What is ACWR in plain terms?

Acute:Chronic Workload Ratio compares recent training load (the acute window, typically 7 days) to longer-term training load (the chronic window, typically 28 days). A ratio near 1.0 means you're training at your recent average. A ratio above 1.5 typically flags elevated injury risk in team-sport athletes based on Gabbett's original research. A ratio below 0.8 flags undertraining relative to your chronic baseline.

Why is rolling ACWR problematic?

Three structural problems. First, rolling sums treat every day in the window equally — a session from 6 days ago counts the same as a session yesterday, which doesn't match how the body experiences load. Second, week boundaries create artifacts: a session on Sunday lands in the prior week's acute sum and a session on Monday in the new week's, even though the body experienced them consecutively. Third, missed sessions create false troughs that trigger undertraining flags when the lifter was simply taking a rest day.

What does EWMA do differently?

Exponentially Weighted Moving Average weights recent sessions more heavily than older ones, with the weight declining smoothly. There's no hard window boundary. A session from yesterday counts more than one from three days ago, which counts more than one from a week ago. The dropoff rate is tunable via a decay parameter, and the resulting ACWR-equivalent is smoother and physiologically more sensible.

Is an EWMA ACWR above 1.5 still a red flag?

The threshold needs recalibration because the math is different. In published work, the EWMA equivalents to the rolling 1.5 threshold tend to sit around 1.4 depending on the decay rate. The more important point: ACWR thresholds are population-level signals, not individual prescriptions. A 1.5 on either math is a flag to investigate context, not a guaranteed injury.

Should the lifter care which version their platform uses?

Yes, at least for interpretation. A platform using rolling ACWR will produce more volatile numbers that spike across week boundaries and dip during rest days. A platform using EWMA ACWR will produce smoother trend numbers that match the physiological picture better. Don't panic over a single spike on a rolling implementation; do pay attention to sustained elevation on either.

EWMA vs Rolling ACWR: Why the Week-Boundary Math Lies

Your platform says your ACWR is 1.6. You had a hard session Sunday, which rolled into the new acute week on Monday. Now the dashboard is red, the recommendation is “reduce volume,” and you haven’t changed your behavior between Saturday and today.

This is a mathematical artifact, not a physiological signal. And it’s avoidable with a different calculation.

This post is about why rolling ACWR creates artifacts, what Exponentially Weighted Moving Average (EWMA) ACWR does differently, and when to trust the signal in either form. It’s part of the adaptive training intelligence cluster.

What ACWR Tries to Measure

ACWR — Acute:Chronic Workload Ratio — became widely adopted in athletic training after a series of papers by Tim Gabbett and collaborators (starting around 2014) showed that rapid increases in training load predicted injury risk in team-sport athletes. The core finding: when acute load (the last 7 days) rises significantly above chronic load (the 28-day rolling average), injury rates climbed.

The ratio formalizes this. Acute workload divided by chronic workload gives a single number. A ratio of 1.0 means you’re doing exactly your recent average. A ratio of 1.4 means you’re doing 40 percent more. Gabbett’s threshold of “approximately 1.5” for elevated risk has been widely cited and widely criticized — a decade of follow-up research has added nuance but hasn’t overturned the basic premise that sharp load increases matter.

ACWR is useful. It catches sudden ramp-ups that recovery might not have caught up with. It flags the weeks where a plan has drifted from recent history. For context, it’s a better signal than bare weekly volume.

Where ACWR gets into trouble is in the implementation details.

How Rolling ACWR Actually Computes

The standard “rolling” ACWR uses two windowed sums:

Acute: Sum of training load over the last 7 days.
Chronic: Sum of training load over the last 28 days, divided by 4 to produce a 7-day equivalent.

The ratio is acute divided by chronic. A simple calculation, which is part of why it’s popular. You can compute it in a spreadsheet.

The simplicity is also where the problems live. Three structural issues.

Problem one: Equal weighting within the window. A session from 6 days ago counts the same as a session yesterday in the acute window. But the body doesn’t experience them the same. Yesterday’s session is still producing active inflammation, glycogen depletion, and neural fatigue. A session from six days ago is mostly recovered from. Treating them as equivalent is a modeling shortcut that doesn’t match physiology.

Problem two: Sharp window boundaries. A session seven days ago is “in the window.” A session seven days and one minute ago is “out of the window.” The body doesn’t respect the boundary. The drop from “fully counted” to “not counted” at the window edge creates mathematical step changes that don’t correspond to physiological reality.

This produces a specific artifact: the “Sunday session” problem. A hard Sunday session is in the acute window on Monday. On the next Monday, that same session is “out” of the acute window entirely. The ratio drops sharply not because the lifter has done anything different, but because the calendar rolled over.

