Why did my lift go down this week when nothing else changed?

Last week an 80 kg lifter pulled 140 kg for three at RPE 8. This week the same three reps felt like RPE 9, and the program had not changed. In the app's model that set is worth 157.0 kg for one against 162.2 kg a week earlier: 5.2 kg or 3.2 per cent down, computed with the model behind the one-rep max calculator. A tested max moves about that much on its own: across 32 test-retest studies the median coefficient of variation for a one-rep max was 4.2 per cent [1]. One session down inside that band is noise. Two sessions down, or a grinder, is fatigue, and the fix is a hold or a small back-off.

One bad session is inside the noise

A one-rep max is a reliable test, and it still wobbles. The review that pooled 32 studies and 1,595 participants found test-retest correlations with a median of 0.97, and coefficients of variation from 0.5 to 12.1 per cent with a median of 4.2, regardless of exercise or training age [1]. Test the same lifter on two days and the bar lands a few kilos apart. An estimated 1RM inherits that wobble and adds one more, the RPE you log. The estimated 1RM guide covers how the chart turns reps and RPE into that number. The same 140 kg triple reads like this.

A point of RPE at a fixed load, or 5 kg off the bar at a fixed RPE, is the same 3 per cent story, inside the band. A grinder is different. RPE 10 for the same three reps is a 6.4 per cent drop, outside it, and the app treats it that way. With a target of RPE 8, its load model returns a hold at 140 kg for the RPE 9 set and a back-off to 132.5 kg for the RPE 10 set, computed with the app's model. It never adds weight to a set that was harder than planned.

What changed, if the program did not?

Sleep is the usual answer. A meta-analysis of 69 studies found that six hours or less in the previous 24 cut exercise performance by 7.56 per cent on average, with strength among the categories that fell, and after a lost or cut-short night the loss grew by about 0.4 per cent for every hour awake before the session [2]. Evening sessions took the hit while morning sessions were largely unaffected [2]. A short night before a 7 pm deadlift is enough on its own to explain the step from RPE 8 to RPE 9.

Soreness is the second. A new exercise, or an old one after a break, damages muscle most at first: in ten men followed through ten weeks of training, damage after a session was highest in week one, lower by week three and minimal by week ten [3]. While it lasts, soreness costs peak torque and joint range of motion, and the review on delayed onset muscle soreness recommends reducing training intensity and duration for one to two days after the session that caused it [4]. Swap the Monday squat variation and Wednesday's deadlift pays for it.

The third is the training around the set. The app reads readiness from two rolling windows, hard sets in the last 48 hours and load in the last 72 hours, so a heavy pull on Tuesday still weighs on Thursday whatever the calendar week says. The fourth is where the set sat: a top set pulled as the fourth working set estimates lower than the same set pulled first, so compare estimates from the same position in a session. The app cannot see your sleep or your calories. It reads the log, and the log only tells the truth if the RPE does.

When two sessions down become a stall

Two sessions is still a wait. By default the app flags a stall after three sessions at the same load with no rep gained, and its first move is a hold: repeat the weight and check rest, sleep and fatigue. If the stall runs on, it suggests a small back-off and leaves the plan alone until you accept it. When the estimate has been sliding across sessions, it skips the hold and goes straight to the back-off. A deload week never emits an increase. More than 30 days off a lift resets it: 140 kg for three at RPE 8, then 35 days away, comes back as 132.5 kg, computed with the app's model.

That is autoregulation in the plain sense of the research: adjusting the next session on measured performance or on how capable you feel, a framework the review traces to the 1940s while noting that even the words readiness and fatigue are used inconsistently [5]. One session down is a Tuesday. Two in a row is a reason to hold. Three at the same load is a stall, and the double progression guide covers what to do when the reps do not come. Top Sets does this arithmetic on every set you log, so the answer sits next to the set before you have racked the bar.

Questions

Should I lower the weight after one bad session?

No. Repeat it. One set that felt a point harder at the same load is a 3.2 per cent change in the estimate for a 140 kg triple, inside the 4.2 per cent a tested max varies by between test days [1]. The app holds the load after an RPE 9 and backs off only after an RPE 10. Change the weight when the log repeats itself.

How many bad sessions before it is a real stall?

Three at the same load with no rep gained, by default. The app then holds the weight and asks you to check rest and fatigue, and if the stall runs on it suggests a small back-off that it never applies on its own. A rising estimate at the same load is not a stall: reps and effort are progress too, the idea behind double progression.

Is my lift going down because I am cutting?

Possibly, and it is usually fine. In the meta-analysis of trials that lifted in an energy deficit for three weeks or more, lean mass gains were impaired but strength gains were not significantly different from lifters training without a deficit [6]. On a cut the app turns every load increase into a hold, so holding your numbers while the scale drops is the plan working.

Does a lower estimated 1RM mean I lost strength?

Not from one session. The estimate reads the set you logged, so a point of RPE at the same load moves it about 3 per cent, and a set logged without RPE drops it further because a blank RPE counts as a max. Strength is the direction across weeks of sessions. Until a lift has three sessions logged the model has no trend to read, because two points are never a trend.