Meet Forge: the thinking behind your next workout

A year of work behind a simple suggestion

We’ve spent over a year designing Forge: studying training research, examining how other workout apps approach programming and progression, and listening to what lifters say when those tools fall short. That included reading complaints and discussions on Reddit, where the details of an ordinary gym session often reveal problems a feature list misses.

Weight jumps that feel impractical. Repetitive workouts. Equipment that doesn’t match the gym. Those frustrations helped shape our priorities: useful variety, respect for your setup, and suggestions you can understand and adjust. Our aim was to build the best science-based algorithm we could, with as much attention to the experience of using it as to the calculations underneath.

The work goes beyond choosing exercises. We’ve built rules for exercise selection, training volume, session timing, history matching, effort feedback and practical weight increments. Then we test awkward cases: a first workout, incomplete history, a long break, an unexpectedly hard set, or a weight you change yourself.

One system, from the plan to the next set

When you ask Forge to build a workout, it chooses exercises and a set structure around your settings, recent training and muscle usage. Auto Progression then works inside a started workout, suggesting weights and reps as you log. You can also use it in a workout you built yourself.

These are different jobs. Building a session determines how much work is in the plan. Auto Progression helps you carry out that plan; it does not add sets or rewrite your completed results.

Build muscle and gain strength use different defaults

Auto Progression reads the fitness goals you saved during onboarding or in settings. When a set has no prescribed rep target, Gain strength uses 4–6 reps and Build muscle uses 8–12. If you select both, the muscle-building default takes priority, just as it does when Forge first chooses a workout goal. Other goals, or no saved goal, use 8–12. These are product defaults, not claims that muscle or strength can only develop in those ranges.

Your program comes first. An existing exact rep target or rep range takes priority over the profile default. A strength program can still prescribe higher-rep accessories. Changing your saved goal updates the fallback for unprescribed sets; it does not rewrite a workout’s existing prescriptions, set count, completed results or manual entries. Both defaults aim for two reps in reserve unless the set specifies another effort target; failure sets target zero.

Why less weight can come with more reps

Imagine your recent history is 80 kg for 6 reps with two reps left. An unprescribed strength set stays in the 4–6 range. With Build muscle selected, the same history is translated into an 8–12 rep target, which can require less weight at the same effort. Compare Gain strength and Build muscle in the simulator to see this with identical history.

Forge does not lower an executable, demonstrated weight merely to move from the bottom to the middle of the same rep range. It keeps that weight when the estimated capacity still supports the minimum reps and target effort. It can reduce the weight when more prescribed reps, a higher reserve target, harder recent performance or fatigue makes it unsuitable. A return after more than six weeks also gets a more cautious starting target, and loads round down to your equipment step.

A lighter target is an adjustment, not proof of lost strength or improved fitness. Compare performance on the same exercise at similar reps and effort over time. More reps can be useful work, but higher weight times reps alone does not establish progress.

The reverse can happen too: more weight with fewer reps. When you reach the top of the range and report at least the requested reserve, Forge can use the next equipment step if the estimated capacity supports it at the bottom of the range. If that step is still too large, it holds the weight. Without effort ratings, the stricter repeated-session rule below applies.

Your history gives Forge a starting point

Auto Progression looks at completed working sets of the same exercise. Each session contributes its strongest working-set evidence, so lighter, fatigued sets later in a workout do not make the next session start weaker. Weight, reps and any effort you reported help estimate a suitable next target. A planned set does not count as something you achieved, and doing more sets does not earn an automatic strength increase.

Recent sessions carry more influence than older ones. A session from six weeks ago has half the weight of equivalent new evidence; at twelve weeks, a quarter. Forge considers up to 24 eligible sessions from the last six months, depending on the history available on your device.

That fading line describes how much we rely on an old log, not how much strength you have lost. After a longer break, Forge also starts more cautiously. With no useful history, Auto Progression keeps your existing targets until a completed working set gives it something to work with.

