A feature is not research-backed just because it exists. Each one below carries an explicit decision: existing evidence applies, new evidence was reviewed and added, no research is needed, or the evidence is insufficient or conflicting. Numbers and thresholds that go beyond the papers are labelled as FitX AI's own choices.
Training
RPE mesocycle
Effort rises week to week across a 5-week block: RPE 7, 8, 9, then 10 on the last set, then a deload.
Existing evidence applies Proximity to failure, RPE/RIR scales and periodization are covered by the library.
Papers Refalo et al., 2023, Robinson et al., 2024, Halperin et al., 2022, Zourdos et al., 2016, Helms et al., 2016, Moesgaard et al., 2022
FitX AI's own choices. The 7, 8, 9, 10 ladder and the 5-week length are FitX AI design choices. The papers support training near failure and RPE/RIR as effort scales, not this exact schedule.
Reps-left feedback
After each set you tap how many reps you had left, and the next targets follow what you actually did.
Existing evidence applies RIR/RPE validity and accuracy are covered by the library.
Papers Robinson et al., 2024, Halperin et al., 2022, Zourdos et al., 2016, Helms et al., 2016
FitX AI's own choices. Reps-left is self-reported; accuracy varies between people, so one rating moves targets only a little.
8-15 working range
Targets stay inside a default 8-15 rep range, chosen as a practical range rather than a biological limit. Barbell squats and hinges default to 6-10 and stop one rep short of failure, even in the RPE 10 week.
Existing evidence applies The repetition-continuum review (Schoenfeld 2021) and the load meta-analyses in the library find similar muscle growth across a wide range of loads when sets are taken close to failure; the proximity-to-failure meta-analyses find that going all the way to failure is not needed for growth.
Papers Schoenfeld et al., 2015, Schoenfeld et al., 2017, Lopez et al., 2021, Lasevicius et al., 2018, Refalo et al., 2023, Robinson et al., 2024, ACSM, 2009, Plotkin et al., 2022, Schoenfeld et al., 2021
FitX AI's own choices. 8-15 is a practical default range, not a research-derived limit. 6-10 for barbell squats and hinges, and RPE 9 as their ceiling, are FitX AI choices to limit the fatigue of the most demanding lifts: no paper sets these numbers. A range the user has set is never changed.
Load progression
Reaching the top of the range at the planned effort adds at least 2% (5% for leg and hip lifts) to the weight.
Existing evidence applies Progressive overload guidance (ACSM 2009) and the progressive-overload trial are in the library.
Papers ACSM, 2009, Plotkin et al., 2022
FitX AI's own choices. The 2% and 5% floors and the rep-ceiling rule are FitX AI choices inside the ACSM's 2-10% guidance.
Estimated 1RM
Every rated set gives an estimated one-rep max from the weight and the reps you could have done, used for targets and progress charts.
Existing evidence applies 1RM prediction accuracy and reps at a given percentage of 1RM are covered by the library.
Papers Reynolds et al., 2006, Nuzzo et al., 2024
FitX AI's own choices. Estimates are approximate; they vary by exercise and person.
Careful heavy-set estimate
A heavy low-rep set is read with the middle of four common formulas before it sets a lighter weight.
Existing evidence applies Between-person variation in reps at a given load is documented in the library.
Papers Reynolds et al., 2006, Nuzzo et al., 2024
FitX AI's own choices. Taking the median of four formulas is an engineering choice, not a validated method.
Weekly sets per muscle
Plans give each muscle a weekly number of sets for your experience, counting half a set for helping muscles.
Existing evidence applies Volume dose-response, trained-lifter ranges and diminishing returns are covered by the library.
Papers Schoenfeld et al., 2017, Pelland et al., 2025, Baz-Valle et al., 2022, Camargo et al., 2025, Aube et al., 2022, Mannarino et al., 2021
FitX AI's own choices. The bands (8-12, 12-16 or 16-20 sets) and the half-set counting are FitX AI choices; the research gives trends, not individual prescriptions.
Once or twice a week
Each muscle is trained twice a week by default, or once if you prefer.
Existing evidence applies With weekly sets matched, frequency meta-analyses show similar growth for once or twice a week.
