Down Weeks: How Much to Drop and How Often

Your plan decided its recovery weeks before you ran a single step of it. Runna slotted a cutback into week 4 the moment you entered your race date. Garmin Coach did something similar, quietly, based on your VO2 max estimate and how many days a week you said you could run. Neither of them knew that week 3 would be the week you flew to Dubai, slept badly for four nights, and ran a 29 km long run at 31°C.

That is the whole problem with the 3:1 pattern. The ratio is fixed in advance by software that cannot see your sleep, your work, your virus, or the heat, and then it holds the down week hostage until the calendar catches up. Sometimes the calendar is roughly right. Often it’s a week late, occasionally two, and by then you’re not recovering from a training block, you’re recovering from a decision.

So, how often should I take a cut back week? Honest answer: every two to four weeks, and the width of that range is the point rather than a hedge. A 34-year-old running 40 km a week for a 10K can usually hold a 4:1 or even 5:1 rhythm. A 48-year-old building to 85 km for a marathon, sleeping six and a half hours, and lifting twice a week will often need 2:1. Your plan almost certainly gave you one number for the whole 16 weeks, which is the tell that it was never reading you in the first place.

The signals that should actually pull the trigger

You need a small set of measurements you check weekly, and a rule for how many have to fire. Two or more of these, persisting four days or longer, means take the down week now regardless of what week the app says you’re in:

• HRV trend. Garmin’s HRV Status sitting in “Unbalanced” or “Low” for four-plus consecutive mornings. On Whoop or Oura, a 7-day rolling average below your 60-day baseline for the same stretch. One bad night is noise, four in a row is a signal.
• Resting HR. Up 4 bpm or more on your own baseline for three mornings. Mine sits at 48; a run of 53s means something.
• Pace-at-HR drift. Your standard easy run costs 6 bpm more than it did a fortnight ago at the same pace, same route, similar temperature. In Intervals.icu this shows as your easy-run efficiency factor sliding; on the watch it’s cruder but still legible.
• Long-run decoupling. Intervals.icu gives you Pw:HR (or HR drift) on every run. Under 5% on a steady long run is healthy. Above 8% on a run that shouldn’t be that hard means you started it tired.
• Performance Condition. Garmin shows this 6 to 20 minutes into a run. Consistently negative for a week of easy running is worth more attention than most people give it.
• A niggle that changes how you move. Not soreness. Anything that makes you alter your gait, or that’s still there on day four, counts as a fatigue signal even if nothing hurts while you’re running.
• Second-day flatness. Your legs feel dead on the easy day after the easy day. That specific pattern, rather than being tired after intervals, is the one experienced runners learn to respect.

Sleep and mood belong here too, and they’re not soft metrics. Waking at 4:40 a.m. wired, losing appetite after long runs, snapping at people: those are sympathetic-nervous-system readouts, they’re just inconvenient to graph.

Auditing whether your plan’s cutback is real

Here’s the part almost nobody checks. A down week is defined by load reduction, not by the word “recovery” in the app. Plenty of AI-generated cutbacks shave volume while leaving the intensity and the long run intact, which reshuffles the week rather than reducing it.

Take a real Runna marathon block, five runs a week, from a 41-year-old aiming for 3:45:

Volume fell 24%, which looks like a down week on the plan screen. But the long run only came down 3 km, the quality minutes barely moved, and Form recovered five points to land at −12. She entered week 5 more tired than she started week 1, and did it three times in a row before her calf let her know.

A genuine down week in that same slot looks like this in Strava’s Fitness & Freshness graph:

Sun wk 3   Fitness 66   Fatigue 83   Form −17
Wed wk 4   Fitness 66   Fatigue 68   Form  −2
Sun wk 4   Fitness 65   Fatigue 58   Form  +7

Fitness drops one point. That single point is the price, and it’s trivially cheap next to 24 points of Fatigue coming off. If your down week doesn’t move Form by at least 10 points and land it above roughly −10, it wasn’t a down week. Intervals.icu gives you the same thing as CTL, ATL and Form for free, and it reads from Strava or Garmin without you doing anything. Garmin users can watch Acute Load fall and the Load Ratio drop toward 0.8 to 1.0 instead. The specific metric matters much less than the habit of looking; the mechanics of why rolling 7-day and 42-day loads behave this way are covered in our guide to training load and injury risk.

What to cut, in order

The cut has to come from somewhere structural, and the long run is usually 50 to 60% of the week. In that 58 km week, the 29 km long run was the week. Cut it to 19 km and you’ve found 10 km without touching anything else.

So the priority order:

1. Long run down 30 to 40%. 29 km becomes 19 or 20. Run it easy, and resist the urge to put a “small” marathon-pace block at the end.
2. Delete the second quality session entirely. Not shorten it. Delete it.
3. Halve the remaining quality session, keep the pace. 6 × 1 km at 4:25/km becomes 3 × 1 km at 4:25/km. You keep the neuromuscular pattern and shed most of the cost, which is the trade you want.
4. Keep every easy run on the calendar, trim each by 10 to 15%. Frequency protects your routine, your tendons and your sleep rhythm. Dropping from five runs to three is what makes people feel worse during a down week, not better.
5. Count the gym. Two heavy lower-body sessions will sit on top of a 30% running cut and cancel a good chunk of it. Halve the volume or drop to one, and keep the weight, same logic as the intervals.

Three depths are worth having in your pocket. A mild reset at 15 to 20% off, used when you caught the signals at two days rather than ten. A standard down week at 30% of volume and 40% of load, which is what most of these blocks should have been doing all along. And a deep reset at 40 to 50% with all quality removed, for when you’ve spent three weeks ignoring a rising resting HR and are now effectively taper-adjacent whether you like it or not.

Making the app go along with it

Runna lets you swap or shorten individual sessions and will accept a manually reduced week without arguing, though it won’t reduce the following week to match, so check week 5 hasn’t jumped straight back to 58 km. Garmin Coach is stubborn in a different way: skip sessions and it recalculates, but it tends to re-add load once your Training Status flips back to Productive, so you may need to move the race date back a week to get a genuine gap. With a ChatGPT-built plan you have the most control and the least protection, so feed it the actual numbers rather than a vibe: “HRV Unbalanced 5 of 7 days, resting HR 48 to 53, long-run decoupling 9.4%, cut week 3 to 40 km with a 19 km long run and one 3 × 1 km session.”

Set a Sunday-night reminder, pull up Form and HRV Status, and write down two numbers. Four weeks of that and you’ll know your own ratio better than any plan generator ever will.