Is it possible to understand, when does semantic predicates work and they doesn't?

[dims12]

Below are two grammars.

In this grammar, semantic predicates do "work". I.e. if they are false, rules don't match and if they are true, rules do match:

parser grammar MyParser;

options {
    tokenVocab=MyLexer;
}

expr
    : term
    | expr asterisk expr (asterisk expr)*
    | expr plus expr (plus expr)*
    ;

plus: {_input.LT(1).getText().equals("+")}? SPECID;
asterisk: {_input.LT(1).getText().equals("*")}? SPECID;


term
    : ALNUMID
    | STRID
    | LPAREN expr RPAREN
    ;

and in this grammar, semantic predicates don't work. I.e. if they are false, rules still match bu error is printed.

expr
    : add
    | mult
    | term
    ;


mult
    : term (asterisk) term ((asterisk) term)*
    ;

add
    : (term|mult) plus (term|mult) (plus (term|mult))*
    ;


plus: {_input.LT(1).getText().equals("+")}? SPECID;
asterisk: {_input.LT(1).getText().equals("*")}? SPECID;



term
    : ALNUMID
    | STRID
    | LPAREN expr RPAREN
    ;

In both cases I am trying to run simple tests to parse 2 + 2 and 2 * 2.

Is it possible to understand beforehand, will definition work or not? What is the logic here?