There is a version of streaming engineering that gets written about, and a version that keeps the site alive. The first involves recommendation models and personalisation. The second is a list of upstream sources, a retry order, and a health check that fires before a user notices anything wrong.
The relaunch from the 123movies team, now running as new 123movies, is a decent case study in the second kind, mostly because the team had no choice. The old operation died when one upstream provider went away, and the rebuild is structured so that no single source can do that again.
Uptime is invisible until it fails
Money, and quality consistency. Every additional source is another contract, another set of health checks, another place where the same film is available at a different bitrate. The user experience of a multi-source catalogue is that some titles start at 1080p and others at 720p, depending on which endpoint answered first.
A catalogue where every title is perfect until the day none of them play is not a better product.
That inconsistency is the price of uptime, and it is worth paying. A catalogue where every title is perfect until the day none of them play is not a better product. It is a more fragile one.
Redundancy is bought with contracts, not code
Most failures are not catalogue failures, they are player failures: a stream that starts, stalls, and leaves the user staring at a spinner with no message. Handling that in-page, with a retry and a fallback, is a small amount of code that decides whether a visitor comes back.
Time to first frame, not page load. Fallback rate per source, so you know which provider is degrading before it dies. And the percentage of sessions that end in an error state, because that number is your real churn metric.
Fallback logic is the whole product on a Friday night
None of it is glamorous. All of it is what separates a site that survived a rebuild from one that will need another in eighteen months.