Today gets a say, too

Ten reps with two more possible tells a different story from ten reps with nothing left. Choose whichever scale is easier for you: reps in reserve (RIR) counts clean reps left; RPE rates effort out of 10. 2 RIR = RPE 8, 1 RIR = RPE 9, and 0 RIR = RPE 10. The simulator and workout loggers use the same conversion. Switching scales keeps the meaning of your entries intact.

The RIR-based effort scale has been studied as a practical way to guide training load. It is useful feedback, even though people’s estimates are imperfect. Research on the effort scale and RIR accuracy.

As you complete sets, Forge adjusts the next unfinished target in that exercise. If a familiar weight turns out much harder than expected, the next suggestion can ease off immediately. It also accounts for the work you have already done, so a lighter later set can be a sensible adjustment.

When you leave effort blank

Leave effort blank if you’re unsure. Forge keeps it unknown. For fatigue only, it uses a moderate 2 RIR / RPE 8 estimate, rather than assuming every unreported set reached failure. That estimate is never stored as something you reported, and it does not add assumed spare reps to your strength estimate. Actual effort feedback takes priority when provided.

You can still progress without rating every set. Before a fresh session, Forge checks your two most recent eligible sessions, at least 24 hours apart and both within six weeks. Each must contain at least two logged working sets, all at the top of the current rep range at the reference load or higher, with no effort ratings. It can then suggest one equipment step, capped at 5%, if the demonstrated capacity supports that weight at fewer reps within the range. A missed ceiling, lighter working set, oversized equipment step or longer break blocks this fallback increase. A reported effort rating uses the effort-aware rules instead.

For example, two qualifying sessions of 80 kg × 12 can support trying 82.5 kg with fewer reps in an 8–12 range. One successful set is not enough. The increase does not repeat just because you reopen the workout; new evidence at the new load must support the next step. Try Progress without effort ratings in the simulator. These thresholds are conservative product rules, not a measurement of your reserve or a guarantee of performance.

Progress does not mean maxing out every set

Our default is to aim for about two reps in reserve unless your workout specifies another effort target. In one eight-week study of trained lifters, stopping with one or two reps left produced similar quadriceps growth to reaching failure, with less acute fatigue. Refalo and colleagues, 2024.

The graph shows Forge’s fatigue estimate, not measurements from that study. Research informs our approach; the exact curves and adjustments are our design choices. The complete algorithm has not been clinically validated, and it cannot measure pain, technique or how ready you feel.

Progress can mean another rep at the same weight before a heavier weight becomes practical. Forge stays inside your workout’s rep range and rounds suggested loads down to the equipment step you choose. Recent performance with reported effort can allow a small increase, but reopening a workout never counts as new progress.

Turn it on, stay in control

1. Open a started workout’s options menu (⋮). Under Set suggestions, switch Last Sets to Auto Progression. Choose a default for new workouts in Workout Settings. These preferences are saved for your account on this device; a choice made for an individual workout takes priority.
2. Choose your load step in kg or lb. Use the change in the weight you log: total weight for a total-load exercise, or per-side weight for an exercise logged per side.
3. Choose RIR or RPE for effort. Tap a set number to report how it felt: two more clean reps is 2 RIR / RPE 8. In the mobile app you can also hold the completion circle, slide to an effort value, and release.
4. Review the suggestion and train. Enter a different weight or rep count whenever you need to. Completing a set accepts the numbers shown; reporting effort helps refine the next target.

Switching back to Last Sets leaves your entries and completed sets intact. Auto Progression currently adapts ordinary weighted rep sets, including failure sets. Warmups, drop sets, myoreps, timed work and assisted or bodyweight exercises keep their existing prescriptions.

See the thinking for yourself

Open the Forge simulator, choose an example, and change one thing. Make the last session older. Leave effort blank. Log a tougher set. Watch the next suggestion change using the same engine as your workouts, without changing your real history.

Run progression checks lets you try the built-in behavior tests with one click. Behind those checks, our automated tests cover manual overrides, completed sets, duplicate history, equipment steps and many combinations of training inputs. We want Forge’s decisions to be understandable, testable and useful when you’re standing in front of the rack.