Estimated 1RM Keeps Changing Daily: Is It Noise or Real Readiness Drift?

You log a 2-rep velocity check at the same reference load three sessions in a row and get three different estimated 1RMs: 138 kg, 146 kg, 133 kg. Nothing else changed — same sleep, same meals, same warm-up. Which number do you program off? Most lifters grab whichever reading they saw most recently, which is backwards, since a single session is the least reliable data point you own.

This is the default behavior of velocity-based 1RM estimation when it is treated as one measurement instead of a signal buried in noise. Part of that 8–13 kg swing is measurement error from the sensor, the rep chosen, and the extrapolation math. Part of it is a genuine change in neuromuscular readiness. One you ignore, the other you act on — and mixing them up is why athletes lose trust in autoregulation within a few weeks.

Two Different Problems That Look Identical

A daily estimated 1RM that bounces around has exactly two explanations, and the fix for one is the opposite of the fix for the other.

Measurement noise: the underlying value barely moved, but the way you measured it introduced random error — sensor placement shifted a centimeter, you compared reps from different points in the warm-up, or you extrapolated a slope built from loads far from where you actually tested. Noise is symmetric, uncorrelated session to session, and tracks nothing in your training log.

Real readiness drift: your force-generating capacity actually changed from accumulated fatigue, sleep debt, illness onset, or a genuine strength gain. Real drift persists across 2–4 consecutive sessions, correlates with wellness scores or jump height, and moves in a direction consistent with your recent training load.

The mistake most lifters make is treating every single-session number as real drift and reprogramming the week around it. Reliability statistics — borrowed from sports science rather than strength coaching — give a formal way to avoid that, and it is the backbone of everything below.

Where the Noise Actually Comes From

Courel-Ibáñez, Martínez-Cava, Morán-Navarro et al. (2019, Sensors) tested five bar-velocity technologies — linear position transducers, an IMU, a smartphone app, and an optical system — across a range of movement speeds. Reliability was strong at moderate-to-fast velocities (CV roughly 3–5%) but degraded sharply at the slow velocities typical of near-maximal loads, with CV climbing to 10–20% depending on device. The closer you test to true 1RM, the noisier the raw reading gets, even though that is exactly the region where accuracy matters most.

A second source is extrapolation distance. Pérez-Castilla, García-Ramos and colleagues, in load-velocity profiling studies published 2018–2020, showed predicted 1RM shifts meaningfully depending on which loads build the regression line and how far the prediction stretches beyond the tested range. A profile built from 40–70% 1RM and stretched to 100% carries more error than one anchored closer to true max. Most daily checks use a single ~65–70% reference load and extrapolate a long way — fast and low-fatigue, but structurally noisy.

Third, rep selection adds variance unrelated to readiness. Logging the first rep off a cold bar versus the second rep after a proper ramp can produce a 3–6% velocity difference from potentiation alone.

Stacked together, these sources alone can produce an 8–15% apparent swing between two sessions where physical readiness never changed — a range that overlaps almost entirely with what people blame on fatigue or bad sleep. That overlap is exactly why the two problems get confused.

Calculate Your Typical Error and Smallest Worthwhile Change

Will Hopkins' reliability framework (Hopkins, 2000, Sports Medicine) answers whether a change is real without guessing. Two numbers matter.

Typical error (TE) is how much your measurement bounces from pure noise when nothing changed. Calculate it from repeated paired-day measurements: take the standard deviation of the differences between trials, then divide by the square root of 2.

Smallest worthwhile change (SWC) is the smallest change that would actually matter for training decisions. Hopkins' standard heuristic sets SWC at 0.2 times the between-athlete standard deviation — for estimated 1RM in trained lifters, that typically lands around 3–5% of 1RM.

The rule that falls out: a change smaller than roughly twice your typical error cannot be distinguished from noise. Only changes clearing both the noise threshold (2×TE) and the meaningfulness threshold (SWC) deserve a training response.

Measuring Your Own Typical Error

1. On two separate days in the same week, run your readiness check exactly as usual — same load, same rep count
2. Record the estimated 1RM each day
3. Repeat this paired test 5–6 times over several weeks, only on days with no obvious fatigue confound
4. Take the standard deviation of the differences and divide by 1.41 — that is your typical error in kg
5. Convert to a percentage of your average 1RM for easier daily reading

Build a Rolling Baseline Instead of a Single Number

Once your typical error is meaningful — and it always is, nobody's is zero — the fix is structural: stop comparing today's reading against yesterday's. Compare it against a rolling baseline built from several recent sessions.

Rolling Median (Simple, Manual-Friendly)

1. Log estimated 1RM from every check for at least 5 consecutive sessions on the same lift
2. Take the median — not the average, which outliers pull further — of the last 5 values as your working baseline
3. Each new session, drop the oldest value and add the newest, then recalculate
4. Reprogram load only when today's reading falls outside the baseline by more than your 2×TE band

EWMA (Better for Automated Tracking)

The exponentially weighted moving average weights today's baseline as: Baseline(today) = α × Today + (1 − α) × Baseline(yesterday). An α of 0.2–0.3 balances things well — recent sessions matter more, but one noisy reading cannot swing the baseline by more than 20–30% of the gap to the prior value. Same logic used in acute:chronic workload monitoring, applied here to 1RM instead.

Diagnostic Checklist: Is This Swing Real?

When today's reading falls outside your baseline band, run this before touching the session:

• Does it clear 2× your typical error? If not, log it and train the planned session — this kills most false alarms.
• Was the setup identical? Same mount, same warm-up length, same rep number logged. A missed detail here is the most common cause of a false swing.
• Does it persist for 2+ sessions? A one-off low reading followed by a normal one is almost always noise. A low reading that repeats or worsens is almost always real.
• Does it correlate with an independent marker? Cross-check jump height, resting heart rate, or a wellness score. Real drift shows up in more than one measure; noise usually shows up only in the velocity number.
• Does the direction make training sense? A drop two days after your heaviest session of the block is expected. A drop with no preceding load spike, illness, or sleep disruption is more likely an artifact.

Fixing the Protocol When It Is Noise

If your typical error comes back above 6%, the swings are mostly avoidable. Work through these in order — ranked by how much variance each removes.

1. Standardize the logged rep. Always use rep 2 of the check, after the same warm-up, never a cold first rep. Typically cuts noise 2–5%.
2. Move the reference load closer to your training zone. If working sets sit at 80% 1RM but the daily check uses 65%, you extrapolate across 15 percentage points every day. Bring the reference load within 10 points of what you actually plan to train at.
3. Lock the device mount. Mark the exact sensor position with tape. A 2 cm shift on an IMU changes the moment arm it measures from.
4. Refresh the full profile every 4–6 weeks. A stale slope adds systematic bias on top of session-to-session noise — the two compound rather than cancel.
5. Test a heavier anchor point periodically. Every 3–4 weeks, add one set at 85–90% 1RM alongside the usual submaximal points. This shortens the extrapolation distance and tightens every daily estimate that follows.

Worked Example: Ten Sessions of Data

An athlete with a baseline back squat 1RM of 145 kg runs a daily check at 100 kg (69% 1RM) for ten sessions. Typical error, from a prior paired-day test, is 4.1% (roughly 6 kg on 145 kg).

A raw-reading approach would have treated session 6 as an emergency and session 9 as a breakthrough, missing that sessions 6–7 formed a real two-session dip while session 9 alone wasn't confirmed until session 10 backed it up. The rolling median caught the dip at session 7 and correctly waited before trusting the apparent gain.