Could you please, explain why any two guesses (even two positive) would work for calculating the IRR? I guess that would be true if at least one gives NPV pretty close to zero or am I wrong?
Any two guesses can be used to estimate the IRR (whether both positive or whatever). The answer is always only an approximation because we assume linearity which does not exist. One positive and one negative give better approximations, but for the exam that does not matter (provided the guesses were not stupid guesses).
For more, watch the Paper F2 lectures on this (because this is revision of both F2 and F9).