Human physiology has been running production monitoring for a few hundred million years, and it does one thing that most engineering dashboards do not: every single regulated variable comes with a declared band, and every sensor exists specifically to detect departure from that band.
That sounds like a small distinction. It is the whole difference between collecting numbers and actually knowing something.
What The Body Regulates, And How Narrowly
Core temperature sits near 37 degrees Celsius. Blood pH stays between 7.35 and 7.45. Fasting blood glucose holds between 70 and 100 milligrams per deciliter. Electrolyte concentrations, blood pressure, oxygen saturation and fluid volumes all live inside similarly tight ranges, and they hold those ranges while you sleep, sprint, starve or sweat.
Notice what is specified there. Not "we track pH." A number, and a band around it, decided in advance. The band is what makes a reading meaningful, because the entire regulatory apparatus downstream is built to answer one question: are we outside it yet.
Most service dashboards fail this test immediately. There is a p99 latency chart. There is no declared band. So when the chart reads 340 milliseconds, nobody on the team can say whether that is normal, drifting or already broken, and the honest answer usually turns out to be that nobody ever decided.
A Sensor Without A Set Point Is Not A Sensor
The physiological control loop has three parts: a sensor that detects deviation from the set point, a control center that evaluates the input, and an effector that acts to correct it. Remove the set point and the sensor degrades into a thermometer nobody reads.
This is what most observability stacks actually are. Enormous sensor coverage, near zero declared set points. Metrics get added because they were easy to emit, not because someone identified a variable that has to stay inside a range for the system to keep working. The result is dashboards that grow every quarter while the on call engineer still opens them and asks a colleague whether this looks bad.
The fix is boring and it is not a tool. For each metric that actually matters, write down the band, and write down what happens when you leave it. If you cannot state the band, that metric is decoration, and you should either derive a band or stop paying to store it.
Effectors Are The Part Everyone Skips
Detection is the cheap half. In the body, glucose rising past the band triggers insulin release, cells absorb glucose, and the value comes back down. Drop below the band and glucagon prompts the liver to release stored glucose. Two opposing hormones, one stable number. The correction is automatic and it is part of the design, not something a specialist gets paged about.
Plenty of teams have alerting and almost no effectors. Every deviation routes to a human, which means the system has no ability to correct itself and the humans slowly stop responding. That is not a monitoring problem, it is a missing actuator.
It is also worth knowing the failure mode on the other side. Positive feedback loops exist in biology, and they are rare precisely because they do not stabilize anything, they run a process to completion. Blood clotting works that way: activated clotting factors activate more clotting factors until the wound is sealed. The terminating condition is the whole design. Build that same loop in software without one and you have a retry storm.
The Takeaway
Biology did not win by measuring more. It won by deciding, for every variable worth defending, exactly how far it is allowed to drift and what happens automatically when it does.
If you want the full physiological version, all eleven organ systems, how each one is regulated and how they interact to hold those bands, the complete guide is here.
Pick your three most important metrics this week. Write a band next to each one. If you cannot, you just learned something more useful than any new dashboard would have told you.
Top comments (0)