How Long Does a Muscle Take to Recover? What Sterka Assumes, and Why

There is a number a lot of training software will show you in the morning: one figure, one bar, one word. Seventy-two percent recovered. Ready. Amber.

Recovered from what, exactly? You benched on Monday and squatted on Saturday. Your chest and your calves are not in the same state, and no single figure can hold both without averaging one of them into a lie. The average is the problem. It is the exact information a coach would use to pick today’s session, and a single score throws it away before you ever see it.

Sterka does not compute a whole-body readiness score. It keeps one number per muscle, each on its own clock.

How long does a muscle take to recover?

There is no measured answer to that question, and anything that gives you one is guessing with a confident face. What there is, in Sterka, is a stated assumption: each muscle clears half its accumulated fatigue in 36 to 72 hours, depending which muscle. Those are the engine’s own calibrated figures, not a reading taken off your body, and the section at the end of this post says exactly how far that claim reaches.

That “half” is the important part. Recovery is not a switch that flips at some number of hours. Fatigue drains off continuously, and the honest way to describe it is a half-life: the time it takes for half of what a session deposited to fade. Sterka’s default is 48 hours, and muscle groups sit in tiers around it.

The spread runs the way a coach would expect it to. Large, eccentrically loaded groups take longer to clear a hard session. Small groups with shorter strength curves clear faster, which is what makes twice-weekly lateral raises reasonable and twice-weekly heavy deadlifts a different question.

The lower back sits alone at the top on purpose. For most lifters, the erector spinae is what actually governs how often heavy pulls can happen, not the prime movers doing the lifting. It is the slowest-recovering, most fatigue-accumulating system in the program, so it gets its own tier above the big movers it works alongside rather than being lumped in with them.

The arithmetic

Every hard set you log deposits fatigue on the muscles it worked, and every deposit decays from its own timestamp:

decayed = dose × 0.5 ^ (hours since the set ÷ that muscle's half-life)

A muscle’s recovery percentage is 100 minus the sum of everything still standing, floored at zero. One hundred is fresh. Zero is maximally cooked.

Here is a real session. You finish five hard chest sets at RIR 2, two reps short of failure. The base cost of a primary hard set is 10 fatigue units, RIR 2 is the neutral effort multiplier, and chest is the primary mover on all five, so the deposit is 5 × 10 × 1.0 = 50. Chest reads 50 percent recovered when you rack the last set.

Chest runs a 54-hour half-life, so 54 hours later exactly half of that 50 has cleared and chest reads 75 percent. Four days out it reads 85.

Now put the identical deposit on every tier at once and watch them separate.

Three days after that identical session, a 36-hour muscle is near 90 percent and the lower back has only just reached 75. Averaged into one score, those two would read as the same lifter having the same day.

One thing the curves make obvious and every readiness display hides: 100 is not the target. A muscle does read a clean 100 eventually, but depending on the tier that is ten days to three weeks out, and training on that schedule is not a plan. Seventy-five is where a bucket runs unrestricted, and it gets there in days.

What a set actually costs

The deposit is not flat, because not every set costs the muscle the same thing.

Which muscle did the work. A primary mover takes the full contribution, a secondary muscle half of it, so a heavy bench charges chest fully and triceps and front delts partially. A dedicated curl, pushdown or lateral raise, where the isolation has a single primary, charges that one muscle at 1.5 times the base: a near-failure finisher deposits more local fatigue than an incidental bystander does.

How close to failure. The effort you log scales the whole deposit: 1.4 at failure, 1.2 at RIR 1, 1.0 at RIR 2, 0.85 at RIR 3, 0.7 at RIR 4 and lighter.

What kind of lift it was. Effort alone cannot separate a 90 percent squat single at RIR 1 from a near-failure dumbbell curl at RIR 1, and those are not the same event for your spine. Free-weight compounds that stack load on the spine get a bracket on top: 1.15 for a set in the 85 to 90 percent intensity range, 1.30 above it. The intensity is inferred from reps and reps-in-reserve, not from the weight on the bar. Machine and spine-sparing versions of the same movements are deliberately left alone, since taxing the safer option would be backwards. This bracket also refuses to be talked down: it takes the harder of what was prescribed and what you logged, so an honest “I had three left” on a heavy squat lowers the effort multiplier above but not your spinal exposure here.

That is why “how many days since I trained chest” is not the question. Twelve hard sets to failure and two easy sets both happened yesterday. Only one of them should still be holding chest down today.

What the number then does to your session

A recovery percentage nobody acts on is decoration. Two different things read these numbers, and they answer different questions.

Every session: the cooked muscle throttles itself

The everyday one is the quieter of the two, and it is the genuinely per-muscle half. Each muscle’s own number sets its own ceiling for the week: below 50 percent it is held at its adaptive sweet spot instead of being pushed toward its maximum, and below 30 it drops to a maintenance dose until it comes back. Today’s session is then trimmed to fit under that ceiling. It happens per muscle, on every session, through every door: the app, the API, a coach’s proposal. A single cooked chest gets throttled while the rest of you trains normally, which is the thing one whole-body score structurally cannot do.

On Get Stronger both mechanisms run at once, on different lifts. The heavy anchor is eased by load: its top set comes down, its sets do not. The cooked accessory around it loses volume exactly as it would on Build Muscle. A hypertrophy frame would cut the cooked heavy lift’s sets; the strength frame keeps the heavy work, dials the bar down, and takes the volume off the accessory instead.

When the engine picks the day: it also gets pointed

The louder one runs whenever the engine is the one choosing. That includes the session already waiting for you in Today, not only a day you regenerate or train anyway on a rest call. It scores each of push, pull and legs from its own primary muscles, reads a band off the best of the three, and sends you there. At 75 or above, the session runs as written. From 50 to 74, working sets stay at least two reps from failure. Below 50, meaning nothing is fresh anywhere, it halves the working sets and eases effort.

