Workout progression
The live workout offers a small, editable recommendation for eligible load-and-repetition exercises. Its purpose is to pre-fill ordinary weight and repetition fields at the useful moment; it is not a separate workout flow.
What the user sees#
With Training Autonomy set to Suggest, the app fills only the first open working set when comparable history is available. The user can overwrite either field and completes the set with the usual check action. There are no policy selectors, completion reasons, review cards, confirmation dialogs, equipment configuration controls, or progression-specific menus in a live workout.
After a working set is completed, only the directly following open working set gets its target. Its completed predecessor is the evidence for that one prescription. A value entered by the user always wins over a generated value, including if an asynchronous history lookup finishes afterwards.
Cardio, duration, distance, variable-load and other exercises without a stable load-and-repetition axis continue to log normally, but receive no strength progression recommendation.
Recommendation rule#
Only the first working set of the most recent completed comparable workout is used to start the next one. Warm-up and drop sets are excluded. Exercise matching uses the canonical exercise UUID, with the existing legacy name fallback for old records.
For a repetition range [L, U], the next first set is simple double progression:
| Previous first set | Suggested next first set |
|---|---|
Below L repetitions | Hold the load and target L |
L through U - 1 repetitions | Hold the load and add one repetition |
At least U repetitions | Advance one standard equipment increment and target L |
For a fixed repetition target, the app projects a load capable of that target from the previous first set's estimated 1RM, while limiting an increase to one standard equipment increment. A recorded RIR contributes to that estimate as additional repetitions to failure; a missing RIR uses only the performed repetitions. Without a repetition target it repeats the prior load and leaves repetitions empty.
For each later set, the app calculates a fresh Brzycki capacity from the immediately preceding real set. Missing RIR or RIR 0 retain 95% capacity for the next set; RIR 1 through 4+ retain 96% through 99%. The next test load is normally the previous real load. When routine weights deliberately define a different positive next-set weight, their ratio to the first template weight is applied to today's real first-set load and rounded to a usable increment before calculating repetitions.
For a range [L, U], the calculated repetitions determine the next target:
| Projection at the rounded test load | Suggested next set |
|---|---|
Above U | One normal increment harder, U reps |
L through U | Hold the test load, projected reps |
Below L | Lower to the highest usable load capable of L reps, L reps |
The engine does not use Brzycki beyond 12 effective repetitions. In that case it holds the test load unless the actual result reached the range maximum, when it advances by exactly one increment. A result of at most three effective reps in a range beginning at six or more reps holds the test load and targets the lower bound, avoiding an implausible deload after insufficient evidence.
Bodyweight exercises receive repetition targets only. Assisted and weighted-bodyweight exercises retain their own load semantics. Assisted work uses effective resistance when body weight is available. Without body weight, it uses a discrete rule: less assistance at the range maximum, more assistance below the range, and the same assistance with one more rep inside the range. Generic equipment floors prevent suggestions below an unloaded standard barbell, the smallest dumbbell, cable or machine increment. The database does not yet store a user's specific gym inventory, so these floors are deliberately conservative class defaults rather than claims about a particular facility.
The recommendation is deliberately a starting point. It does not diagnose fatigue, pain, form, recovery or readiness, and it never changes the routine.
Persistence and updates#
Saving a workout writes every retained set and the completed workout record in one SQLite transaction, then verifies the retained rows. Untouched empty rows and unfinished generated placeholders are discarded, so workout history shows only used sets. Manually entered open sets and all completed sets are retained. This avoids the former delete-before-save path that could lose entered sets if finalization was interrupted.
The simplification does not require a schema migration. Existing progression metadata remains readable for backup, restore and historical records, while new workouts use algorithm version progression_v1.8_jit.