Feedback creates groups
Imagine testing one guess against every answer still possible. Each answer produces a five-tile feedback pattern. Answers with the same pattern remain indistinguishable after that guess, so they stay in the same group. For example, a guess that splits 20 possible answers into five groups of four tells you more than one that leaves 16 answers together.
Expected remaining answers
One readable measure is the expected size of the group you will face after playing a guess. Large feedback groups count more heavily because they occur for more answers. The solver uses these simulated groups to estimate elimination power. Lower expected remainder means the guess is likely to narrow the board more sharply, although the exact outcome still depends on the hidden answer.
Why raw letter frequency is different
Letter frequency is a useful shortcut before any clues exist, but it cannot see positions, duplicate rules, or the shape of the current candidate set. Two guesses may contain equally common letters yet divide the remaining answers very differently. That is why the word library is only a general starting point, while live solver results are calculated for your clues.
Reading a recommendation
Treat the top result as a supported option rather than an instruction. Check whether it is a possible answer, how many candidates remain, and whether the guess tests the letters that distinguish them. With two attempts left, a slightly weaker information score may be worth accepting for a real chance to solve. With a large ambiguous set, a non-answer probe may be the safer decision.