Fixing Velocity Readings That Vary Between Sessions

You loaded 100 kg on the back squat last Tuesday and the sensor read 0.61 m/s mean concentric velocity. This Tuesday, same bar, same plates, same athlete who says he feels identical, and it reads 0.71 m/s. That is not a 2% adaptation bump - it is a 16% swing on paper, large enough to move an estimated 1RM by 8 to 10 kg and flip an autoregulation call from add weight to hold. Coaches who see this pattern usually blame the sensor or the athlete's readiness. In most cases neither is the real cause. Garcia-Ramos et al. (2018) found between-session velocity reliability at fixed submaximal loads (coefficient of variation 4.7-9.1% depending on exercise) was consistently worse than within-session reliability (CV 2.1-4.3%) in the same subjects, meaning the extra noise gets introduced somewhere in the session setup, not the lift itself. The three usual suspects are sensor position, how the bar is loaded, and the order of the warm-up ramp - all fixable in one training block without touching the device.

Before changing anything, it helps to know where the error accumulates. The table below separates the three sources this guide addresses from device-level noise, covered separately in our noisy velocity readings guide.

Stacked together, the first three sources alone can account for a 0.09-0.23 m/s swing at a given load - enough to explain the 0.10 m/s gap in the Tuesday-to-Tuesday example above with no real change in the athlete. Each has its own fix, and they compound, so partial fixes leave partial noise.

An IMU or LPT clip that lands even 1-2 cm away from last week's spot changes the lever arm and the axis it reads from. On a back squat, mounting 2 cm closer to the sleeve collar versus 2 cm further out changes how much bar whip the sensor picks up, a difference measured at up to 0.03-0.05 m/s at loads above 70% 1RM. Athletes and assistant coaches who reattach the clip by eye every session are the single biggest source of this error - a clip that looks identical can sit a centimetre or two off in either direction, and the error is nondirectional, so it can inflate or deflate the reading depending on which way it drifted.

The fix takes under two minutes to set up permanently. Mark the exact mounting point on the bar sleeve with a paint pen or colored tape for every exercise you track, since squat, bench, and deadlift each need their own mark if they use different bars or positions. Photograph the mounted sensor from two angles the first time you set it correctly and pin that photo to the athlete's session notes; before every session, whoever mounts the sensor matches the live setup to the photo rather than to memory. For LPT tethers, mark the stand's floor position too - moving it 15 cm sideways changes the cable angle enough to introduce a real trigonometric error that grows through the range of motion.

Banyard, Nosaka and Haff (2017) reported between-session intraclass correlation coefficients (ICC) of 0.68-0.88 for velocity at submaximal back-squat loads when sensor remounting was not standardized, versus ICCs consistently above 0.90 once a fixed mounting protocol was enforced - a meaningful jump from a fix that costs nothing but a paint pen and a photo. The limitation: visual matching only works if it depends on a shared reference, not memory. When mounting duty rotates between coaches or the athlete self-mounts on off days, treat the photo protocol as mandatory, not optional.

Two bars loaded to the identical 100 kg do not necessarily move the same way. Courel-Ibanez et al. (2019) measured bar whip - the oscillation of a loaded barbell during the lift - at 2 to 5 Hz on standard 20 kg training bars above 70% 1RM, producing velocity overestimates of 0.04-0.09 m/s versus a stiffer 29 mm powerlifting bar under the same load. If your gym has more than one 20 kg bar and athletes grab whichever is free, or the same athlete alternates between a stiff bar on max weeks and a whippier one on volume weeks, the sensor is reading two different physical systems and reporting the difference as a performance change.

Plate loading order matters too: how load is distributed along the sleeve changes the bar's moment of inertia and, on bars with any flex, its oscillation frequency. Fix two habits: loading heavy plates innermost one week and outermost the next (pick one convention and write it on the wall by the rack), and mixing bumper plates with iron plates on the same working sets, since bumpers have a larger diameter and different mass distribution that changes effective barbell length. A loose or missing collar adds the same kind of noise - plates shift slightly during the concentric phase, and a single-axis sensor registers that wobble as extra velocity.

