The Cartographer's Magnetic North: On the Drift of a Shifting Baseline

For centuries, mariners have trusted the compass, a needle drawn unerringly to magnetic north. It is a fixed point, a constant in a world of variables. Yet every seasoned cartographer knows a quiet truth: magnetic north is not fixed at all. It drifts. It wobbles. It is a living point on the globe, slowly migrating year after year. The maps must be updated, the compasses recalibrated, lest a course plotted by an old chart leads a ship gently, imperceptibly, into danger.

In our world of service reliability, we have our own magnetic north: the baseline. It is the set of thresholds and performance metrics we establish to define ‘normal.’ A latency under 200ms. An error rate below 0.1%. A memory footprint that hovers around 60%. We set our alerts to these fixed points, confident that any deviation will sound the alarm. But like the cartographer’s north, our baseline is not a permanent truth. It is a temporary agreement with a dynamic, evolving system.

This is the phenomenon of ‘shifting baseline syndrome,’ a term borrowed from ecology. It describes how each generation comes to accept the state of the environment they first experience as normal, even if that state is already degraded from a previous, healthier one. We see it in our dashboards. The p95 latency creeps up by 5ms each month. The database query, once snappy, now takes a few milliseconds longer. The memory leak is so slow it’s measured in kilobytes per day. Individually, these changes are trivial, well within our tolerances. They don’t trip a single wire. But collectively, over a year, they represent a fundamental drift in the health of our service. Our ‘north’ has moved, and we haven’t noticed.

The lesson from the cartographer is not to find a more permanent north—that’s impossible. It is to build the practice of recalibration into the very fabric of our work. We must periodically re-survey the territory. This means intentionally reviewing and questioning our baselines. Is 200ms still the right threshold for latency, or has user expectation shifted? Are we measuring the right things, or have new failure modes emerged that our old gauges can’t detect?

True observability provides the tools for this—not just the dashboards showing current state, but the historical trends that reveal the drift itself. It allows us to see not just where the needle points today, but the entire path it has traveled. By acknowledging that our baselines are living entities, we stop being passive observers of a slow decline and become active navigators, constantly correcting our course against the true north of user experience and system integrity, not the drifting magnet of what we’ve merely grown accustomed to.

Notes & further reading

A few pages I came back to while writing this: