[richieartoul]

Isn’t that ignoring the point that the read is running concurrently with w1 and w2? Your comment about “have the effect” only applies to transactions beginning after a write has completed, but the anomaly cockroach suffers from is for a read running concurrently with two writes with no overlapping execution. In that scenario they fail to be strictly serializable, but so would any system meeting only the two criteria you outlined I think.

Edit: Ah ok I see you edited it. Ok that makes sense if it works.

[_benedict]

yeah, to clarify also textually the "has the effect of" applies to all witnesses, including those that are concurrent. So if r executes after w1 or w2 then r must see that w1 applies before w2, since w1 "has the effect of" applying before w2, whether or not r is concurrent with them.