Exercise 3: Why the 30%/3-of-10 Match Mattered — Possible Solution
====================================================================

WHY THE MATCH WAS SIGNIFICANT
------------------------------
Per this chapter, checking each instance's version "confirms exactly
the mid-rollout pattern above - 3 of 10 instances on the new version.
3 out of 10 is 30%, matching the failure rate almost exactly." A
coincidental match between the proportion of requests failing and the
proportion of instances running new code is strong, near-conclusive
evidence that the failures are specifically tied to which version
handles a given request - if the cause were unrelated to version skew,
there would be no reason for the failure rate to line up numerically
with the new-version instance count at all.

WHY THIS WAS BETTER THAN JUST NOTICING "IT STARTED AFTER A DEPLOY"
------------------------------
Timing alone (failures starting right after a deploy) is suggestive
but not conclusive - plenty of things could coincidentally start
around the same time as a deploy. The specific numeric match between
failure rate and new-version instance proportion is a much stronger,
quantitative confirmation that ties the symptom directly to the
rollout's own instance mix, not just its timing.

THE ACTUAL ROOT CAUSE
------------------------------
Per this chapter, "the new version's code expects a display_name field
that a migration was meant to backfill onto every existing user
record - but the backfill is still running and only partially
complete. Requests landing on new-version instances fail specifically
for the subset of accounts the backfill hasn't reached yet." The root
cause was new code deployed ahead of a still-incomplete backfill
migration - exactly the lockstep-timing mistake the expand-contract
pattern exists to prevent.

WHY THIS WORKS AS AN ANSWER
------------------------------
It explains why the specific numeric correspondence (not just the
timing) was strong evidence, and correctly identifies the underlying
cause as an incomplete backfill combined with code deployed ahead of
it, connecting back to this chapter's own expand-contract discussion.