Your measurement technique is invalid. See Fig. 1. Imagine that instead of using an oscilloscope you have a multimeter, and you connect only the meter's "+" probe to the test point on the device under test (DUT); you leave the meter's "-" probe disconnected from the DUT. The multimeter performs a voltage measurement and displays a voltage value. Is the displayed voltage value a valid representation of the voltage between the DUT's test point and the DUT's ground reference potential?
The answer is "It depends." If the multimeter's "-" input randomly happens to be at the same potential as the DUT's ground reference potential, then the voltage value shown on the multimeter is probably valid; otherwise it is definitely invalid.
Now imagine using two DMMs. The negative terminals of the two DMMs are connected together, but as before the negative terminals are not connected to the DUT. The positive terminals on the two DMMs are attached to two different test points on the DUT. Both DMMs make a voltage measurement. In this scenario, does each individual multimeter make a valid single-ended voltage measurement? If the answer is "No" (and it is "No"), one cannot now synthesize a valid differential voltage measurement by calculating the difference of the two invalid single-ended voltage measurements.
Also consider Fig. 2. Ask yourself this question: "What is the voltage across the resistor on the left side of the figure?" (which represents the input impedance at the oscilloscope's analog input). It's impossible to say what the voltage is because you cannot know voltage ΔV, the change in voltage between the oscilloscope's chassis ground (which is usually Earth ground) and the DUT's "floating" ground potential. (HINT: Using Kirchoff's Voltage Law, start at the oscilloscope's chassis ground potential and add up the voltage drops from that chassis ground node to the top of the resistor at the oscilloscope's input; that sum—i.e., the voltage across the resistor— is the voltage the oscilloscope measures.) Because voltage ΔV is undefined, the oscilloscope measures a random voltage.
Now imagine repeating Fig. 2 with two analog input channels on the oscilloscope, leaving the probe grounds disconnected and connecting the probes to different test points on the DUT. Both analog inputs perform a single-ended voltage measurement that yields a random voltage value for each channel. These two random voltages cannot be synthesized into a valid differential voltage measurement by subtracting one from the other.
To make this measurement you should use
A battery-powered oscilloscope that is not connected to Earth ground potential, OR
A differential voltage oscilloscope probe (Keysight Labs video on YouTube), OR
An oscilloscope whose input channel(s) can be safely isolated from chassis (Earth) ground potential, OR
A custom measurement setup that uses (for example) an isolation amplifier that provides galvanic isolation between the oscilloscope ground potential and the DUT's floating ground potential.