Adaptive Training Intelligence: The Load Signal Your Wearable Isn't Showing You

You just finished a heavy squat session. Your wearable says your readiness is 68. Your lift log says this week’s lower-body volume is sitting at 18 hard sets and trending up. Your last deload was 11 weeks ago.

Should you do the back-off bench session you had planned for tomorrow? Or swap it for something lighter? Or skip it entirely and push it to Thursday?

A readiness score can’t answer that. It wasn’t designed to.

---

The Gap Between a Readiness Score and a Training Decision

Every major consumer wearable now ships a daily training-readiness number. Oura calls it Readiness. Garmin calls it Training Readiness. WHOOP calls it Recovery. They are all built from a similar recipe: overnight HRV, resting heart rate, sleep duration and quality, and — on the watches — some notion of recent training load. The output is a single 0–100 score that tells you something about how well-recovered your autonomic nervous system looks this morning.

The scores are useful. They catch illness before you notice symptoms. They catch the tail end of a bad trip. They flag a week where the training block has outrun recovery. The problem is that they answer the wrong question.

When you plan a training session, you are not asking “how recovered am I?” You are asking a longer, more specific chain of questions: given how recovered I am, given what I did yesterday and last week, given how my chest versus my back versus my legs specifically recover, given how close to failure I should be training, given where I am in my load progression on bench versus squat versus deadlift — what exact session should I do today?

A readiness score collapses all of that into a single number and hands you the collapsing. You still have to do the chain-of-reasoning yourself, and most people skip several links. The usual failure modes are predictable:

• A high readiness day after a hard session yesterday gets interpreted as “green light,” and the same muscle group gets trained again 36 hours later.
• A medium readiness day during a deload gets read as “push harder” because the number is up from yesterday.
• A low readiness day on a scheduled deadlift session gets overridden because “I feel fine.”
• A string of medium readiness days during a high-volume block gets ignored because the color never turns red.

None of these mistakes are the score’s fault. The score is honest about what it represents: a snapshot of autonomic recovery. The mistakes happen because the score is being asked to be a prescription, and it isn’t one.

Turning the score into an actual prescription requires a second layer. That layer needs to know which muscle group you trained yesterday, how close to failure you took those sets, what your personal weekly ceiling is for that muscle, what your current position on the progression curve for each major lift is, and how your body specifically handles accumulated training stress. That layer — adaptive training intelligence — is what we are going to spend the rest of this post describing.

It is also worth being clear up front about what this is not. Omnio’s adaptive training work is generalized methodology, not a prescriptive AI coach. It narrows the decision space. It makes explicit the signals you already care about. It catches the days when the plan and the body are out of sync. It does not replace a human coach’s eye for technique, injury management, or the nonverbal parts of knowing when an athlete is off. The rest is data literacy, and this post is a walk through what that literacy looks like.

For context on the other clusters in this series: we cover composite scores and their confidence, nutrition intelligence beyond calories and macros, and energy availability and RED-S in companion pillars.

Why Generic Volume Rules Fail Individuals

The modern strength literature converges on volume landmarks: MV (maintenance volume), MEV (minimum effective volume), MAV (maximum adaptive volume), and MRV (maximum recoverable volume). Published tables give ranges — typically 8 to 10 hard sets per muscle per week for MEV, 12 to 20 for MAV, 16 to 25+ for MRV — and a decade of evidence-based programming has used those numbers as a starting point. They are a meaningful improvement over “train until you can’t.”

The ranges fail, though, when they are applied as prescriptions rather than starting points. Three failure modes are common enough to name.

Failure mode one: population ranges hide individual variance. A 12-set MAV for chest is the median response. Around that median, individuals range from roughly 8 to 20 depending on training age, sex, recovery capacity, stress exposure, and genetics. A 45-year-old office worker who sleeps 6.5 hours a night under high job stress and a 28-year-old athlete who sleeps 8.5 hours with no other training demands do not have the same MAV, even if both have been lifting for five years. The population-median prescription works for neither.

