Sorry, "reply" instead of "reply all" again.

Well, it seemed impossible to me, but here what I found in d_math.c (pd 0.40.0 or something, I'm a bit lazy to look for version):

static t_int *sigwrap_perform(t_int *w)
{
    float *in = *(t_float **)(w+1), *out = *(t_float **)(w+2);
    t_int n = *(t_int *)(w+3);
    while (n--)
    {
        float f = *in++;
        int k = f;
        if (f > 0) *out++ = f-k;
        else *out++ = f - (k-1);
    }
    return (w + 4);
}

Here it definitely says:
if f <= 0 then return f - ((int f) - 1) = f - (int f) + 1 = frac(f) + 1
which is 1 for each non-positive int, of course, and assumes float-to-int conversions (int k = f) take 'neares to zero' which seems a dangerous assumption for me (I guess, it should return frac(f) + 1, where 0<=frac<1 for each negative number; perhaps, some compiler flags fix it, it depends on compiler, but for integers the result is always 1).
I think, the whole situation is worthy a bug report.