The primary concern arises from the possibility that the actual block composition by miners may not align with the "all chunks up to x" model, leading to discrepancies in the chunk feerates when compared to the expected values based on one's own mempool data.

A key issue is the recognition that miners may have different mempool compositions, rendering any predictions based on an individual mempool potentially inaccurate for determining a miner's behavior. This raises the question of whether it is appropriate to base estimations solely on one's own mempool information.

Furthermore, there are risks associated with allowing miners to influence fee estimates. If miners can manipulate their block formations to cause substantial deviations from expected transaction inclusion, this could lead to skewed fee estimates. Such a situation would be undesirable as it might provide miners with undue leverage over fee prediction mechanisms.

Another layer of complexity involves the ability, or lack thereof, to identify the specific clusters or chunks that miners utilize during block construction. Any attempt to relinearize block clusters on our end could result in models that diverge from the miners' actual methods, which might not only be based on older protocols but could also involve proprietary strategies.

Despite these challenges, the suggested approach leans towards relying on our mempool's chunk feerate estimates at the time of block discovery. This decision is rooted in the rationale that our mempool likely provides the most accurate forecast of upcoming blocks, and therefore, the assumption that miners have similar visibility into transaction pools should be maintained, despite potential variations in actual access.