Direct-drive smart trainers typically read within 1-2% of a reference across normal training power, with larger errors at the extremes and some unit-to-unit variation; whether ERG-mode training affects pacing skill is untested
Interpretation
Accuracy part, consistent across small studies: bias -1.1% at 250-700 W but 4.5% at 100-200 W and 13.0% above 750 W in one rig-tested unit (Zadow 2016), stable after 14 months (typical error 1.6%; Zadow 2018); five units averaged -1.5% (range -3.1% to 0.0%) with two shifting between tests (Hoon 2016); three units of another trainer agreed with a reference crank meter (ICC at least 0.996) in 11 cyclists (Lillo-Bevia 2018); a recent trainer matched dual-sided pedals in 10 riders aged about 46 (Morais 2024). Graded moderate for accuracy only: small samples, mostly rig tests, cyclists not triathletes, no wheel-on trainers, models that date quickly. The ERG part is limited: no study tested ERG-mode training; indoor power varied about half as much as outdoor (Jeffries 2019, 20 men), enforced constant pace was more demanding than self-pacing in 9 male rowers (Lander 2009), but even pacing lowered lactate in 10 male cyclists (Thomas 2012).
Practical implication
Treat trainer power as accurate to within a few percent across normal training power, check it against a second source if you compare indoor and outdoor numbers, and set zones from the device you train on. Use ERG mode for structured intervals but ride some sessions in free mode to practise pacing; that advice is inference, because ERG-mode training has not been tested against free riding.
Sources
8 sources, independently verified.
At 250-700 W and 80-120 rpm the trainer showed bias -1.1% (95% limits of agreement -3.6% to 1.4%); across 100-999 W bias was 4.2% with wide limits, driven by 4.5% bias at 100-200 W (limits -2.3% to 11.3%) and 13.0% at 750-999 W. Caution advised below 200 W and above 750 W. Single unit, 2016 model.
Over 100-600 W at 80-100 rpm, ratio bias between tests 14 months apart was 1.002 (limits 0.992-1.011); ICC 1.00; typical error 3.1 W (1.6%). Single unit.
Rig-driven incremental test from 50 to 400 W. KICKR mean error -1.5% (range -3.1 to 0.0%) vs -1.6% for the SRM; typical error 1.5% vs 0.7%; two KICKR units changed significantly between tests. One trainer model, 2016. Cited by repository URL because the listed DOI did not resolve.
At 100-500 W, 70-100 rpm, seated and standing, the three trainer units agreed with a reference crank power meter (ICC at least 0.996), with bias -5.5 to 3.8 W in most conditions and minor differences at the lowest and highest workloads; CV under 1.2% from 150 to 500 W.
A direct-drive smart trainer and dual-sided pedal power meters showed no significant differences at power stages from 100 to 270 W, good reliability, and more than 80% of Bland-Altman points within the limits of agreement. Small sample.
Randomised 20-minute time trials outdoors on a racing circuit and indoors on an electromagnetically braked ergometer, same bike and power meter. Mean power 280 W indoors vs 284 W outdoors, not different (P = .256), r = .94; within-person power variability far higher outdoors (SD 69 vs 33 W). All-male.
At matched average power (about 84% of maximal time-trial power), an enforced constant pace raised core temperature (38.72 vs 38.46 °C), post-test lactate (6.19 vs 5.24 mmol/L) and muscle activity compared with self-pacing at a fixed effort rating; performance times did not differ, and power varied more stroke to stroke when self-paced. Rowing, men only, single sessions.
In time- and work-matched 20-km trials, even pacing (at the self-paced mean power) lowered blood lactate compared with self-paced and variable-paced trials; the abstract reports lower effort ratings and more positive affect with even pacing, though its reported statistics for this are unclear. Men only, single sessions.