Failure mode two: the ranges were measured on populations that don’t include you. Most volume-landmark research is on young, trained, mostly-male, mostly-healthy subjects. Generalizing those ranges to women, older adults, people with chronic conditions, people returning from injury, people coming off deloads of unusual length, or people stacking training on top of another sport is an extrapolation that the papers themselves don’t support. In the best case the numbers are right on average. In the worst case they are systematically off by 20% and nobody notices because the user just accepts that they are “injury-prone” or “a slow recoverer.”

Failure mode three: the ranges are static in literature but dynamic in life. Your MAV for chest in April — well-rested, eating well, light work schedule — is not the same as your MAV in November when you’re sleep-deprived from a new baby, under stress at work, and traveling twice a month. Using April’s number in November gets you injured. Using November’s number in April leaves progress on the table.

The usual resolution is “add RPE to the program and let the lifter auto-regulate.” This helps, but it shifts the error rather than eliminating it. RPE drifts across weeks in ways that are hard to notice from the inside, and the RPE of the lifter who is already mildly overreached is biased downward — they believe they are training at 8 when the objective effort is 9.5. Without a calibration layer that compares self-reported RPE to concrete outcomes like rep velocity, the drift compounds into the volume prescription.

The fix is not better tables. The fix is personalization: each lifter’s own per-muscle ceiling, learned from their own training history, updated continuously, with the system being explicit about how confident it is in the estimate.

Maximum Adaptive Volume: Per-Muscle Bayesian Ceilings, Explained Without Math

The concept of Maximum Adaptive Volume (MAV) already exists in the literature. The translation to a personalized number is where Bayesian methods come in.

Here is the idea without the equations. For each muscle group — chest, back, quads, hamstrings, shoulders, arms, and so on — we want to know the largest weekly hard-set count you can currently handle and still recover for next week. Call that your personal MAV. The number you want is not a single point estimate; it is a credible range, with a lower bound that is the conservative “this is almost certainly safe” volume and an upper bound that is the optimistic ceiling.

A Bayesian approach builds that credible range by combining two sources of information.

The prior. Before the system has any of your training data, it has to start somewhere. The prior is the best-evidence population distribution from strength literature — for chest, say, a median around 12 to 14 hard sets per week, with a wide band around that median. The prior is deliberately wide because population variance is high. The point of the prior is not to be right; it is to be less wrong than starting from nothing.

The evidence. Every week of your training logs shifts the prior. Weeks where you successfully completed all hard sets and recovered by the next session — defined by your own post-session markers, like HRV rebound, sleep, and subjective fatigue — pull the estimate up. Weeks where you missed reps, got sick, or stalled progression pull it down. The width of the band shrinks as more evidence arrives: a lifter with 4 weeks of data has a wider band than one with 20 weeks.

The lower confidence bound. This is the part that makes the system conservative by design. Instead of prescribing from the median of the posterior distribution — which would be a 50/50 bet — the prescription uses a lower percentile, typically the 10th or 25th. The system is saying: based on your history and population priors, there is a 90% chance your true MAV is at least this number. Prescribing at that number biases toward “definitely recoverable” rather than “probably adaptive.”

In plain English: the system gives you a volume target it is pretty sure your body can handle, not one it hopes your body can handle. The cost is a slight conservatism — you might be leaving a set or two on the table compared to your true ceiling. The benefit is that weeks that exceed the prescription rarely blow up into injury or multi-week overreach.

The implementation detail that matters for interpretation: the estimate is per muscle, not per workout. Chest MAV and back MAV are different numbers on different update schedules. A chest-heavy week followed by a back-heavy week isn’t the same kind of load as two mixed weeks, and a system that treats them the same is giving up information.

