If you connect the "high" side of two different single-ended signal sources which both have their lows sides connected together (grounded or connected to a "common") to the two ends of a "floating" single transformer winding, the secondary of the transformer will have output which is the difference between the two signals, not the sum.
If the signals are unrelated, it will appear to be a "summed" signal, but with a phase difference of 180 degrees between the two input sources.
If the two input sources represent the exact same signal (same voltage, in-phase), then there will be no output because at any point in time the instanteous voltage on the two transformer pins is always exactly equal, thus there is no voltage difference across the winding, so no output in the secondary.
If the two input signals are 180-degrees out of phase, then the output will be twice the amplitude of the two input signal voltages. That's how balanced inputs work to eliminate induced noise on mic cables (which is in-phase on both conductors), while transmitting the mic signal which is of opposite polarity (180 deg. "out of phase") on one conductor.
"Summming" transformers usually have two independent primary windings that keep the two inputs isolated (4 input wires, two "grounded" (single-ended) or connected to signal "low-side" (balanced), and two connected to the signal "high-sides") and the input windings are marked with a "polarity" identification. By connecting the inputs properly, two inputs can actually be summed (added in-phase).
Summing = adding signals in-phase
Subtracting (or "taking the difference") = adding signals 180 degrees out-of-phase
Try setting your two signal generators to the same frequency and same amplitude (say 400 Hz) As they drift in and out of "sync" the signal at the transformer's output will go through a point where the signals cancel for an instant, and then they add or subtract as soon as there's a phase difference between the two generators.
To better observe what's actually happening, try synching your 'scope (externally) to one of the signal generators to use as a stable time reference, and then slowly change the frequency of the other generator.
Synching to the output signal will make it more difficult to understand what's happening since the 'scope will alternately sync to one or the other generator signal. If that's how you were observing the output, it may explain why you were seeing "somewhat unpredictable" results.
Hope this makes sense.