There is a known asymptotic expansion of harmonic numbers $H_n$ for $n\to\infty$: $$\begin{align}H_n&=\gamma+\ln n+\sum_{k=1}^\infty\left(-\frac{B_k}{k\cdot n^k}\right)\\ &=\gamma+\ln n+\frac1{2n}-\frac1{12n^2}+\frac1{120n^4}-\frac1{252n^6}\,+\,\dots,\end{align}\tag1$$ where $B_k$ are Bernoulli numbers. We can take a linear combination of harmonic numbers to cancel constant and logarithmic terms, compensate for $O(n^{-1})$ term, and get the following series that is possible to evaluate in a closed form (e.g. using generating function): $$\sum_{k=1}^\infty\left(H_n-\,2H_{2n}+H_{4n}-\frac1{8n}\right)=\frac18-\frac\pi{16}.\tag2$$ Rather than compensating for $O(n^{-1})$ term, we can take a series with alternating signs, that is also possible to evaluate in a closed form: $$\sum_{n=1}^\infty\,(-1)^n\,\Big(H_n-\,2H_{2n}+H_{4n}\Big)=\frac{3\pi}{16}-\frac\pi{4\sqrt2}-\frac{\ln2}8.\tag3$$ Generalizing, we can consider two families of series: $$\mathcal A_m=\sum_{n=1}^\infty\,(-1)^n\,\Big(H_n-\,2H_{2n}+H_{4n}\Big)^m,\tag4$$ $$\mathcal B_m=\sum_{n=1}^\infty\Big(H_n-\,2H_{2n}+H_{4n}\Big)^m,\tag5$$ and try to evaluate them in a closed form.
So far I have the following conjectured result:
$$\large\sum_{n=1}^\infty\Big(H_n-\,2H_{2n}+H_{4n}\Big)^2\stackrel{\normalsize\color{gray}?}=\frac\pi8-\frac\pi{16}\,\ln2-\frac{\pi^2}{96}+\frac3{16}\,\ln^22-\frac{G}{4},\tag{$\diamond$}$$
where $G$ is the Catalan constant.
Could you please help me to prove this result and, possibly, find other values of $\mathcal A_m,\mathcal B_m$?
Just my thoughts for now: I would try to exploit Parseval's identity. For first, we have: $$ H_n-2H_{2n}+H_{4n} =\int_{0}^{1}\frac{-x^n+2x^{2n}-x^{4n}}{1-x}\,dx \tag{1} $$ and: $$ \sum_{n\geq 1}\frac{-x^n+2x^{2n}-x^{4n}}{1-x}e^{niy} = \frac{1}{1-x}\left(\frac{-1}{1-e^{iy}x}+\frac{2}{1-e^{iy}x^2}+\frac{-1}{1-e^{iy}x^4}\right).\tag{2}$$ The poles of the RHS (as a function of $x$) are located at $x\in\left\{e^{-iy},\pm e^{-iy/2},\pm e^{-iy/4},\pm i e^{-iy/4}\right\}$.
By using the residue theorem we may compute an explicit representation for: $$ g(y) = \sum_{n\geq 1}\left(H_n-2H_{2n}+H_{4n}\right)e^{niy},\tag{3} $$ then Parseval's theorem gives: $$ \sum_{n\geq 1}\left(H_n-2H_{2n}+H_{4n}\right)^2 = \frac{1}{2\pi}\int_{-\pi}^{\pi}g(y)g(-y)\,dy \tag{4}$$ and the resulting integral should be not to difficult to evaluate in terms of dilogarithms.
Another chance may be to apply summation by parts (like I did in this question), but it looks lengthy.