The Sourdough's Quiet Peak: On the Starter That Thrives on Neglect

In the world of operations, we are all apprentice bakers. We tend to our digital starters—those foundational services and data stores—with a ritualistic precision. We feed them at scheduled intervals, monitor their activity with vigilant probes, and panic at the first sign of sluggishness. The received wisdom is one of constant, careful management. A healthy system, we're told, is an actively managed system. But what if our compulsion to meddle is the very thing that saps its resilience?

I propose a heretical thought: sometimes, the most reliable services are those we learn to neglect. I call this the "sourdough principle." A sourdough starter, that living culture of flour and water, doesn't need—and indeed, suffers from—daily feeding if it's kept in a cool, dormant state. In fact, a period of quiet neglect in the refrigerator allows the culture to develop deeper, more complex flavors and a hardier constitution. Constantly fussing over it at room temperature forces frantic, weak growth and risks collapse. Our digital systems can exhibit the same traits.

We see this in the over-engineered backup regime. The common advice is to verify, test, and validate backups with relentless frequency. Yet, I’ve seen environments where this process itself becomes a primary source of instability. The backup windows consume resources, the verification scripts introduce their own failure points, and the constant I/O chatter obscures the genuine signal of the production workload. In striving for perfect assurance, we build a second, shadow system of complexity that can, and does, fail. Contrast this with a simpler, more robust system: a hands-off, incremental snapshot to immutable storage, checked not daily, but perhaps monthly, with a clear, manual process. It’s a less "active" system, but its stability comes from its simplicity and its infrequent interaction with the delicate ecosystem of production.

This philosophy extends to logging. The modern doctrine is to log everything, to create a firehose of data so that when something goes wrong, the answer must be in there somewhere. But this creates a problem of abundance. The sheer volume becomes a form of silence. We build elaborate parsing pipelines that themselves require maintenance and monitoring. What if, instead, we practiced strategic neglect? What if we defined a small set of critical, high-fidelity signals—the equivalent of watching the dough for a specific kind of bubble or rise—and let the trivial chatter fall away? The system’s health becomes clearer not by adding more data, but by having the confidence to ignore most of it.

The goal is not abdication, but cultivating a different kind of attention. It’s the shift from being a nervous helicopter parent to a wise gardener who understands dormancy. It means building systems with inherent stability, designed to be left alone, rather than systems that require constant intervention to remain upright. It’s about trusting the quiet hum of a process that has been given the space to find its own equilibrium. The peak of reliability isn't always a frantic peak of activity; sometimes, it's the quiet, robust vitality of a starter jar in the back of the fridge, silently gaining strength from the neglect we’ve finally learned to grant it.

Notes & further reading

A few pages I came back to while writing this: