There is no true state any more

Why does a reader's choice of copy, not the writes, produce the anomaly once there are many copies?

Transactional isolation had one true state of the database. The only question was whether a reader saw a coherent cut of it (a consistent slice, as if taken at one moment). Here there is no true state. There are n copies. None is privileged. Each holds a different suffix of the write history (some have received the latest writes, some are still behind). "Is this value correct?" stops being a meaningful question. Only one question survives: "is this value part of an admissible history?"

Three categorical changes from Transactional isolation.

The anomaly is produced by readers, not writers. In Transactional isolation, fixing the schedule of writes fixed the problem. Here the identical set of writes, delivered in the identical order, is anomalous or not depending on how many vantage points exist and whether they compare notes. Add a read replica and you manufacture anomalies that did not exist before, although you changed no write path. This is the observer-count rule from A promise about histories. It is why consistency is the one cost across the platform that does not depend on data volume. It scales with the topology of observation: how many observers, how far apart.

The constraints become physical. The limits in Transactional isolation were combinatorial: NP-completeness, undecidability, shapes of graphs. The limits here are the speed of light and, more precisely, uncertainty about it. This matters when you judge a claim. These limits are permanent, not a description of today's engineering. r + w ≥ d (explained below) will not be optimized away by better software, and no amount of money moves it. Money only moves which side pays.

The hierarchy breaks in a new way. In Transactional isolation the levels failed to form one ladder because two families of mechanisms had collided in the names. Here the models fail to form one ladder because one of them is not the same kind of claim as the others (the next section).