Foleil builds training around the equipment, days, session length and preferences you provide. Its rules determine valid workouts and progression options. AI can arrange existing exercise options or choose among allowed next targets; it does not have an unrestricted licence to invent loads and repetitions.
Here is what the implemented rules do. This explains the programme's mechanics. It does not establish that Foleil produces better strength gains than Fitbod, or that an example is an expected result.
What shapes the programme?
The programme builder takes a goal, level, split, chosen weekdays, session minutes and equipment. Supported splits require compatible day counts: upper/lower uses an even number of days, push/pull/legs uses a multiple of three, and a body-part split uses five. Custom arrangements are also supported.
Equipment filters restrict eligible exercise choices. Load targets use the equipment available at the chosen place, such as an inventory of loads or known increments. An unavailable load is not treated as attainable. Personal exclusions, safety ceilings and manually pinned targets limit changes.
Saving a programme puts it in the local library. Adding its sessions to the timetable uses a separate scheduling preview, preserving the distinction between a draft and scheduled sessions. Duration is estimated from work, rest and transitions. A selected routine can exceed the requested budget; the builder shows that conflict rather than claiming it fits. An AI-arranged session is checked against its duration budget and supplied exercise options.
How do effort and RIR affect progression?
Foleil stores effort as RPE. Its RIR presentation uses RIR = 10 − RPE, with half-step RPE values from 6 to 10 corresponding to 4 to 0 repetitions in reserve. These are reported estimates, not measurements of muscular failure.
In the automatic strength-candidate rule, an increase requires permission to increase, all working sets recorded, every target reached, and no recorded working-set effort at or above 9.5. Missing effort is not itself treated as a high-effort reading; reaching the targets and the other constraints still matter. The AI's “completed targets” explanation has a stricter check requiring valid effort below that limit for every set.
In the rep-range double-progression rule, effort of 9.5 or more selects a lighter attainable load at or below 90% of the relevant existing load when equipment information allows it. Incomplete or mismatched evidence cannot silently stand in for a completed comparable session.
When do repetitions or load go up?
For a row with a confirmed rep range, double progression keeps earned repetitions set by set. When the upper end has not been reached for every working set, it adds one repetition to the first eligible completed set, within that set's range.
When every working set reaches the upper end at the comparable load, the rule can move to the next attainable load and reset repetitions to the range's lower end. That next load must be no more than the greater of a 10% rise or 0.5 kg above the current load. A manually pinned row stays pinned. Missing equipment information holds the progression instead of inventing a load.
Rows without that confirmed range use the existing history-based target rule. The automatic candidate layer bounds loaded repetitions to the supplied range or its cap, unloaded work to its own cap, and timed increases to five-second steps up to two minutes. It offers at most five distinct choices for a row, for at most twelve rows in a session.
These are separate rule paths. A rep-range worked example should not be mistaken for the behaviour of every generated workout.
What does the AI decide?
For strength progression, the rules create the possible targets first. The AI receives those lettered options, recorded work and relevant permitted context, including recovery and the scheduled week. It returns an option letter and an allowed reason code for each row.
Unknown options, duplicate answers for a row and unsupported reasons are refused. If AI is switched off, unavailable, out of quota, cancelled or too slow, the rules' own choice remains. A later change to the underlying scheduled row also prevents applying a stale answer.
Programme composition is a different task. It arranges supplied exercises into sessions and keeps their existing prescriptions; it does not predict future training loads. The resulting choices must pass the same session-size, catalogue and time checks.
How are volume and lighter weeks handled?
Recorded repetition volume counts completed repetitions multiplied by the recorded load, accounting for whether a load is one item or a pair. The session total sums recorded repetition-set results; it is not a working-set-only tally. Timed and distance work are not relabelled as repetition volume. scheduled volume and completed work remain different figures.
The periodisation engine supports static, linear, reverse and undulating patterns. Its configured lighter-week interval marks lighter weeks explicitly. For affected unpinned working sets, a lighter week uses a load factor of 0.8, rounded down to an attainable load, and reduces an existing target RPE by two within the allowed bounds. Preparation sets remain unchanged. The programme builder also offers a lighter-fourth-week preference; the generated-plan rule reduces working-set count in that path. These are different mechanisms, not a universal automatic deload prescription.
