Yeah I think most even degreed mets don't understand the true value of analogs, many create them, initializing with current conditions and treat their output as gospel and often only produce one or maybe a few sets of them without giving any due consideration to other variables. Appallingly, many even compare the analogs to climatological base periods that are far outside the temporal range of the analogs of interest (occurs most often with NOAA US PSD data where post 1950 analogs are used by are compared against the 1981-2010 base period and are thus depicted cooler than what they should be). If you really boil it down conceptually, analogs are basically a determinstic, statistical, historical model. If everyone treated these analogs like a numerical weather prediction model, with many of the same basic nuances, sampling, resolution, and dynamical uncertainties and biases, and that skill increasingly degrades with time and may even be state dependent (meaning that analogs in general may more perform better in reanalyzing and forecasting the weather), we would be a lot better off and wouldn't have nearly the issues we currently do wrt analogs. Any model is going to be an imperfect representation of the current state ocean and atmosphere in some way, shape, or form because we can't measure and evaluate the weather at every single point on the globe at all possible times to perfection, we are limited in our spatial and temporal coverage, instrumental and statistical interpolating uncertainities as well as other unforeseen, & often less significant sources of error. For analogs specifically, a few other sources of error emerge, from the broad assumption that the weather that's occurred in the past will repeat itself exactly in the future with no statistically significant dynamically changes at any point in time or space in any of the observed phenomena, & that the observed frequency distribution of said phenomena is Ironclad and/or completely set in stone and contains more than enough events s.t it can may yield significant results and similar deterministic historical depictions of the ocean and atmosphere would not diverge much from observations. Rather, (as I mentioned in a previous heated discussion with Larry), they should be treated as only one solution of infinitely many that are possible for a given climate background, thus, there exists an inherent amount of uncertainty that a forecaster, climatologist, or other scientist should know and they need to adjust or word their forecast accordingly. This all of course means that even if your analogs initialized the current state of the ocean and atmosphere perfectly, you have to be able to understand the physical processes, trends, and potential for stochastic external events that are and may drive the overall weather pattern in the future and only use your analogs, like an operational NWP model, as as a tool, not a forecast. Henceforth, this also means that you must dynamically adjust them to things that may not be adequately captured by the historical record, such as global warming and random variability, & the alterations to the general circulation that may accompany it.