The Wobble: What the Unstable Chair Teaches Us About Feedback Loops

There is a particular kind of instability that every person who has ever worked in an office, sat in a café, or owned poorly assembled furniture knows intimately. It is the wobble. One leg of your chair is a fraction of an inch shorter than the others. The correction—a coaster, a folded napkin, a wad of paper—is immediate but imperfect. You spend the day in a low-grade, persistent negotiation with gravity, making micro-shifts in your posture, learning the exact angle of lean that brings temporary equilibrium. The wobble is not a complete failure; the chair holds. But it demands constant, conscious feedback.

Our services are full of wobbles. They are the latent instabilities we learn to live with, the slight degradation in a database connection pool, the incremental memory creep in a long-running process, the third-party API that adds 50ms of latency every Tuesday morning. Like the chair leg, they do not trigger a catastrophic ‘down’ event. Our health checks, often binary pass/fail gates, sail right past them. The service is ‘up,’ technically answering its ping. But it wobbles. And everyone who leans on it feels the unsettling shift.

This is where the simple uptime check fails and a more nuanced observability must begin. The wobble teaches us that stability is not a state, but a dynamic, continuous process of correction. It is a feedback loop. Your body on the wobbly chair is a superb observability platform: it senses the tilt (metrics), interprets the discomfort (traces), and executes a compensatory shift (action). The problem isn’t the initial imperfection; it’s the absence of a built-in, automated feedback mechanism to correct it. We tolerate the wobble because the manual process of sensing and adjusting has, tragically, become part of the ‘normal’ workload.

From Detecting Failure to Sensing Strain

To build truly reliable systems, we must move beyond monitoring for failure and start instrumenting for strain. A health check that only asks “Are you alive?” is like testing the chair by seeing if it collapses under full weight. Useful, but far too late. We need checks that measure the pressure on each leg. What does the wobble look like in our systems? It’s the 95th percentile latency creeping upward while the median holds steady. It’s the error rate for a specific user cohort. It’s the growing gap between committed and virtual memory. These are the precursors to the fall.

The goal is not to eliminate every microscopic imperfection—that’s the pursuit of the perfectly level floor in a world of settling foundations. The goal is to shorten the feedback loop to the point where the correction is automatic, seamless, and often, preemptive. It’s the engineering equivalent of a self-levelling chair leg, one that senses the uneven load and adjusts before the sitter does.

Tomorrow, notice the wobbles. In your code, in your infrastructure, in your own routines. That slight resistance when a container starts, the extra query that ‘shouldn’t matter,’ the manual script you run every Thursday. Each is a leg needing adjustment. Reliability isn’t built in the grand gestures of post-mortems and failovers alone, but in the quiet, relentless work of spotting the tilt and slipping the metaphorical coaster underneath, long before anyone has to consciously shift their weight.

Notes & further reading

A few pages I came back to while writing this: