Juggling rings are different from juggling balls in one obvious way: they are circles.
That sounds trivial until you have juggled both. Rings fly differently. They have to be kept vertical or they wobble. They spin on their own axis as they travel. To throw one cleanly, you have to release it with a consistent spin and trust that it will stay aligned across the arc.
And every time a ring comes back to your hand, it looks the same as when it left. A ring has no front or back. It has no natural resting position. The throw does not change it. The catch does not change it. It simply returns.
This is, structurally, the definition of a loop.
What MLOps borrowed from the ring
Machine learning operations - MLOps - is the discipline of keeping a model working after it has been deployed.
The naive assumption is that this is about maintenance. You train a model, you deploy it, and then you watch it to make sure it does not break. But that is not quite right. The world that the model was trained on is not the world it is operating in. Data changes. User behavior shifts. What the model learned last year may not describe what is true this week.
Rings require a different kind of attention
When you juggle balls, small inconsistencies in a throw are self-correcting. A ball that is slightly off-axis still lands roughly where you expect it.
Rings are less forgiving. A ring thrown with the wrong spin will wobble and make the catch awkward. You cannot fix a bad ring throw mid-flight. You can only notice it, catch what you can, and make the next throw cleaner.
This teaches something specific: feedback has to be immediate, and correction has to happen on the next cycle. Not three cycles from now. Not after you have accumulated enough data to feel confident. The pattern depends on each throw being slightly better informed by the one before.
In an MLOps context, this looks like having monitoring in place before you need it. When something drifts, you want to know at the next throw, not after the pattern has already broken.
The ring reminds you of its shape
Here is the thing about rings that makes them good teachers.
When a ball comes back to your hand, it looks the same as every other ball you have ever caught. It does not carry a record of its own path. But a ring is literally shaped like its trajectory. Its form is a circle. It is, at all times, showing you the loop.
MLOps systems work best when the loop is visible in the same way. When the team can see: here is the model, here is what it is producing, here is how that compares to what we wanted, here is what we are going to do about it. Not as a one-time review but as a continuous display.
Three rings, three loops
When you juggle three rings, you have three separate loops running through your hands simultaneously.
Each one is independent. Each one has its own timing. But they share a rhythm - a common beat that keeps them from colliding or getting out of phase.
This is how production ML systems work at any real scale. Multiple models, multiple feedback loops, different data sources and update cadences - but all running on a shared understanding of what good performance looks like and how drift gets detected and addressed.
The rings do not wait for each other. They trust the pattern.
That trust is earned by making the individual loop reliable first. Get one ring moving cleanly before you add two more. The same principle as balls. The same discipline, applied to a different shape.
Start with one loop. Make it tight. Make it visible. Make it run on its own. Then add the next.
Read next: 99 Percent - the training metric and the loop that produced it. Three Props, Three Physics - why the ring’s circular shape produces gyroscopic stability that balls and clubs cannot match. The Physics of the Throw - the angular momentum equations behind the ring’s stable orientation.