Calibration is the process of comparing a sensor's reading against a known, trusted reference and recording how far off it runs, at one or more points across its working range. It does not fix the sensor. It tells you, in writing, how much to trust the number it reports, and by how much that number can be wrong before the sensor fails its own tolerance.
Every temperature data logger used in a regulated cold chain, and every fixed sensor built into a cold room or truck, needs a current calibration behind it before its readings mean anything to an auditor or a quality team. Calibration is not limited to the electronic loggers used in shipping either. Any fixed thermometer, controller probe, or handheld reference used to check a cold room, a stability chamber, or a delivery vehicle needs the same treatment, whether or not it ever leaves the building.
Traceable standards
A calibration is only as good as what the sensor was checked against. Traceability means the reference used, a calibration bath, a dry block, a certified reference thermometer, can be followed back through an unbroken chain of comparisons to a national or international measurement standard, each link in that chain carrying its own certificate and its own uncertainty figure. A sensor calibrated against a reference with no traceable chain behind it has a number on a certificate but no way to prove that number means what it claims to mean.
Independent accreditation, issued by a national accreditation body against an international standard for testing and calibration laboratories, is the usual way a calibration lab proves its own chain holds up. A certificate from an accredited lab carries that accreditation mark and scope; a certificate from an unaccredited lab is a claim with nobody independent standing behind it.
Calibration intervals
A sensor drifts over time: repeated temperature cycling, physical shock, humidity, and simple component aging all shift a reading away from what a fresh calibration recorded. An interval sets how long a calibration is trusted before it has to be repeated, commonly a year for a sensor in continuous service, though a sensor used in a harsh environment, or one feeding a high consequence application like a vaccine cold room, often sits on a shorter interval. Missing an interval does not make the sensor wrong; it makes its accuracy unproven from that date forward, which for an auditor is functionally the same problem. An interval is not fixed forever either: a sensor with a clean history of small, consistent drift across several calibrations can sometimes justify a longer interval, while one showing erratic or growing drift needs a shorter one. Calibration history is itself evidence for how long the next interval should run.
Tolerance and uncertainty
A calibration certificate reports two different things that are easy to mix up. Tolerance is how far a reading is allowed to drift from true before the sensor fails its calibration, set by whoever owns the equipment or the standard it operates under. Uncertainty is a property of the calibration itself, how confident the calibration process can be in the comparison it just made, expressed as a range around the recorded result. A sensor can pass its tolerance and still carry meaningful uncertainty; a usable certificate shows both, because a tolerance figure with no stated uncertainty behind it is an incomplete claim. A narrow tolerance paired with high uncertainty is worth catching early: claiming a sensor holds within a tight band while the calibration process itself cannot measure that precisely means the tolerance figure is not actually being demonstrated.
Certificates as the actual deliverable
The calibration certificate, not the sensor, is what an auditor asks to see. A usable certificate names the reference standard and its own traceable chain, states the as found readings before any adjustment and the as left readings after, records the conditions the comparison ran under, and carries a date, an expiry, and a signature. A sensor calibrated with no certificate to show for it might genuinely be accurate, but there is no paper trail proving it, and in a regulated cold chain an unprovable claim is treated the same as a false one. Certificates also expire; a sensor running past its certificate's expiry date, even if it was accurate the day it was last checked, is functionally the same as a sensor that was never calibrated, because nothing on file proves what it is doing today.
Uncalibrated readings prove nothing
A logger with no current calibration behind it can still record a clean, uninterrupted, plausible looking temperature curve for an entire shipment or an entire thermal mapping study, and that curve still proves nothing about the temperature the product actually experienced, because there is no basis for trusting the number the logger reported in the first place. The same gap undermines a thermal validation study built on that data: the report is only as strong as the calibration behind the instruments that generated its readings, and a validation built on an uncalibrated sensor is not a weaker version of a valid one, it is not evidence at all. This is why an auditor reviewing a monitoring program asks for calibration certificates before looking at a single temperature chart. A perfect looking chart from an unproven sensor is worth less than a messier chart from a sensor with current, traceable calibration behind it.