The other detail that matters: the estimate updates with your life. A new baby, a job change, a 70-hour work week, a bout of illness — any of these move the posterior. The system doesn’t reset to the prior, but it widens the credible band, flags the change, and recomputes with updated uncertainty. A lifter who logs and lifts for two years has a tight, personalized MAV. A lifter who logs and lifts for two years with two significant life disruptions has a slightly looser but still-personalized MAV. The estimate tracks reality.

This is the generalized idea. Implementations differ on the prior’s exact shape, how aggressively weeks update the posterior, and what percentile of the credible interval gets used. What is universal is that the prescription is a personalized range, not a population range, and the range is narrower for lifters with more data.

For a longer writeup on the per-muscle specifics, see our post on Maximum Adaptive Volume.

Recovery Half-Lives: Why Some Muscles Need 36 Hours and Others 72

Volume ceilings assume a recovery window. The window is different for different muscles.

Consumer training apps usually pick a single recovery constant — 48 hours between sessions for a muscle group, give or take — and apply it across the board. The assumption is convenient but wrong. Per-muscle recovery differs by a factor of roughly two in published research, and the pattern is stable enough to name.

Fast-recovering muscles. Shoulders, calves, forearms, and small upper-back muscles typically return to baseline in 24 to 36 hours after a moderately hard session. They can be trained three to five times a week in experienced lifters, and the limiting factor is usually tendon rather than muscle.

Medium-recovering muscles. Chest, biceps, triceps, and mid-back recover in 36 to 60 hours. Two to three sessions per week per muscle is the useful operating range. Heavier sessions push the recovery time up; higher-rep sessions at lower loads often recover faster.

Slow-recovering muscles. Quads, hamstrings, glutes, lats, and the lower back all routinely take 48 to 96 hours depending on session intensity. A hard squat session at RPE 9 can leave quads sore and under-recovered for 72 to 96 hours in a way that a comparable-volume chest session does not.

The reason these differ is structural. Larger muscle groups produce more muscle damage per session, recover through slower mechanisms, and tax more systemic recovery resources (glycogen depletion, CNS fatigue, mechanical tension on connective tissue). Slow-twitch-dominant muscles tend to recover faster than fast-twitch-dominant ones. Muscles with long lever arms in their primary movements (quads in a squat, hamstrings in a deadlift) accumulate more eccentric damage than muscles in shorter-range movements.

An adaptive training system encodes this by attaching a per-muscle recovery half-life — the time it takes for the fatigue signal from a session to decay to half its initial value. Half-life is a more useful framing than a hard recovery window because the decay is exponential: 80% of recovery typically happens in two half-lives, 95% in three. A chest session with a 30-hour half-life is roughly 80% recovered at 60 hours and 95% recovered at 90. A quad session with a 60-hour half-life is 80% recovered at 120 and 95% at 180.

The numbers aren’t static across individuals, either. Older lifters have longer half-lives than younger. High-stress life periods lengthen the half-life for everyone. Sleep debt lengthens it. Underfueling lengthens it significantly — which is one of the pathways where nutrition intelligence and training intelligence intersect.

The practical output: on a given day, the system can tell you that your chest is 92% recovered from Monday’s session but your quads are only 58% recovered from Tuesday’s squats. A dashboard reading just “recovered” or “not recovered” is hiding both numbers.

More on this in the per-muscle recovery half-life spoke.

RPE Calibration: Learning Your Personal Effort Scale

Rate of Perceived Exertion is the bridge between objective load and subjective effort, and it’s the scale that makes auto-regulated programs work. The problem is that self-reported RPE is drift-prone in ways that are hard to notice from the inside.

A lifter who reports RPE 9 in week one and RPE 9 in week six may be reporting two very different things. In week one, the RPE 9 corresponds to an actual bar velocity 30% below their top set. In week six — maybe under more life stress, maybe mildly underrecovered from the accumulated block — the RPE 9 corresponds to a velocity 45% below their top set. The lifter feels the same 9. Their body is working noticeably harder.

That drift is not a failure of the lifter. It’s a structural property of subjective effort scales. Effort is context-dependent, mood-dependent, and calibrated against recent memory rather than against an objective reference. Without a separate layer, every downstream calculation that uses RPE — set-by-set adjustment, load progression, volume tolerance estimates — inherits the drift.

