This question is a follow up on the question about construction the triangle given distinct altitude, bisector and median.
The answer provides an expression for the side length $a$ as the two roots of quadratic expression, and the expression for the other two side length, $b,c$ in terms of that $a$:
\begin{align} a_1&= 2\sqrt{m_a^2-2h_a^2 + (2h_a^2-\beta_a^2)\sqrt{\frac{m_a^2-h_a^2}{\beta_a^2-h_a^2}}} \tag{1}\label{1} ,\\ a_2&= 2\sqrt{m_a^2-2h_a^2 - (2h_a^2-\beta_a^2)\sqrt{\frac{m_a^2-h_a^2}{\beta_a^2-h_a^2}}} \tag{2}\label{2} ,\\ b,c&= \tfrac12\sqrt{4m_a^2+a^2 \pm 4a\sqrt{m_a^2-h_a^2}} \tag{3}\label{3} . \end{align}
Numerical testing on the wide range (like, about 40000) of the different triangular shapes suggests that it's always the $a_1$ that results in a valid solution.
Any ideas of how to prove that analytically or a counterexample?


Let $AH$, $AK$ and $AM$ be respectively the given altitude, bisector and median from vertex $A$ of triangle $ABC$. Notice that $K$ always lies between $H$ and $M$ and suppose WLOG that $B$ and $H$ lie on the left of $M$, while $A$ is on the right (i.e. $AB<AC$, of course for $AB=AC$ the problem is underdetermined). I'll set: $$ h=AH,\quad k=HK=\sqrt{AK^2-AH^2},\quad m=HM=\sqrt{AM^2-AH^2} $$ and then denote by $y$ distance $AH$, by $x$ the signed distance $BH$: positive if $B$ is on the left of $H$ and negative if $B$ is on the right of $H$ (i.e. between $H$ and $K$).
From $BM=CM$ we get a first equation: $$ y=2m+x $$ while from the bisector theorem we have $$ (x+k)\sqrt{h^2+y^2}=(y-k)\sqrt{h^2+x^2}. $$ The second equation can be squared and simplified by factoring out $(x+y)$, finally yielding: $$ (h^2-k^2)(x-y)+2kxy+2kh^2=0. $$ Inserting here the first equation gives a quadratic equation for $x$ with solutions: $$ x=-m\pm\sqrt{m^2-h^2+{m\over k}(h^2-k^2)}. $$ But $x>-m$, hence the solution with a negative square root must be discarded, leaving a unique solution: $$ x=-m+\sqrt{m^2-h^2+{m\over k}(h^2-k^2)} $$ which gives: $$ BC=x+y=2x+2m=2\sqrt{m^2-h^2+{m\over k}(h^2-k^2)}. $$ Substituting here $m=\sqrt{m_a^2-h_a^2}$, $k=\sqrt{\beta_a^2-h_a^2}$ and $h=h_a$ one recovers your first solution. And this should also make clear why your second solution is not acceptable.