The Odds of a Breakdown

Same pipeline as last time, different question. The walk-off effect post was about what happens after a ride goes down. This one is about the part before that: how often does it happen in the first place, and is there any pattern to when?

The overall odds

For the 10 rides with the most breakdowns since June, here’s the share of every 5-minute reading where the ride showed DOWN instead of OPERATING — the honest answer to “what are the odds this ride is broken right now.”

Two of these aren’t like the others. Disneyland Monorail is down 40% of the time. Golden Zephyr is down 34% of the time. Everything else on this list — including some real headliners in earlier posts — sits under 14%. That’s not those two rides breaking down more often than everything else; it’s that each outage runs long. Both have some of the longest average downtime-per-incident in the whole dataset.

When it happens

Same 10 rides, broken out by day of week and hour of day — a heatmap of the odds, and a scatter of the actual breakdown events, whichever reads clearer to you. Hover a cell for the exact number.

The heatmap makes Monorail and Golden Zephyr’s problem look like a solid block through the middle of the day, every day of the week — not a time-of-day thing so much as an always-somewhat-broken thing. But switch to the scatter and a second pattern shows up that the aggregate numbers hide entirely: both rides also cluster breakdowns in the last couple hours before close, separate from their daytime pattern. I don’t know yet if that’s a real end-of-day mechanical thing or a data artifact from how closing procedures get logged — flagging it as something to watch rather than a conclusion.

Does the weather matter?

Widened this part to every ride in the park, not just the top 10 — joined each wait-time reading to the nearest hourly weather observation.

Temperature is the cleaner signal: a steady, monotonic climb across five buckets and roughly 950,000 readings, from 4.1% down under 70°F to 9.0% down at 100°F+. That's not noise — it's a real doubling, and it tracks with the obvious mechanical-strain story.

Rain looks even more dramatic — breakdown rate roughly triples when it’s raining, and average wait climbs too, plausibly because fewer rides running concentrates the crowd onto whatever’s still open. But it barely rains here: only 5,834 of the 949,000 readings in this comparison were logged during rain, against 943,000 dry. Directionally real, statistically thin. I’ll revisit this once more rainy days accumulate in the dataset.

What’s next

Same as last time — this is query code now, not a one-off, so it’s re-runnable as the dataset grows. Halloween season is about to show up in the data, and I’d bet a Saturday in October looks nothing like a Tuesday in July for crowd-driven wear on the popular rides. The original plan was for this kind of pattern to eventually surface inside the app itself — a quiet “this one’s historically unreliable around now” hint on the rides that earn it.