Calibration is the fix. The idea is to maintain an objective anchor point against which RPE is regularly checked, and to apply a per-user correction when the two drift apart.

Objective anchors can be:

• Bar velocity. A set ending at a specific velocity loss (say, 20%) is a more consistent definition of “near failure” than self-reported RPE 9.
• Reps in reserve vs actual reps. A calibration set where the lifter predicts RIR 2 and then is instructed to go to failure, and the actual reps completed are logged. A calibrated lifter predicts within 1 rep. A drifting lifter predicts 2, completes 5.
• Next-day recovery. A set truly at RPE 9 produces a specific next-day HRV/RHR signature. A set the lifter reported as 9 that produces the signature of an 8 reveals the drift.

A system that compares these signals across weeks can correct for drift. If a lifter’s reported RPE 9 now produces the biomarker signature of an RPE 8 from three months ago, the system either flags the drift and asks the lifter to recalibrate, or applies a correction to the load prescription so that the effective effort stays constant.

This matters most during long blocks. An auto-regulated program running on uncorrected RPE produces a familiar arc: the lifter feels fine weeks one through three, mildly tired weeks four through six, noticeably beat up by week eight, and injured or sick by week ten. The RPE was drifting the whole time. A calibrated system catches it in week four.

More on this in RPE calibration.

Load Progression: Why Bench, Squat, and Deadlift Follow Different Curves

If you plot weekly top-set load over a long training history, the curves for different compound lifts don’t look the same. Squat tends to progress faster than bench. Deadlift is usually the slowest, with long plateaus punctuated by abrupt jumps. The overhead press looks different again, with tighter variance but lower absolute slope.

The generic “add 2.5 lb a week to your bench” rule treats these differences as noise. They aren’t. They are structural, driven by the geometry of the lift, the muscle groups involved, and the ratio of technical to metabolic limitation.

Squat involves the largest absolute muscle mass and trains a compound movement where fine technique matters less than raw force production. Progress is relatively fast and linear for the first year or two of serious training, then transitions to block-periodized progression where weekly load fluctuates within a planned envelope. A reasonable intermediate lifter might add 10 to 20 lb on their squat top set over a 12-week block.

Bench press involves a smaller absolute mass, a shorter range of motion, and more technical sensitivity. Progress tends to be slower and more variable. The same 12-week block might yield 5 to 10 lb on the top set. Bench also plateaus earlier and benefits more from accessory work — a lifter who has plateaued on bench often sees progress resume when they add triceps and back volume rather than pushing the bench itself harder.

Deadlift is the high-impact lift from a recovery standpoint. Repeated hard deadlifts can accumulate fatigue faster than their volume implies. Progression tends to be jumpy — periods of stable weekly loads followed by sudden 10 to 20 lb jumps as a block peaks. An adaptive system that treats deadlift like bench will either under-load or overtrain it.

Overhead press is the slowest progressor on average, but also the most consistent. Small weekly jumps (1 to 2.5 lb for intermediates) stack over long time horizons. The overhead press rewards patience.

The implication for adaptive training intelligence is that load progression should be estimated per-lift and per-user. The system watches your actual progression history — not the block plan, the actual recorded top sets — and learns each lift’s expected weekly rate. When a scheduled jump outpaces the learned rate by more than a credible interval, the system flags it. When progression stalls for longer than expected, it suggests a deload or a block transition.

For deeper treatment, see personal load progression rates.

ACWR, Monotony, and Strain: Three Signals That Together Predict Overtraining

Acute:Chronic Workload Ratio (ACWR) is the overtraining signal that consumer platforms love to surface. It’s a simple idea: divide this week’s training load (acute, typically a 7-day rolling sum or exponentially weighted mean) by the rolling 28-day average (chronic). A ratio near 1.0 means you’re training at your recent average. A ratio above 1.5 means you’re doing much more than your recent average. A ratio below 0.8 means you’re undertraining relative to your chronic baseline.

