SOREN · Sleep · Outcomes · Recovery · Exercises · Nutrition
Meet SOREN — your personal research engine
SOREN uses machine learning to study your training, recovery, sleep, nutrition, and performance over time — running the statistical equivalent of an n-of-1 study on you.
No chatbot. No AI-generated workouts. No invented conclusions.
What is SOREN?
SOREN is the research engine inside Mass in Motion. It runs the statistical equivalent of a study with a single participant — you — over the training you have already logged, and reports what actually holds up.
It is not a chatbot and it does not give opinions. It does not generate your workouts. Every line it shows you is a measured result from your own data, and when it does not have an answer yet, it says so.
S · O · R · E · N
The five signals it reads
Five streams, named up front. SOREN reads these and nothing outside them.
Sleep
Duration, timing, and how you sleep after training.
Outcomes
What your sessions actually produced — lifts, paces, effort.
Recovery
Readiness, resting heart rate, HRV, soreness, and load.
Exercises
The movements you did, how they were loaded, and how you responded.
Nutrition
What and when you ate and drank around your training.
How it learns
SOREN doesn’t jump to conclusions
01
Collect
Training, sleep, recovery, nutrition, and performance data, from the sessions you actually logged.
02
Compare
SOREN compares similar situations against each other instead of treating every workout as identical. You are your own control group.
03
Learn
Potential relationships start to appear as evidence accumulates, and SOREN shows you which ones it is still working on.
04
Validate
Only the patterns that keep holding up, after a correction for the ones that look real by chance, become personal insights.
What it can discover
The kind of thing SOREN looks for
SOREN reports relationships, not causes, and it says which rung of the ladder each one is standing on.
Meal timing before sleep
So far, when the gap between your last meal and when you fell asleep was higher, your sleep that night has tended to be better by about 35 minutes, across 7 weeks of your data. The early signal has been consistent, but we need more of your data to confirm it.
What was comparedYour nights after a 4–5 hour gap, measured against your own nights after a 2–3 hour gap. 23 nights, spread over 51 days.
Movement Response: 48+ hours since a hard lower-body lift
Following 48+ hours since a hard lower-body lift, estimated strength carryover averaged +6.0%.
What was compared19 of your sessions at 48+ hours of rest, against 12 of your own sessions under 36 hours, over 118 days. It survived the false-discovery-rate correction.
Caffeine and your sleep
So far, when the caffeine still in your system at bedtime was higher, your sleep that night has tended to be lower by about 31 minutes, across 5 weeks of your data. We need more of your data to confirm this early pattern.
What was comparedYour nights following an afternoon coffee, against your own nights without one. 14 nights, spread over 37 days.
A pattern that has not shown up yet
So far, when your training volume over the previous week was higher, your next top-set strength hasn’t shown a clear change, across 3 months of your data. The early signal has been consistent, but we need more of your data to confirm it.
What was comparedA reported non-effect is a result, not a gap. Knowing a lever does nothing for you is what stops you spending months pulling it.
Running and lifting
Still learning how your running and lifting sessions interact.
What was comparedNeeds at least 6 comparable observations, 3 of them at repeated exposure levels, spread across a minimum of 10 days.
Example SOREN insights. Written in the wording the app itself would render, with demonstration figures. The effect sizes and session counts are illustrative, not one person’s real result presented as a typical one — SOREN only ever shows you findings that came out of your own logged training.
Emerging to validated
A pattern has to earn its way up
Nothing arrives as a conclusion. Every observation starts at the bottom of this ladder and only moves when your own data pushes it there — and it can move back down.
01
Still learning
Not enough comparable evidence to say anything yet.
SOREN has noticed something, but it has not seen it enough times, across enough weeks, to say anything about it. It says so instead of guessing.
02
Emerging
Showing up in your data, not confirmed yet.
The relationship has now appeared across several sessions and is holding up as more data arrives. It is shown to you with the evidence behind it, and labelled as unconfirmed.
03
Validated
Repeated enough times in your own data to rely on.
Enough evidence has accumulated, and survived a false-discovery-rate correction, for SOREN to treat the pattern as meaningful for you specifically.
It compares like with like
Sessions where something was present are measured against your own sessions where it was not. You are your own control group.
It waits for repetition
A pattern has to hold across enough sessions, spread over enough time, before it is allowed to say anything at all.
It corrects for luck
Test enough relationships and a few look real by chance. SOREN applies a false-discovery-rate correction across the whole set at once, so a coincidence does not get promoted to a finding.
