Writing up the argument you lost

Everyone publishes the optimisation that worked. Far fewer publish the one they were certain about that did nothing, or the objection from review that turned out to be correct. The second kind is more useful to read and much more convincing about the author, because it demonstrates the thing the whole phase has been training: updating on evidence rather than defending a position.

Lead with the numbers and the method that produced them. What you measured, on what device and connection, at what percentile, before and after. A claim of "40 per cent faster" with no method behind it is worth nothing to a reader and, more importantly, worth nothing to you in six months when you want to know whether it held.