The original Gabbett research suggested that ACWR above approximately 1.5 increases injury risk substantially in team-sport athletes. The number has since been adopted — and arguably over-adopted — across the training tech space.

It is important to be precise about what ACWR is and isn’t in a modern system.

ACWR is a context signal, not a prescription driver. A ratio of 1.4 doesn’t tell you to reduce tomorrow’s session by 30%. It tells you that tomorrow’s session lands on top of a body that has recently been doing less, and that the downstream load is trending toward a zone where injury risk climbs. The prescription comes from MAV. ACWR is the alarm that flashes when the prescription and the recent history are drifting apart.

ACWR is sensitive to the math. A rolling 7-day sum has different dynamics than an exponentially weighted moving average of the same window. Week boundaries create artifacts: a Sunday-heavy session lands in the prior acute window and spikes Monday’s ACWR in ways that don’t reflect how the body experienced the load. EWMA-based ACWR smooths this out and is usually the better default.

ACWR under-weighs the shape of the training week. Two weeks with identical ACWR can have very different physiological meaning. A week where every session was medium-hard has the same acute load as a week with one hard day, two easy days, and one off day — but the second week is more recoverable. The signal that captures this is monotony: the ratio of mean daily load to its standard deviation over the week. Monotony close to 1 means you had high-variance days with clear hard/easy contrast. Monotony above 2 means you were doing roughly the same thing every day.

Strain combines the two. Strain is weekly load multiplied by monotony. In the Foster-style model, high strain — the combination of high weekly volume and high monotony — is a stronger predictor of overtraining than either alone. A week of high volume with high variance is usually fine. A week of moderate volume with no variance is often where overtraining shows up.

Putting the three together:

• MAV is the prescription. How much you should do per muscle this week.
• ACWR is the trend alarm. Whether this week’s total load is escalating too fast relative to recent weeks.
• Monotony and strain are the structure alarms. Whether the week, as distributed across days, has enough variance to be recoverable.

When all three are in the green band, the plan is working. When MAV says push but ACWR is above 1.4 and monotony is above 1.8, the plan is probably producing a week the body won’t recover from. When MAV says pull back but ACWR is 0.7, the system is catching an unnecessary deload on already-undertrained muscle groups.

A dashboard that shows only ACWR — or only readiness, or only recovery — is showing a fragment of the picture. The three signals together are what lets a training system be adaptive rather than generic.

More on the overtraining signal itself in monotony and strain.

Detraining and Re-Entry: What Happens After a Break

Training breaks happen. Travel, illness, injury, life disruption, a planned off-season. The adaptive training question is not “should I have taken the break” — you already did — but “what should the first session back look like, and how fast can I ramp?”

The research baseline here is Mujika and Padilla’s detraining studies (short-term and long-term reviews), plus subsequent work that has refined the picture. The summary:

• 0 to 2 weeks off. Strength is remarkably durable. Short breaks often produce no measurable strength loss and sometimes a small rebound due to recovered fatigue. Aerobic fitness declines faster than strength, with meaningful drops in VO2 max after about 10 days.
• 2 to 4 weeks off. Small strength declines begin, especially in trained lifters. Fast-twitch fibers atrophy faster than slow-twitch. Neural adaptations are retained better than muscular ones. Most lifters return to baseline within 2 to 3 weeks of re-training.
• 4 to 8 weeks off. Strength losses become noticeable, typically 5 to 15% depending on the muscle group, training age, and age. Return-to-baseline takes longer — often as long as the layoff.
• 8 to 12 weeks off. Significant losses, and re-entry needs conscious caution to avoid injury. The first week back at prior loads is the highest-risk window.
• 12+ weeks off. Re-entry should treat the lifter as a returning trainee, not an advanced one. Starting loads of 60 to 70% of prior top sets for the first two weeks is a reasonable default.

