It depends on the current sensor, so here's the honest ladder. With the factory-paired rbAmp CT — module and clamp calibrated together at the factory — it's ±0.5 % of reading per channel, guaranteed for your unit, across the bulk of the range (accuracy widens near the very low end, as it does for any meter). It measures true active power — the mean of instantaneous voltage × current over the line cycle, with per-channel phase compensation — not current × assumed-voltage apparent power.
With a generic SCT-013 clamp of your own, the clamp's unit-to-unit spread sets the budget: a few percent out of the box, tightening to ±0.5–1 % after a one-point calibration against a reference (supported in all four libraries). So a random clamp gets you close, but only the paired CT is guaranteed per unit — that guarantee is the difference, not a bigger headline number.
For a whole-panel figure summed across several modules, expect around ±1 % — module-to-module errors add up and don't average out, so we never quote a summed total as tighter than a single channel. More on the physics: how CT energy measurement really works.