Problem three: Missed sessions and rest days. If a lifter takes two rest days in a row during a hard block, the acute sum drops significantly. The ratio drops. The platform might flag undertraining or suggest ramping volume back up. But the lifter didn’t undertrain — they took appropriate rest. Rolling math doesn’t know the difference between a strategically easier week and an accidental ramp-down.

These aren’t edge cases. They happen every week in every consistently-logged training program. The effect is that rolling ACWR is noisy, jumps around for mathematical rather than physiological reasons, and produces flags that don’t track the actual injury-risk trajectory.

What EWMA Does

Exponentially Weighted Moving Average replaces the window sum with a weighted sum that decays over time. The idea:

The most recent session counts the most.
Each previous session counts less, with weights declining exponentially.
There’s no hard boundary — old sessions eventually contribute a negligible amount but the dropoff is smooth.

The decay rate is controlled by a parameter called alpha (between 0 and 1). A higher alpha makes the EWMA respond faster to recent changes; a lower alpha makes it smoother. For training load, common defaults correspond to effective windows of 5 to 7 days for acute and 21 to 28 days for chronic.

Mathematically, an EWMA updates each day by a fixed fraction of today’s value plus a fixed fraction of yesterday’s EWMA. No windows to manage, no boundaries to worry about. Every session has a smoothly declining influence on the current estimate.

The practical effects compared to rolling:

Fading influence, no cutoffs. A session from eight days ago still counts a little — it just counts less than a session from three days ago. The step change at the window boundary disappears.

Smoother day-to-day trend. Small day-to-day load variations produce small trend changes. A single missed session or a single hard extra session is absorbed into the trend rather than spiking it.

Sunday-session problem fixed. A hard Sunday session has its peak influence on Monday, fades a bit by Tuesday, more by Wednesday, and becomes background by the following week. There’s no sudden drop when the calendar rolls over.

Weighting that matches physiology. Recent sessions are still producing acute effects; older sessions are mostly recovered. The weighting structure of EWMA — heavier on recent, lighter on older — is more physiologically sensible than uniform weighting.

What Published Research Says

The EWMA-vs-rolling debate is well-covered in sports science literature. Murray and colleagues (2017) compared the two methods in Australian football players and found EWMA-based ACWR had a stronger relationship with injury risk than rolling ACWR. Subsequent work has supported the finding across different sports, though the literature isn’t unanimous — some studies find comparable performance between the methods and debate which is the right default.

Where the literature does converge: the choice of decay parameter matters. Too fast a decay (alpha too high) produces noisy EWMA. Too slow (alpha too low) washes out the acute signal. Reasonable defaults come from matching the effective window to the population the model is calibrated on — a 7-day acute window corresponds to an alpha around 0.25 to 0.3; a 28-day chronic window corresponds to around 0.07.

Threshold values for EWMA ACWR are slightly different from rolling. The commonly cited 1.5 rolling threshold for elevated injury risk maps roughly to 1.4 on an EWMA with standard decay. The mapping isn’t exact and shouldn’t be taken as a precise conversion — it’s more that both versions produce flags at approximately similar effect sizes when calibrated to the same population.

The Broader Point: ACWR Is Context, Not Prescription

Regardless of whether a platform uses rolling or EWMA, the larger interpretation issue deserves emphasis.

ACWR is a trend alarm. It flags that the acute load has climbed above the chronic baseline in a way that historically correlates with injury risk in large populations. What it doesn’t do:

Tell you how much to reduce. A 1.5 ratio doesn’t prescribe a 50 percent volume cut. It flags elevated risk. The prescription still comes from elsewhere — in the adaptive training stack, from per-muscle MAV.
Account for which muscles the load hit. A 1.5 ratio driven by two hard leg sessions and a 1.5 driven by two hard shoulder sessions have very different injury-risk implications. ACWR is a global signal.
Distinguish intentional ramp from accidental ramp. Peaking blocks intentionally run acute above chronic. A coach programming a peak might deliberately hit 1.4 to 1.5 in the weeks leading to a meet. The platform’s red alarm doesn’t know the intent.
Replace readiness signals. ACWR is a load-based signal. It doesn’t know you slept four hours or skipped meals. The composite readiness signals handle that; ACWR handles the load side.

The correct interpretation of an elevated ACWR: “Your recent training load is climbing faster than your chronic baseline. Consider whether this is intentional, whether recovery markers are keeping up, and whether the next week’s plan still makes sense.” The correct interpretation of a depressed ACWR: “Your recent load is below your chronic baseline. Is this planned rest, a context shift, or a drift you didn’t mean to have?”

Neither interpretation is a prescription. Both are flags to look at context.

Monotony and Strain as Companions to ACWR

ACWR on its own, whether rolling or EWMA, under-weighs the shape of the training week. Two weeks with identical ACWR can have very different physiological meaning.

Week A: five moderate-hard days, two off days. ACWR 1.15.
Week B: seven consistent medium days with no variation. ACWR 1.15.

Week A is more recoverable. Week B is more likely to produce overreach, even though the ratio is identical.