Papers Pelland et al., 2025, Schoenfeld et al., 2016, Schoenfeld et al., 2019
FitX AI's own choices. Twice a week is a default that spreads the sets, not a requirement.
Per-session ceiling
No muscle gets more than 10 sets in one session; the rest is moved to another day. This is a practical limit, not a research finding.
Evidence insufficient or conflicting The one study found on per-session volume is a preprint with no abstract in Crossref, PubMed or Europe PMC, so it could not be verified or summarised and is tracked as a candidate.
Limits. No verified, peer-reviewed evidence supports the 10-set ceiling. It is a practical product limit.
FitX AI's own choices. 10 sets per muscle per session is a product choice.
Planned exercise variety
The second weekly session for a muscle uses a different exercise when your equipment allows.
Existing evidence applies Exercise variation is covered by a systematic review in the library.
Papers Kassiano et al., 2022
FitX AI's own choices. Rotating the second session's exercise is a product choice that fits the review's direction; the review does not prescribe it.
Session length
Plans fit your session length, estimating each set with its rest (about 2 minutes for big lifts, 90 seconds for small ones).
Existing evidence applies The inter-set rest meta-analysis in the library supports resting more than a minute.
Papers Singer et al., 2024, Grgic et al., 2018, Grgic et al., 2017
FitX AI's own choices. The per-set time estimates (2.75 and 2.25 minutes, plus warm-up) are planning figures, not research findings.
Rest timer
After a set, a timer shows a recommended rest (3 minutes after heavy compound lifts, 2.5 after other compound lifts, 2 after isolation exercises) that you can end early.
New evidence reviewed and added Rest-interval evidence was reviewed and added: a Bayesian meta-analysis for muscle growth, two systematic reviews (strength; hypertrophy) and a trial in trained men.
Papers Singer et al., 2024, Grgic et al., 2018, Grgic et al., 2017, Schoenfeld et al., 2016
Limits. The evidence is mixed. For muscle growth alone, the pooled analysis found little difference beyond about 90 seconds, while trained lifters gained more strength (and in one trial more thigh growth) with rests of about 2-3 minutes. No paper sets different rests for heavy compound, other compound and isolation exercises.
FitX AI's own choices. The 180, 150 and 120 second defaults and the three exercise categories are FitX AI product choices; the timer is a recommendation and never forces a rest.
Deload week
Week 5 keeps the weight, halves the sets and lowers the effort.
Existing evidence applies A consensus on deloading is in the library.
Papers Bell et al., 2023
FitX AI's own choices. Halving the sets in week 5 is a FitX AI choice; the consensus describes practice and does not prescribe it.
End-of-block review
Each block's sessions set the next block's starting point, with increases capped at 10%.
Existing evidence applies Periodization and progressive overload papers are in the library.
Papers Moesgaard et al., 2022
FitX AI's own choices. The 10% cap and the review rules are FitX AI choices. Periodization trials compare programme designs; they do not validate this method.
Day split selection
Your days are split into upper/lower, full-body or similar templates; each is tried against your days, time, equipment and experience, and the better fit is kept.
Evidence insufficient or conflicting The frequency papers support training each muscle twice a week; no reviewed paper compares split designs at equal weekly sets and frequency.
Limits. No evidence ranks one split above another. The templates and the selector's thresholds are product heuristics.
FitX AI's own choices. Templates, the 2-muscle and 5-point selection thresholds.
Session check
Each day in Edit plan shows its working sets and main muscles, and points out what may make it more tiring than it needs to be: long near-failure sets on heavy squats and hinges, too many of them in one session, more than about 10 sets for one muscle, an isolation exercise before the main lift, or several exercises doing the same job. It never changes your plan.
Existing evidence applies Similar growth across loads when sets are hard (load meta-analyses, repetition continuum), failure not being needed (proximity-to-failure meta-analyses), the per-session ceiling (session-cap) and longer rest on demanding lifts (rest-interval reviews) are in the library.
Papers Schoenfeld et al., 2017, Lopez et al., 2021, Refalo et al., 2023, Singer et al., 2024
FitX AI's own choices. Flagging a heavy squat or hinge with a top of range above 12 and 3+ sets, more than 6 hard sets of those lifts in a session, and three same-movement exercises for one muscle are FitX AI heuristics. Exercise order (early exercises get more reps and progress more for strength) is from reviews logged as candidates, not yet in the library. Fatigue varies by person, so these are pointers, never limits.
