One workout in, one decision out.
Two steps of plain arithmetic work out your numbers. A language model then reads the pattern for each muscle group and proposes changes. A set of checks can throw those out before anything reaches your templates.
Finished working sets per muscle group, with the weight, reps and reps in reserve. Warm-ups, sets you did not finish and sets stopped four or more reps short are left out.
What the plan asked for against what you did, plus the direction your strength is going on that lift.
A language model reads the pattern for each muscle group and proposes one change, with a confidence score.
Rules that throw out a change built on too little evidence, and limits on how far one session can move anything.
With Auto-Pilot on, accepted changes go straight into your templates. With it off, they wait for you to say yes. You can also write your own plan, and then the automatic engine steps aside.
Hard sets per muscle, per week.
Volume is counted in hard sets per week for each of ten muscle groups. A hard set is a working set you finished with three reps in reserve or fewer. It counts in full for the muscles doing the work. Warm-ups, unfinished sets and sets stopped four or more reps short do not count. That number is what every decision is made against.
Enough, productive, too much.
Each muscle group has a range, in sets per week. MEV is the least that still makes it grow. The MAV band is where most of your progress happens. MRV is the ceiling, where fatigue builds faster than you can recover from it. Together these are called volume landmarks.
These are the defaults. If you write your own plan it sets the target band, and the floor and ceiling stay where they are. In a week built around heavier weights and fewer sets, the overreaching check uses a lower ceiling.
Under the floor, the model is told to add sets before it adds weight. Over the ceiling, no new exercise gets added. A muscle that sits under its floor for eight weeks gets more work, without you having to spot it.
Reps in reserve matters more than anything else you log.
Reps in reserve is how many more you could have done. Weight and reps on their own cannot tell a hard set from an easy one. That is why it is logged for every set, and why it decides whether you progress.
Effort steps up through the session and goes back to the start when the weight goes up, so you are not pushed to failure on every set. Compound lifts run 3, 2, 1. Isolation and machine work run 2, 2, 1. Your own plan keeps effort level unless it says otherwise.
12, 10, 10 with effort going up is what tiredness looks like. It is not a bad session. The app reads the whole session together, so a normal drop-off never turns into a higher target for the later sets.
RIR is your own estimate, and most people are worse at it than they think. If you never log it, the app has less to go on. It counts an unlogged set as hard, and it will not move a compound lift up without one.
A strength number you never test for.
Estimated 1RM turns any set into one number, so 5 reps at 100 kg and 9 at 85 kg can be compared. No single session is taken as a result. The app fits a line through your best set from each of the last six weeks and uses the slope. Three flat weeks on a lift you are working hard is what it calls a plateau. The strength climb on the Intelligence tab looks back eight weeks instead of six, so the two can disagree about a lift, and each one tells you which window it used.
Load goes up when it is earned.
Reps go up first. Most lifts use double progression. You add reps inside the rep range, and only when you hold the top of the range at the target effort does the weight go up.
A weight increase has to pass all of these. Fail one and the weight stays.
Free weights need three good sessions at the weight. Machines need two. In a high-volume phase it wants one or two more.
Most of them with the planned sets actually done. One session cut short does not hold you back.
Your last set has been close to failure. Machines get one rep more slack. A compound lift with no RIR logged waits.
On the most recent session, on the first working set.
If two or more increases on this lift have failed or been undone in the last four weeks, the app waits. It does not try again.
Three full sessions at 120 kg, last set at RIR 1, top of the rep range reached. The step is capped at the lift's usual increment and rounded to a weight you have actually used on it. No progression in a deload week, on a lift that reads as stalling, or into a muscle that reads as tired, whatever the numbers say.
What overrides the plan.
An automatic deload halves your working sets, takes about a fifth off the bar and asks for three reps in reserve. Your own plan can ask for a lighter cut, anywhere from 30% to 60% of the sets. Either way it is a planned easy week.
This week's work divided by your average over the last few weeks sets your fatigue status. A big jump moves you toward a deload, unless your average is so low that the ratio means nothing.
A session you cut short still counts. It cannot earn a heavier weight, though. The gate wants three quarters of your recent sessions on that lift finished, and a session where you did fewer than three quarters of the sets does not count as finished.
An active injury blocks recommendations that would load that area, and a severe one adds a deload. After three pain-free sessions it drops back to watching, and you can close it yourself whenever you like.
Once your training block has run past its planned length, progressions pause and a deload is suggested until the block resets.
Choices made, and things it will not do.
A muscle that is not getting enough work gets more sets before it gets more weight. That is a reasonable choice. It is not proven best for every goal.
It would rather hold you a week too long than move you up a week too early. If you like to push, you will find it cautious. You can write your own plan instead.
It reads what you log. It does not watch you lift, and nothing here replaces a coach standing next to you.
Calories are not tracked. Bodyweight is read from your profile once a week. If it is dropping, the coach is told that a dip of up to about 3% on a lift is the diet, not a stall, before it suggests a deload. The strength trend itself still comes from the bar. We would rather leave a gap you can see than fill it with a made-up number.
Pain is a reason to see a physio. The app will back your training off, and it will not diagnose anything.
No cardio plans, no mobility work, no sport-specific programming.
Every suggested change shows its reason on the card, and how confident the model was. If you turn one down, the app remembers. Turn down a swap and the lift stays where it is.
