Exercise 3: "Auto-Scaling Kept Us Online, So Nothing Needs to Change" — Possible Solution
====================================================================

What's wrong with this reasoning:

This treats "the site remained technically reachable" as equivalent to
"there is no real problem" -- but per this chapter's own Step 9 and
`cloud1-3`'s original material, auto-scaling reacting to load is a
completely SEPARATE fact from whether the underlying issue that caused
the load spike to become a problem in the first place was ever
actually addressed.

In THIS capstone's specific scenario, per Step 9: "the scaling was a
reasonable reaction to real load, but it was also masking the
connection pool issue underneath the whole time." Even accounting for
the extra instances that were added, checkout STILL FAILED
intermittently for a growing share of customers during the incident --
auto-scaling adding capacity did not actually prevent the connection
pool exhaustion from causing real, customer-visible errors. "Kept the
site technically online" glosses over the fact that a meaningful
portion of checkout attempts were still failing throughout the
incident -- the site being reachable is not the same as the site
working correctly for everyone trying to use it.

Beyond that, per `cloud1-3`'s original warning (directly restated in
this chapter): auto-scaling masking a root cause means the underlying
problem "goes unaddressed... until the auto-scaling group eventually
hits its configured maximum instance count," at which point the
symptom reappears, "now at a larger and more urgent scale." If nothing
is changed here, the SAME connection pool exhaustion issue would very
plausibly recur during the NEXT high-traffic event -- and if that
future event pushes traffic even higher, or the max instance limit is
reached sooner, the failure could be considerably more severe than
this one, precisely because the actual root cause (Step 6's missing
load-testing process, and the undersized pool default) was never
fixed.

The correct framing, per Step 7: auto-scaling handling the immediate
symptom is not a substitute for the mitigation (fixing the pool size
directly) or the long-term fix (adding load-testing to the deployment
process) -- both of which the postmortem's action items specifically
address, and neither of which "the auto-scaler handled it, technically"
would have resolved on its own.

WHY THIS WORKS AS AN ANSWER
------------------------------
This directly refutes the colleague's premise using two pieces of
evidence from the chapter itself: (1) the scenario's own facts show
checkout was STILL FAILING for real customers despite auto-scaling
reacting, and (2) `cloud1-3`'s own warning about what happens once the
max instance count is eventually reached -- both showing that "stayed
online" was never actually true in the full sense the colleague's
argument requires, and that the underlying risk remains completely
unaddressed regardless.