It changes its mind
A finding is withdrawn or replaced when newer data stops supporting it, and every card carries the confidence behind it.
Experiments · In development
Structured n-of-1 experiments
Watching your history can only tell you what already happened. SOREN is being built to test individual hypotheses using controlled protocols based on your own baseline data: one variable changed on purpose, everything else held still, measured against you rather than against an average.
Protocols are reviewed before being made available, and none are currently active.
Example protocol currently under review
Does eating a complete meal 60–90 minutes vs 150–180 minutes prior to training affect session energy and top set output?
Condition A
Finish a complete meal 60–90 minutes before training.
Condition B
Finish the same meal 150–180 minutes before training.
- Baseline required
- 8 comparable sessions
- Protocol length
- 12 sessions
- Assignment
- 2×2 crossover blocks
- Measures
- Change in your top-set output
How one would run
01
A baseline first
A protocol will not start until you have at least 8 comparable sessions behind you. Without a baseline there is nothing to measure a change against.
02
Two conditions, alternated
You run condition A and condition B in a 2×2 crossover across 12 sessions, so the order of the blocks cannot masquerade as the effect.
03
Adherence is checked, not assumed
Each session is scored against the protocol window, with a 15-minute tolerance. Sessions that drifted outside it are counted as such rather than quietly included.
04
It can stop early
You can cancel at any point, and the protocol stops itself on an unscheduled deload, an injury, or repeated discomfort.
Other protocols under review
Nutrition
Pre-workout carbohydrate timing
Does consuming carbohydrates 30–60 minutes vs 90–120 minutes pre-session improve your lower-body training performance?
Nutrition
Pre-workout caffeine timing
Does caffeine consumed 30–45 minutes vs 60–75 minutes prior to lifting lead to higher strength expression and readiness?
Training
Warm-up duration and specificity
Does an extended 12–15 minute dynamic warm-up vs a concise 5-minute warm-up enhance top-set performance without inducing fatigue?
What an experiment can never be
Machine learning is structurally forbidden from inventing an intervention. Every experiment must come from a versioned registry that a human wrote and approved, and these are excluded from it outright:
- Prescription medication or dosage changes
- Dangerous supplements or unapproved stimulants
- Intentional dehydration or sleep deprivation
- Injury aggravation or pain exposure
- Extreme caloric restriction or unsafe fasting
- Extreme training overload
When a protocol is approved and your baseline training history is large enough, SOREN will be able to propose the ones your own data has a question about. Running one would always be your choice, and you would be able to stop it at any point.
The distinction
Machine learning without generative AI
SOREN uses machine learning to recognize patterns and model how an individual athlete responds over time. It does not use a language model to invent workouts, recommendations, or conclusions.
| SOREN | Generative AI fitness apps |
|---|---|
| Analyzes your recorded data | Generates content |
| Looks for measurable relationships | Produces conversational answers |
| Builds evidence over time | Can answer immediately |
| Shows its uncertainty | Often gives a definitive response |
| Studies the individual | Usually relies on broad instructions |
| Does not write your workout | Can generate workouts |
Both are useful tools. They are built for different jobs, and only one of them belongs anywhere near a training prescription. ML learns. Algorithms program. Exercise selection, sets, reps, load, progression, running prescriptions, and scheduling all stay under the deterministic engine described in the methodology.
Not this
What SOREN is not
- Generate your workouts
- Write or rewrite your training program
- Act as an AI chatbot
- Invent recommendations that the numbers do not support
- Override the deterministic programming algorithms
- Claim a pattern is a cause
Your data
You decide whether SOREN runs at all
Personalized Coaching is a permission you grant, and you can withdraw it at any time in Settings > Privacy & Data. When you do, new analysis stops, active findings are removed from display, and your programs, logging, recovery, and nutrition keep working exactly as before.
Using SOREN for your own coaching is separate from the optional consent to include eligible data in population research. Agreeing to one does not enroll you in the other. Personalized findings are historical statistical observations: they do not guarantee outcomes and are not medical advice — see the medical disclaimer.
Personalized Coaching & Data Use NoticeML & Algorithm Improvement ConsentPrivacy Policy
Early access
Carry the momentum
Join the list and we will email you the launch date and your founder pricing details. Founder pricing is $64.99/year for the first 50 eligible members, against $89.99 normally.
Want to train on it now? Beta testers get the build before launch. Four questions, about thirty seconds.
Apply for the beta →