The practical standard: assign one specific bar per exercise per athlete (label it if the gym owns duplicates), load heaviest plates innermost every time, use spring or lockjaw collars, and never mix bumper and iron plates within a working set. Log the bar's ID alongside the session so a bar swap shows up immediately instead of surfacing as unexplained drift three weeks later.

The warm-up is not just preparation - it is the last variable that sets the nervous system's readiness state before the number that gets logged. A ramp that jumps straight from an empty bar to 70% 1RM in two big steps leaves the athlete in a different potentiation state than a five-step ramp with even increments and consistent rest, and that state genuinely changes bar speed at the working load, not just perceived effort. Athletes who warm up faster on a day they are running late, or add an extra unplanned set because the bar felt heavy, are introducing a real physiological variable that the sensor faithfully records as inconsistency.

Fix this by writing the ramp down as a fixed protocol per exercise rather than leaving it to feel. A workable back squat ramp for a 140 kg 1RM athlete: bar x8, 60 kg x5, 90 kg x3, 110 kg x2, 125 kg x1, each set separated by 90 seconds, with the first working set starting exactly 3 minutes after the final ramp rep. Write the loads, reps, and rest into the program itself rather than reconstructing it from memory each session. Keep the rest interval before the first working set within a fixed 30-second window - too little and too much rest each shift velocity in opposite directions - and always run the ramp on the same bar and sensor setup used for working sets, since warming up unmounted and attaching the sensor only for working sets skips the chance to catch a mounting error before it contaminates real data.

Garcia-Ramos et al. (2018), in the same analysis cited above, noted that a meaningful share of the between-session variance occurred specifically in the first working set of a session, consistent with incomplete or inconsistent potentiation rather than a device or bar issue - a useful diagnostic clue: if sets two and three are consistent between sessions but the first is not, the warm-up ramp is the likely source. One limitation: optimal ramp length is somewhat individual - athletes over roughly 40, or training early morning, often need a longer ramp to reach stable readiness - so personalize the protocol once, then hold it constant rather than re-optimizing session to session.

Rather than changing all three variables at once and not knowing which one mattered, isolate them across four sessions at a fixed reference load - typically 70% of the athlete's most recent tested 1RM.

Record MCV at the reference load each session and track running CV. Above 6-7% after session 1 is consistent with the uncontrolled baseline this guide describes; below 3-4% by session 4 means the three fixes removed most of the artificial noise, leaving what remains attributable to genuine day-to-day readiness. If CV stays elevated once all three are in place, the remaining variance is more likely a device-level issue - sampling rate, filtering, magnetic interference - covered in our troubleshooting guide, or worth a check against our sensor calibration protocol.

An intermediate lifter (back squat 1RM 135 kg, training age 2 years) tracked MCV at a fixed 95 kg reference load across six weeks. In weeks 1-2, with no standardized sensor mount, mixed usage between two visually identical 20 kg bars, and a warm-up that varied between 4 and 7 steps depending on available time, MCV at 95 kg ranged from 0.58 to 0.74 m/s across six sessions - a CV of 9.8%, with an estimated 1RM swinging between 128 and 149 kg despite no real strength change.

Starting week 3, the coach applied the four-session protocol above: sensor position marked with tape and photo-matched, one bar labeled and reserved for this athlete's squat sessions, heaviest plates loaded innermost with lockjaw collars, and a written five-step ramp with fixed 90-second rest. Over the following four sessions, MCV at 95 kg tightened to 0.63-0.68 m/s - a CV of 3.9% - and the estimated 1RM band narrowed to 138-143 kg, tight enough to support week-to-week load decisions with confidence. The athlete's actual tested 1RM eight weeks later was 141 kg, inside the stabilized band and nowhere near the noise-driven extremes from weeks 1-2. Nothing about the athlete's capacity changed in week 3; only the measurement conditions did.