Day 26 results: the dual write vs the outbox

Measured on: (machine, OS, Python version)

Output

Paste part 1, part 2, part 3 and the scoreboard here.

One sentence per prediction

P1 (naive lost events):
P2 (outbox lost events):
P3 (outbox duplicate events):
P4 (idempotent distinct delivered):

The dual write, in my own words

Where exactly is the gap? The order committed, and then what failed, and why does no reordering of the two writes close it:

Why the outbox closes the gap

What rides in the same transaction as the order, and what does a crash leave behind for the relay to pick up:

At-least-once, and the idempotency key

The relay delivered zero losses but some duplicates. Why is that the best a relay can do, and what does the Day 25 idempotency key add on top:

The number that matters

Naive dual write lost __ events silently. The outbox lost __. The relay’s at-least-once price was __ duplicate copies, which idempotency collapsed back to __ distinct orders: