s are not the same 60 minutes. Patch it with a second, much simpler number. Count your quality minutes, meaning minutes at threshold pace or faster, including the hard portions of intervals but not recoveries. Keep them to roughly 15% of weekly duration, so around 45 minutes in a 300-minute week. Two sessions of 20 minutes of tempo, or a 5 × 1000m workout, and you’re there.

Prefer something with a bit more nuance? Layer sRPE on top. It costs you five seconds per run and catches the sessions where the pace was easy but you weren’t. Log RPE × minutes in a notes field, total it weekly, and compare to duration. When your sRPE total climbs faster than your minutes for two weeks straight, you’re accumulating strain that isn’t showing up in the volume, and that’s the earliest warning any of these methods will give you.

Keep the load scores, demote them

None of this means deleting Garmin’s numbers. Acute Load Ratio and Load Focus are useful as a second opinion, particularly Load Focus, which is genuinely good at catching the classic recreational-runner error of having almost no low-aerobic volume. TrainingPeaks CTL is worth watching if you’ve tested your threshold pace in the last eight weeks and you actually keep it updated. Relative Effort is fine for comparing two similar runs in similar weather.

Demoting them just means they don’t get a vote on whether you run tomorrow. That decision comes from minutes, quality minutes, and how your calves feel walking down the stairs.

Your AI plan is going to hand you distances, because every plan does. Open next week’s block, take your honest easy pace (not your Strava-flattering one, the one you actually run on a Wednesday evening), and rewrite each session as minutes. Then add the week up. If the total is more than 10% above last week’s, the plan doesn’t know that yet, and you do.