An adaptive system uses the layoff length to interpolate between your pre-break MAV and a literature-derived floor. The longer the layoff, the closer the prescription moves to the floor. The first week back is deliberately conservative — usually 60 to 75% of pre-break loads with reduced volume — and the system monitors the recovery markers closely. If HRV, RHR, and sleep tolerate the re-entry well, the prescription ramps up quickly (week two at 80%, week three at 90%, week four back to baseline for short breaks). If the recovery markers show stress, the ramp slows.

The key principle: detraining is not only a strength drop. It’s also a tendon and connective tissue detraining, and connective tissue re-adapts slower than muscle. A lifter who comes back at pre-break loads on week one often feels fine on the lift and gets a tendon injury in week three. The conservative ramp exists for tendons as much as for muscle.

More in detraining: how fast you actually lose strength. For broader context on how breaks interact with underfueling, see energy availability.

Putting It Together: A Week in the Life of Adaptive Training Data

Concrete example. A 38-year-old intermediate lifter, five-day split (push / pull / legs / upper / lower), nine months of consistent logging, no recent injuries, using a wearable and a lift tracker that both feed into a single adaptive system.

Sunday night. Last night’s HRV was back to the personal p50 after a hard Saturday legs session. Sleep 7.5 hours. Resting heart rate 54 — at p50. The system has Sunday as a rest day. Confidence high.

Monday (push day). Morning readiness at p60. Chest and triceps are 96% recovered from last week’s Thursday push session. MAV lower bound on chest is 14 sets/week; current week pace is 0 (this is the first session). Scheduled session: bench 5x5, incline DB 3x10, overhead press 4x6, cable flyes 3x12, tricep extensions 3x12. Total: 10 chest sets, 4 shoulder sets, 6 tricep sets. The system confirms the session as-planned because every muscle is well within its credible MAV band and recovery is clean.

Tuesday (pull day). Mid-back and biceps fully recovered. Lats were at 88% recovered from a Friday accessory session — close to a threshold. MAV lower bound on lats is 12 sets/week. Scheduled session includes 8 lat-heavy sets. The system flags: “lat recovery marginal, consider reducing top-set RPE by 0.5 or shifting two sets to Thursday.” The lifter takes the suggestion and does 6 hard lat sets + 2 back-off.

Wednesday (legs day). Squats and deadlifts. Morning readiness at p45 — below personal normal. HRV is at p30. The system looks at the context: hard push Monday, moderately hard pull Tuesday, quads and hamstrings are 72% recovered from last Saturday. Scheduled: squat 5x5 at RPE 8, RDL 4x6 at RPE 8, leg press 3x10, hamstring curls 3x12. ACWR for lower-body load is at 1.35 — still in the green band. The system suggests “readiness signal mildly low; consider capping top-set RPE at 7 and dropping leg press one set.” The lifter takes it.

Thursday (upper accessory). Recovery markers are back to personal normal. ACWR dropped because Wednesday was lighter than planned. Chest is at 78% recovered, shoulders at 85%. Scheduled upper work lands comfortably within MAV bands. The session runs as-planned.

Friday (lower accessory / conditioning). Quads are at 65% recovered (48 hours post-squats, half-life of roughly 60 hours for this lifter). Hamstrings at 70%. The planned session is conditioning, not heavy lower — sled work, core, a short row piece. No heavy quad or hamstring load. The system is happy.

Saturday (off). Pure rest day. The system uses the day to update posteriors. Chest MAV posterior barely moves — one normal week. Lat MAV posterior widens slightly because of the within-week schedule shift, then tightens again as the logged sets confirm the revised plan was fine. Squat progression slope updates with Wednesday’s RPE-7 data point, which the model weights less heavily than an RPE-8 reference session.

