RIR vs RPE: what to actually write down after a set
RIR and RPE are the same scale read from opposite ends. Which one to log, why RPE 10 is not failure, and why half points are usually finer than the instrument.
Written while building GETSIXPACK — a training log that scores the session against your range and calls the next load.
RIR is reps in reserve. RPE is the same measurement read backwards. On a 10-point scale the conversion is subtraction:
| RPE | RIR | What it means |
|---|---|---|
| 10 | 0 | Nothing left. The next rep does not happen. |
| 9.5 | 0–1 | Maybe one, and you would not bet on it. |
| 9 | 1 | One more, clean. |
| 8 | 2 | Two more. |
| 7 | 3 | Three more. Working, not hard. |
Pick whichever end you think in and stay there. Switching mid-block and then comparing the numbers is the mistake, not picking the wrong one.
RPE 10 is 0 RIR. It is not failure.
This is the distinction that matters most and gets collapsed most often.
A set taken to 0 RIR is one where you could not have completed another rep with good form. A set where you tried another rep and it did not go up is a missed rep — a different event, and one worth recording separately, because it tells you something 0 RIR does not.
Labelling 10 as “failure” has a specific, measurable consequence: people stop logging 10s. A set that genuinely had nothing left gets written down as a 9, because “failure” sounds like something went wrong and nothing did. Every autoregulated load that reads those numbers then drifts down, quietly, because its inputs have been shaved by half a point across the board.
If you log by hand, keep a separate mark for a missed rep. If your app treats RPE 10 and a failed rep as the same event, it is throwing away the more informative of the two.
Half points are usually finer than the instrument
The literature on this is not flattering. Trained lifters, asked to predict how many reps they have left, tend to underpredict proximity to failure by one to two reps, and they are worse at it the further from failure the set is. At 5 RIR the estimate is close to a guess.
So a scale with 0.5 steps is offering a resolution the reader does not have. That is not an argument for never using half points — at the top of the scale, between 9 and 10, the difference is real and most people can feel it. It is an argument against building anything important on the difference between 7 and 7.5.
Practical version: log whole numbers by default. Reach for a half only near the top of the range, where you can actually tell.
Log it per set, not per session
“Today was an 8” is a mood. Log the set.
Effort belongs to a set because that is the unit progression decides on. Three sets of 8 reps at RIR 2, 2, 2 and three sets of 8 at RIR 3, 2, 0 are different sessions with identical rep counts, and only one of them says the load is ready to move. If you are running double progression, the effort on the last working set is the number the load increase hangs on, and a session-level average deletes exactly that.
What it is for
Two things, and they are different.
Deciding today’s load. You planned three sets of 8 at RIR 2 and the first set came in at RIR 0. Something is different today: sleep, food, a hard session two days ago. Drop the load rather than complete the prescription and call it a win. That is autoregulation in the narrow sense, and it is the reason to log effort at all.
Deciding next session’s load. Here effort is the gate. A rep sweep at 0 RIR on the last set is weaker evidence than a rep sweep at 2 RIR, and only the second one earns an increase.
What it is not for
Self-reported RIR is the least reliable input in any progression model, and it should never be the only thing a decision rests on.
Concretely: “the last set was 0 RIR” is not sufficient grounds to cut the load. Reps falling out of the bottom of the prescribed range is evidence. A 0 RIR last set is corroboration. One without the other is a number moving on a guess — and if any of that RIR was reconstructed rather than reported, which is what happens when history gets imported from a spreadsheet or another app, it is a guess dressed up as a report.
Log effort per set. Keep the scale coarse. Keep misses separate.
Where this comes from
The 1–10 RPE scale anchored to reps in reserve is Zourdos and colleagues (2016, Journal of Strength and Conditioning Research). The finding that trained lifters underpredict how close they are to failure, and get worse at it the further from failure the set is, is Hackett and colleagues (2012, Journal of Sports Sciences). Helms and colleagues (2016) is the standard treatment of prescribing load by RPE rather than by percentage.