Solar flares exhibit complex variations in elemental abundances compared to photospheric values. These abundance variations, characterized by the first ionization potential (FIP) bias, remain challenging to interpret. We aim to 1) examine the spatial and temporal evolution of coronal abundances in the X8.2 flare on 2017 September 10, and 2) provide a new scenario to interpret the often observed high FIP bias loop top, and provide further insight into differences between spatially resolved and Sun-as-a-star flare composition measurements. We analyzed 12 Hinode/Extreme-ultraviolet Imaging Spectrometer (EIS) raster scans spanning 3.5 hours, employing both Ca XIV Ar XIV and Fe XVI S XIII composition diagnostics to derive FIP bias values. We used the Markov Chain Monte Carlo (MCMC) differential emission measure (DEM) method to obtain the distribution of plasma temperatures, which forms the basis for the FIP bias calculations. Both the Ca/Ar and Fe/S composition diagnostics consistently show that flare loop tops maintain high FIP bias values of $>$2--6, with peak phase values exceeding 4, over the extended duration, while footpoints exhibit photospheric FIP bias of sim 1. The consistency between these two diagnostics forms the basis for our interpretation of the abundance variations. We propose that this variation arises from a combination of two distinct processes: high FIP bias plasma downflows from the plasma sheet confined to loop tops, and chromospheric evaporation filling the loop footpoints with low FIP bias plasma. Mixing between these two sources produces the observed gradient. Our observations show that the localized high FIP bias signature at loop tops is likely diluted by the bright footpoint emission in spatially averaged measurements. The spatially resolved spectroscopic observations enabled by EIS prove critical for revealing this complex abundance variation in loops. Furthermore, our observations show clear evidence that the origin of hot flare plasma in flaring loops consists of a combination of both directly heated plasma in the corona and from ablated chromospheric material; and our results provide valuable insights into the formation and composition of loop top brightenings, also known as EUV knots, which are a common feature at the tops of flare loops.