Sunday. Light active recovery — a walk, some mobility. The weekly summary:

• Chest: 13 hard sets (inside MAV band of 11–15 lower bound to upper bound).
• Back: 14 hard sets (inside band 12–16).
• Quads: 9 hard sets, one RPE step below plan (inside band 8–12).
• Hamstrings: 9 hard sets (inside band 8–11).
• Shoulders: 10 hard sets (inside band 8–12).
• ACWR for total load: 1.12 (green).
• Monotony: 1.4 (green).
• Strain: moderate, no flag.
• Readiness trend: stable.

The week wasn’t the week the lifter planned on paper. The plan said 10 chest sets, not 13. It said 5x5 hard squats, not 5x3 at RPE 7. It said 8 lat sets on Tuesday, not 6 on Tuesday plus 2 on Thursday. The week was, however, the week the lifter’s body was ready for. The plan and the system adapted to each other, rather than the lifter grinding through a plan the body wasn’t recovered for.

That is what adaptive training intelligence looks like in practice. It is not a replacement for the program. It is the program talking back.

---

Putting It Together

The gap between “I have a wearable” and “I know exactly what to train today” is not a UI problem. It is a modeling problem. Consumer platforms collapse the model into a single readiness number because that number is easier to render than the structure underneath. The structure is what actually drives training decisions: per-muscle volume ceilings, per-muscle recovery half-lives, calibrated RPE, per-lift progression rates, and overtraining signals used as context rather than prescription.

The components are:

• Personalized per-muscle MAV with credible intervals, updated weekly
• Recovery half-lives that differ by muscle group and by lifter
• An RPE calibration layer that corrects for subjective-scale drift
• Per-lift load progression rates that respect that bench, squat, and deadlift are different curves
• ACWR (preferably EWMA-based) as a trend alarm, not a prescription driver
• Monotony and strain as the structure alarms inside the week
• A detraining/re-entry model that uses layoff length to scale the return-to-training ramp

None of these are speculative. The science behind each is well-established, even where the consumer-product industry has flattened it into a single wellness score. What is hard is wiring them together into one system that reads your logs, your wearable data, and your actual training history, and that talks back when the plan and the body are drifting apart.

The honest framing: a system like this makes you a better programmer of your own training. It will not replace a coach. It will not catch technique errors or notice that you look flat when you say you feel fine. It will catch, reliably, the week your prescribed volume is outrunning your chronic baseline; the lift whose progression has stalled for longer than expected; the muscle group whose recovery half-life has quietly lengthened under life stress; the deload you needed three weeks ago.

If your current training software shows you a readiness score and a workout for the day, you’re using a recommendation engine. If it shows you per-muscle MAV bands, per-lift progression trends, and the full ACWR/monotony/strain stack with confidence intervals on all of it, you’re using a model. The difference matters most on the weeks where the plan and the body disagree.

Omnio builds the second kind. The adaptive training feature writeup is at /features/adaptive-training, and the deep-dives on each component are linked throughout this guide.

---

More in this Series

This is the pillar post for a cluster on adaptive training intelligence. Companion spokes are scheduled over the next three months:

• Maximum Adaptive Volume: Why Per-Muscle Training Ceilings Beat Generic Volume Rules
• Detraining: How Fast You Actually Lose Strength (and What the Research Shows)
• Per-Muscle Recovery Half-Lives: Why Legs ≠ Shoulders ≠ Chest
• RPE Calibration: Why Your 9 Is Someone Else’s 7 (and Why It Matters)
• Your Bench Doesn’t Progress Like Your Squat: Per-Movement Load Progression
• EWMA vs Rolling ACWR: Why the Week-Boundary Math Lies
• Monotony and Strain: The Overtraining Signal Wearables Miss
• Your Training Split vs Your Actual Training Split: Schedule Detection from Behavior

And existing posts in the same territory:

• What Is ACWR and Why It Matters for Training
• How Wearables Measure Stress and Strain
• What Is HRV and How Do Wearables Measure It
• Predicting Health Dips Before They Happen

Companion pillars on other clusters:

• Composite Scores and Confidence
• Nutrition Intelligence
• Energy Availability and RED-S