This is where the per-muscle model pays for itself most visibly: a cooked chest does not get you a half-volume chest day, it gets you legs, at full volume, because legs are fine. The half-volume session is reserved for the day when there is no fresh bucket left to send you to.

Name the focus yourself and it trains what you asked for, on any path, as does a session composed through the API or proposed by the coach. The caps still travel there, but they are sized off your freshest bucket while you train the one you named. That is the trade for making the call yourself, and the per-muscle ceiling above is what still holds the cooked muscle down when you do.

How much the effort eases depends on your goal. On Build Muscle a light day holds working sets at three reps in reserve, the protective version. On Get Stronger the ceiling is two, deliberately closer to failure: a strength lifter’s hard day should not be converted into a pump session because their bars are low. What that ceiling takes depends on where the block already sits. The strength wave cues working sets differently across a block, so on most weeks a ceiling of two eases the top set alone, and on a week already cued closer than that it pulls the working sets back too.

A few more places the number lands:

• A light day latches until the muscles clear 60 percent, not the 50 that triggered it. Without the gap, an eased session lets recovery rebound just past the edge, the next session runs full, that re-fatigues you, and the block settles into alternating half-sessions with full ones while going nowhere.
• On Get Stronger, a focus muscle sitting between 30 and 70 percent is “accumulating but not wrecked”, and the engine floors your working sets one notch further from failure so the volume around the main lift stops digging. The heavy single is governed separately, and more protectively: the session’s anchor lift is graded on its least recovered primary muscle. Between 50 and 70 the top set is capped at 85 percent and brought back to your accumulation effort rather than left at the harder prescribed number, and the opportunistic near-max rep-outs are withheld. Below 50 the heavy block is gated off entirely.
• A low computed lower-back score does not, on its own, cancel leg day. It sets a ceiling instead: when a session programs a free-weight spinal compound and the lower back reads below 50, the top set is capped at 87 percent of your estimated max. It is a cap, not a haircut, so it can only move a prescription that reaches 87 percent in the first place, which is strength-band territory; a hypertrophy working set already sits below it. It also stands down during a deload, where the whole session is already light.
• When every primary muscle across push, pull and legs is below 30 percent, the engine calls a rest day rather than inventing a session for you to fail.

This is the machinery under why your workout app never adjusts. When the fatigue is systemic rather than local, the scheduled deload is the planned answer that does not wait for your bars to go red.

Where these numbers come from, and where they do not

An honest caveat, since this is the point at which fitness software usually starts implying more than it knows.

The half-life tiers are a coaching calibration, not a physiological constant someone measured in a lab and Sterka copied out. The ordering follows the recovery-ramp asymmetry between large eccentric groups and small ones, and the values are deliberately conservative: faster than the old “a muscle needs a full week” folklore, slower than the fastest recovery claims in circulation. They are written down where you can read them, which is the most any fatigue model should claim.

The scale is also set deliberately hot. Five sets at RIR 2 landing a muscle exactly on the depleted line is a fast burn, and it is fair to argue it should be slower. The half-life and the per-set base are the two knobs that set it.

What this is not: a health measurement. It does not read your heart rate, your sleep or your nervous system. It is bookkeeping on training you actually did, which is a narrower claim than a readiness score makes, and a checkable one.

Treat the tiers as the engine’s assumption about you, not a finding about you.

Per-muscle recovery gets sold as a way to hit a frequency number, and it is worth saying what it does not promise. Current ACSM guidance puts each major muscle at two or more sessions a week. But for size specifically, the largest current dose-response meta-regression finds that once weekly volume is equated, spreading it across more days adds little on its own: for hypertrophy, frequency is a delivery mechanism rather than a driver. Strength is the exception, where the same work finds a real, diminishing effect of training a lift more often.

So the recovery model is not there to defend a twice-a-week target, and it does not promise you one. It answers a smaller question on the day: how much dose can this session carry, and which muscles should carry it. How often a muscle actually comes around is set by how many days you train and how your split divides them. Three sessions a week can be well-judged on every one of them and still reach each category once.

Nothing here is stored

One last property, because it changes what the number means. Fatigue is never saved as a running total. It is recomputed on every read by summing your surviving training log, each set decayed from its own timestamp.

That sounds like an implementation detail and is not. A stored total is a claim about your history; a derived one is a reading of it. Delete a set you logged by mistake, retract a session, fix an RIR you guessed wrong, and the next read is correct, because there is no accumulated total to unwind and no path that has to remember to unwind it. Nothing accumulates phantom fatigue you cannot see and cannot clear.

One thing is stored, and it is yours: if you tell Sterka a muscle is still wrecked, that override stands over the computed number until you next train that muscle or the day ends. You are a source of evidence about your own body, and the model treats you as one.

A muscle you have never trained reads 100 percent, and the engine records that it is an absence of measurement rather than evidence of freshness. The two are not the same, and treating them as the same is how a plan talks itself into a volume jump it has no reason to believe in.

A recovery model earns its place by changing what you are told to do today. One number cannot do that, because by the time it is one number the information that would have changed the session is already gone.

References

1. American College of Sports Medicine position stand. Resistance training prescription for muscle function, hypertrophy, and physical performance in healthy adults: An overview of reviews. Medicine & Science in Sports & Exercise, 2026.
2. Pelland JC, Remmert JF, et al. The resistance training dose-response: meta-regressions exploring the effects of weekly volume and frequency on muscle hypertrophy and strength. SportRxiv preprint, 2024; subsequently published in Sports Medicine, 2025.