Three copies, one missed your write. Your read still sees it.

transcript

Three copies of your data. No leader. One copy missed your write. Can your next read still see the new value? Every time.

The client writes x equals 5 to two of the three nodes. A and B. Node C misses it. It still says 1. Now read from two nodes. Worst luck: you pick B and C. One says 5, one says 1. Keep the highest version. 5.

Why did that work? You wrote to 2. You read from 2. 2 plus 2 is 4. That's more than 3 nodes. So the two sets must share a node.

But here's the catch. Write 2, read just 1? Faster, right? 2 plus 1 is exactly 3. Not more. That one node might be C. Read 2 instead, and one of them always has the write.

And you can tune it. Read 1, write to all? Fast reads. Read all, write 1? Fast writes. 5 nodes, 3 each way? Survives more failures.

No leader needed. Just one inequality. Reads plus writes must beat the replicas. Then they always overlap.

Full breakdown in the lecture.