Посчитаем погрешность вычисления коэффициента подобия: погрешность координат 1e-5, значит, погрешность расстояния — 3e-5, погрешность коэффициента при максимальном стократном отношении размеров — 3e-3.
Понятно, что точности 1e-3 недостаточно для сравнения коэффициентов подобия.

I tried to follow the topic but the translator russian - english was not enough. I have a program which gets WA in test 44. Actually I proved changing precision issues and the best I got was WA in test 90, so I think the only bug in the code is about precision.
The function that deals with precision in my code is:

bool check(par *q, double t) {
  int i, j, orden[12];
  bool usado[12];
  memset(usado, 0, sizeof(usado));
  for (i = 0; i < n; i++) {
    for (j = 0; j < n; j++) if (!usado[j])
      if (dist(q[i].first - p[j].first, q[i].second - p[j].second) < eps) break;
    if (j == n) return false;
    usado[j] = true;
  }
  return true;
}

Array p keeps the input points and array q the points in a possible domino tile. The function check returns if the points in both arrays are "more or less" the same

The function dist is:
double dist(double x, double y) {
  return fabs(x)+ fabs(y);
}

and the eps is defined to be 1e-3

It would be great if anybody who got AC explains how he dealt with precision in his code