Decide what counts as a movement before you look
With forty questions run three times each, a change of one or two percentage points is within the noise. Set the threshold in advance - commonly five points, or a change that holds across two consecutive rounds - and apply it to competitor figures too. Deciding afterwards which movements count is how a series becomes a story rather than a measurement.
Attribution is approximate, and that is acceptable
Several things shipped between rounds and the platforms changed underneath you. Clean attribution is not available. What is available: which specific questions moved, and whether the content and placements aimed at those questions were the ones that shipped. That is a weaker claim than causation and strong enough to decide the next round's budget.
Three outcomes
It moved. Identify the questions that moved, check whether the answers now cite anything of yours, and do more of what that was.
It did not move. Before concluding the content failed, check retrieval: were the new pages fetched at all, are the placements indexed, did a CDN rule change. A surprising share of flat rechecks are access failures rather than content failures.
It moved down. Usually a competitor shipped something, or a platform changed how it retrieves. Check whether your own figures held while the competitor's rose - a relative loss with stable absolutes is a different problem from an absolute decline.
What to carry into the next round
The same question set, the same conditions, and a short written note of what you changed and what you expect it to do. The note is what makes the next recheck interpretable; without it, the round after this one starts from nothing again.