Next-workout status and stall check
Each exercise says in plain words whether the next workout is a step up, the same, an easier load, a restart after a break or a deload, what the last comparable session was and why; a stall check notes when estimated strength has not risen across several rated sessions. It explains the engine's targets and never changes them.
Evidence insufficient or conflicting The status and explanations only restate the engine's own targets (estimated 1RM, RPE and rep-range rules already in the library). The stall check and the repeat/restart rules after missed weeks rest on FitX AI heuristics: no reviewed paper gives a threshold for a stall in recreational lifters' estimated 1RM, and a candidate review of strength loss after time off (Bosquet 2013) is logged but not audited, so no number comes from it.
Limits. Estimated 1RM from submaximal sets varies from day to day and depends on how accurately reps left are rated; the check says possible, never diagnoses recovery or readiness, and suggests only small next steps.
FitX AI's own choices. A possible plateau needs at least 4 rated, comparable sessions spread over 3 or more weeks with no gap over 3 weeks, and the best estimated 1RM of the newest 2 within +1% / -3% of the best of the ones before them; one lower session is read as normal variation; a missed week repeats the last trained effort and 2 or more weeks in a row without training the exercise restart at RPE 7, after which the ladder is picked up one rung at a time. All are configurable FitX AI product heuristics, not validated cut-offs.
Exercise history and targets from it
Your logged sets stay with each exercise whatever happens to your plans, and an exercise coming back into a plan starts from what you last did, worked out again at the week's effort.
Existing evidence applies The starting point is an estimated 1RM from reps to failure (reps done plus reps left), the same model as the e1rm feature, re-fitted to the rep range at week 1's effort. Reps-left estimates are imprecise for a single set, so the median of up to three rated sessions is used and one set is never taken alone.
Papers Halperin et al., 2022, Reynolds et al., 2006
FitX AI's own choices. Pooling sessions within six weeks of the newest, keeping 95% of the estimate after 29-90 days without the exercise and 90% after longer, and recalibrating after 90 days or with only one rated session are FitX AI choices. Strength is lost gradually with time off and by different amounts for different people; the library has no paper that sets these numbers. Sets without an effort rating are read as to failure (the most conservative reading); an effort is never filled in.
Warm-up, ramp-up sets and optional cool-down
Each workout suggests a short general warm-up, a few dynamic movements for the day's lifts, easy ramp-up sets before the first demanding lift, and an optional cool-down. None of it is logged or changes your targets.
Evidence insufficient or conflicting A systematic review with meta-analysis found warm-ups improved performance on most measures studied (Fradkin et al. 2010). Long static stretching just before effort can slightly reduce performance, and dynamic movement doesn't (Behm et al. 2016), which is why the warm-up is dynamic. Cool-downs and post-exercise stretching have no reliable effect on soreness, injury or recovery (Van Hooren and Peake 2018; Afonso et al. 2021), so the cool-down is offered as optional and makes no such claim.
Limits. Evidence for the exact ramp-up pattern before resistance lifts is limited, and the review evidence for warm-ups is mostly short trials in athletes. These papers are not yet in the audited research library: verify them there before citing them on the site.
FitX AI's own choices. The 3-5 minute general warm-up, the movements chosen for each pattern, the ramp-up fractions (40/60/80% for 8/5/3 reps; 50/75% for beginners), at most three ramped exercises a day, and the 3-5 minute cool-down are FitX AI choices. Ramp-up sets are never logged and never count toward volume, effort, estimated 1RM, bests or progression.
Starting weight from a related exercise
An exercise you have never logged starts from the closest one you have (same movement and main muscle), not from a generic estimate, and your first session fine-tunes it.
Evidence insufficient or conflicting The donor's strength is the same estimated 1RM from reps to failure that exercise-history uses (median of up to three rated sessions).
Limits. No study gives a reliable ratio between two different exercises or kinds of equipment for a given person, and machine and cable loads depend on the make, so the carried-over number can only be a rough starting point.
FitX AI's own choices. The equipment ratios (0.55-1 of the donor's estimate), matching on movement, main muscle, class and side, and recalibrating in the first session are FitX AI choices, set on the light side.
