Russian strata (sing.: stratum) are a mechanism used to divide players in a tournament system aiming to funnel the best players to the top by breaking players into smaller tiers (strata) over time. Players with positive results after a given number of rounds get placed into upper strata, and players with negative results will eventually filter to lower strata. Of course, as times go on, large positives and small positives end up with different treatment on different strata. The final round of the format usually ends with strata of size 4 or 8 to determine final tables. Over the course of the tournament, positions are permeable with players from a middle stratum able to reach an upper stratum if their record improves. Depending on organizers, the final table may or may not have permeable rankings.
This is not to be confused with King of the hill tournaments, although in its most extreme interpretation could be interpreted as a Russian strata system with an enforced size of 4 every round after the 1st.
In order to induce an intended best-versus-best final, after a given number of rounds randomly drawn (e.g.: 2), future rounds will establish a stratum of players with an intended size, usually half the field. Pairings for these future rounds would be done within their stratum, with a desired pairing method (random, Siberian, or a vertical slice as examples (note: horizontal slices would lead to the King of the hill format and do not require forming strata)). After one or two more rounds, depending on tournament size and length, the field could be divided into three or four strata. Further rounds could keep either the same number of strata, add a few, or increase to double the previous round. Aside from timing a strata division, the other thing is establishing a ranking differentiation.
In order for the strata to have meaning, a ranking differentiation method should be public knowledge for the tournament in question. Accumulated score is one common method to derive a differential, however a rank score or average is also a method that can be used as a primary ranking condition. Secondary differentials might be required, given the discrete nature of the primary differential (using score will have fewer ties than rank, as there are only four ranks possible to obtain (barring the rare in-game tie).
Depending on the size of a tournament and strata parameters, the format could indeed induce a gradual separation of the best from the rest. While skilled players will likely make their way to the top and unskilled players to lower strata, this can equally be applicable to the lucky/unlucky, as well as people scoring/paying yakuman-grade hands. The paradox is that the steeper the stratification, the less likely that the best are retained and not just there due to luck. The later the start, and the flatter the top strata curve line is, the more likely that the most skilled players can be retained and/or the best can pierce through from lower strata. It also provides for people with a similar ranking to compete with each other, which is a desirable intent of such a format.
Aside from the direct opposite of the advantage related to early and steep curves favoring luck over skill, the tournament format does not have any inherent guarantee of limiting multiple matchups. Unchecked for, this format can easily create duplicate and even triplicate pairings.
The 2016 Japan House Cup held on March 12-13, 2016 in Moscow, Russia, is one prominent example of the use of Russian strata. Sessions 1 and 2 used random seating, and future rounds used random seating within each stratum. From Round 3, the stratum sizes were 24, 20, 26, 12, with a final seventh round using vertical slices in groups of 16 (thus, [1, 5, 9, 13], [2, 6, 10, 14], etc.).
The premise of the tournament format is interesting, more data would have to be collected as to judge for accuracy of the intended goals.