This is what monotony and strain capture.

Monotony = mean daily load / standard deviation of daily load, computed over a window (typically 7 days).

Monotony close to 1.0 means high variance — clear hard/easy days.
Monotony above 2.0 means low variance — every day looks similar.

Strain = weekly load × monotony.

Low strain = either low volume or high variance or both. Recoverable.
High strain = high volume with low variance. Overtraining risk rises.

Foster’s original work on monotony and strain predated the ACWR framing and showed that these signals catch overtraining that ACWR alone misses. Modern adaptive systems use all three — ACWR, monotony, strain — as the overtraining context signal, with none of them acting as a prescription driver on their own. Monotony and strain get their own spoke in this series.

What to Do If Your Platform Uses Rolling ACWR

Most consumer platforms still use rolling ACWR. If yours does, practical implications:

Watch the trend, not the snapshot. A single day’s rolling ACWR of 1.5 is often an artifact. A sustained trend of 1.4+ over 5 to 7 days is a real signal.

Mentally subtract the week boundary. If you know a hard Sunday session just rolled into the acute window, the Monday spike is probably the artifact. Look at Tuesday or Wednesday to see what the ratio settles to.

Ignore dips from rest days. A scheduled rest day or two will drop the acute sum sharply on a rolling implementation. That’s not undertraining; it’s rest.

Look at the chronic trend. Chronic load (the 28-day sum) is much smoother and more physiologically meaningful than acute. A rising chronic trend is a training adaptation signal; a falling chronic trend during a planned block is something to investigate.

Ask whether EWMA is available. Some platforms offer both and let you pick the display. EWMA is usually the better default for interpretation.

What a Good Platform Shows

An adaptive training platform that handles ACWR well should:

Use EWMA as the default, with transparent decay parameters.
Show both acute and chronic trends independently, not just the ratio. You want to see whether the acute is rising, the chronic is falling, or both.
Surface the ratio as a trend band, not a single number. The band shows typical weekly variance so the lifter knows which changes are meaningful.
Combine ACWR with monotony and strain for the overtraining context signal. ACWR alone is incomplete.
Treat ACWR as context, not prescription. Prescription should come from MAV or coach-supplied plans. ACWR flags when the prescription is drifting from recent history.
Acknowledge intentional ramps. Peaking blocks legitimately push acute above chronic. Flagging “risk” in these cases without accounting for intent is false alarms.

A Concrete Comparison

Consider a lifter with a 7-day acute and 28-day chronic window. They train Monday, Wednesday, and Friday at moderate load, with variable intensity:

Mon: 6 TRIMP
Wed: 5 TRIMP
Fri: 7 TRIMP
Total weekly load: 18 TRIMP
Daily average: ~2.6

They’ve been in this pattern for four weeks. Chronic load is steady around 18 TRIMP/week.

Saturday. They add an extra session, 6 TRIMP. Weekly acute now 24. Rolling ACWR = 24/18 = 1.33.

Sunday. Rolling ACWR is still 1.33 because nothing has changed yet. The Saturday session is in the window.

Monday. The new week’s Monday session is 6 TRIMP. Acute window now covers Monday (6) through last Tuesday (0). The prior Saturday session is still in the window. Acute = 24. ACWR ≈ 1.33. But this is the first Monday where the acute is dominated by recent sessions, not the extra Saturday.

Next Saturday. Last Saturday’s extra session rolls OUT of the 7-day window. Acute drops from 24 to 18. ACWR drops from 1.33 to 1.00.

The drop from 1.33 to 1.00 at the window roll-off is a pure mathematical artifact. The lifter has been consistent for 6 days. Nothing has changed. A rolling implementation shows a sharp “dashboard green” change that isn’t physiologically meaningful.

EWMA version. The same events produce a smoother curve. The extra Saturday session peaks the EWMA mid-week, fades gradually over the following 7 to 10 days, and settles back toward baseline without a step change. The ratio is informative without being jumpy.

Both calculations are detecting the same underlying event — one extra session on a Saturday. EWMA frames it in a way that matches how the body actually processes it.

In Summary

Rolling ACWR creates window-boundary artifacts, spikes around weekend sessions, and false troughs from rest days.
EWMA ACWR smooths these artifacts, weights recent sessions more heavily, and produces a trend that better matches physiology.
Thresholds differ slightly between the two (roughly 1.5 rolling ≈ 1.4 EWMA), but both are population-level flags, not individual prescriptions.
ACWR belongs with monotony and strain as the overtraining context signal, not as a standalone prescription.
In the adaptive training stack, prescription comes from MAV; ACWR is the alarm that flags when the MAV prescription and the recent history are drifting apart.

Platforms that surface ACWR as a single number, without context and without the monotony/strain companions, are showing a partial picture. The number is useful. The interpretation depends on the rest.

Omnio’s implementation: /features/adaptive-training.