Apparently in some contexts, it's not 'survival of the fittest [genome]' or 'survival of the fittest [allel]'; but when the mutation rate is sufficiently high, it's more like "survival of the flattest [set of genomes' fitness]" - with viroids, a viroid can have the most fit set of RNA but if the 'nearby' RNA sets are all low-fitness, it's so likely to stray into them that it becomes unfit compared to another viroid whose RNA set may be lower fitness but has less disastrous possible mutations. Weird!

"Viroids: Survivors from the RNA World?", Flores et al 2014 https://pdf.yt/d/1mRWEvpPH2aAp0dl / https://dl.dropboxusercontent.com/u/5317066/2014-flores.pdf / http://libgen.org/scimag6/10.1146/annurev-micro-091313-103416.pdf :

"As a consequence of their high mutation rates (see below), some viroids display high genetic variability. This phenomenon has been observed after inoculating plants with infectious viroid cDNAs (usually head-to-tail dimeric constructs) or their transcripts (mimicking replicative intermediates) and quantifying the heterogeneity of the resulting progeny. The diversity observed in the family Pospiviroidae, e.g., in PSTVd (57), Citrus exocortis viroid (5, 64, 135), Citrus bent leaf viroid (52), Chrysanthemum stunt viroid (CSVd) (16), and Citrus dwarfing viroid (131), is relatively low, and the consensus sequence does not deviate significantly from the wild type. However, in the family Avsunviroidae, e.g., PLMVd (2, 94) and CChMVd (16, 23), the resulting population is a complex spectrum of mutants and the consensus sequence often changes over time or is difficult to define. This difference is most likely caused by the distinct replication fidelity of the RNA polymerases involved (see below), with selection imposed by the host also shaping the final distribution of genetic variants (5, 122, 131, 139).
Given their small genomes and high mutation rates, viroids have often been treated in the framework of the quasi-species theory, originally developed to describe natural selection acting on primitive RNA replicons undergoing error-prone replication and consisting of mutant distributions (mutant spectra or clouds) (6, 34). This theoretical scheme was first applied to bacteriophage Qβ (33) and then to many other RNA viruses (84, 100, 134). A unique prediction of the quasi-species theory that distinguishes it from classic population genetics models is that selection operates on sets of variants from a particular region of the fitness landscape rather than on individual virus variants. As a result, the fitness associated with a specific sequence depends on the average fitness of its neighbors in the sequence space (84), as first experimentally shown with the RNA bacteriophage φ6 (11). This collective behavior can lead to the so-called survival of the flattest effect, whereby a population located in a region of the fitness landscape where a large proportion of sequence neighbors are selectively neutral can outcompete another population located in a higherfitness peak but with a more deleterious sequence neighborhood (121, 141). This effect was first demonstrated in vivo by competition experiments between CSVd and CChMVd, performed under different conditions: At the physiological mutation rate the fittest CSVd outcompeted CChMVd, but increasing mutation rate by UV irradiation reversed the outcome and the flattest CChMVd outcompeted CSVd (16). Later, the survival of the flattest was also shown in Vesicular stomatitis virus populations subjected to increased mutational stress by chemical mutagenesis (115)."