An optional personal weekly-set range needs six complete weeks of data. It uses 80–120% of the median of those weeks, with rounded bounds. Missing weeks do not become zero and the range is not a population recommendation.
An eight-week worked example
Example only: a rule illustration, not a forecast or a personalised programme. One exercise is encountered once per week, with three working sets in an 8–10 rep range, a 40 kg starting load and a next attainable load of 42.5 kg. Every preceding prescription is completed exactly at RPE 8, with confirmed comparable set records, no pins or exclusions, and permission to increase. This example selects the rules' choice, leaves AI selection off and configures no lighter weeks so the progression path is visible.
| Week | Load per set | Repetitions in sets 1, 2, 3 | Assumed record after this exposure |
|---|---|---|---|
| 1 | 40.0 kg | 8, 8, 8 | All targets completed at RPE 8 |
| 2 | 40.0 kg | 9, 8, 8 | All targets completed at RPE 8 |
| 3 | 40.0 kg | 10, 8, 8 | All targets completed at RPE 8 |
| 4 | 40.0 kg | 10, 9, 8 | All targets completed at RPE 8 |
| 5 | 40.0 kg | 10, 10, 8 | All targets completed at RPE 8 |
| 6 | 40.0 kg | 10, 10, 9 | All targets completed at RPE 8 |
| 7 | 40.0 kg | 10, 10, 10 | All targets completed at RPE 8 |
| 8 | 42.5 kg | 8, 8, 8 | All targets completed at RPE 8 |
The table is generated by a transparent transcription of the rep-range rule. Weeks are simply labels for consecutive exposures. Extra sessions, a missed target, different recorded effort, a lighter week, an equipment limit or an AI choice can change the next target. For example, repeating the same load after incomplete evidence is a valid outcome; steady increases are not promised.
What does sports science support?
The ACSM progression position stand (2009) describes progressive resistance training and adjustment to the goal and training status. It supports the general rationale for scheduled progression; it did not study Foleil or validate these exact thresholds.
Robinson and colleagues (2024) analysed estimated proximity to failure and found that its relationship with adaptation differs between strength and hypertrophy. Their analysis cautions that RIR was estimated. It does not establish that Foleil's RPE-to-RIR display is a precise physiological measurement or that its effort cutoff is optimal.
Bell and colleagues (2023) developed expert consensus on deload design. Deloading was described as under-researched. That supports explaining lighter weeks as a programming option, not claiming that a fixed interval or Foleil's exact reduction has proven superior results.
What has not been demonstrated?
These rules make programme behaviour inspectable. We have not published a head-to-head strength study against Fitbod or long-term evidence of improved consistency. A complete daily routine remains Foleil's product proposition; effectiveness requires evidence from real use.
See Ask Sprout's shared context and the nutrition accuracy check for the other parts of that routine.
Frequently asked questions
Does AI invent the next weight?
No. The strength progression rules supply bounded candidate targets and AI can choose a letter from those options. Invalid or unavailable AI responses leave the rules' own choice in place.
Does every session automatically add weight?
No. Progression depends on comparable recorded work, completed targets, effort, equipment, pins and permission to increase. The rule can add a repetition, repeat a target, suggest a lighter load or hold a change.
Is the eight-week table a promise of progress?
No. It assumes every prescribed set is completed at the stated effort and follows one rep-range rule without lighter weeks or AI selection. It illustrates the algorithm rather than predicting a person's results.
Sources
- American College of Sports Medicine. Progression models in resistance training for healthy adults. March 2009. Position stand.
- Robinson ZP and colleagues. Exploring the dose-response relationship between estimated resistance training proximity to failure, strength gain, and muscle hypertrophy. September 2024; online 6 July 2024. Meta-regressions.
- Bell L and colleagues. Integrating deloading into strength and physique sports training programmes. 21 September 2023. Delphi consensus study.
- Foleil's implemented training and programme rules, reviewed 19 October 2026. First-party implementation evidence; no outcome trial is implied.
