Numbers-First Buyer’s Framework · Procurement Perspective · Published: October 9, 2026
Two sensor datasheets land on the evaluator’s desk. Both claim high accuracy, wide temperature range, fast response, and tough construction. On paper they look like twins; in the plant they behave like strangers. This rarely means a manufacturer printed something false. It means the same parameter can be defined in different ways, with the reference conditions buried in the small print, and comparing datasheets without checking those definitions means comparing figures that do not mean the same thing [1]. The cure is a fixed comparison grid: eight fields, each with a defined basis of comparison and a named trap. Run two candidate models through the grid and the “similar” datasheets separate quickly into one instrument and one brochure.
Table of Contents
Key Takeaways
- Accuracy is the classic trap. A figure quoted as a percentage of reading and one quoted as a percentage of full scale are different promises — at low concentrations they can differ by an order of magnitude, so force both vendors onto the same basis [1].
- Response time needs a definition attached. T90 is the time to reach 90% of the final value after a step change, T63 the time to 63%. A bare number in milliseconds without the definition is decoration, not data [2].
- Drift is only comparable when the period and the reference conditions are stated with it. Definitions anchored to a stated period under constant reference conditions are what make a per-year drift figure auditable [3].
- IP68 does not mean one thing. The code certifies dust-tightness plus protection against continuous immersion under conditions the manufacturer specifies — so ask for the depth and duration behind the rating [4].
- Temperature compensation has its own limits worth reading line by line. A laboratory-grade ATC probe, for example, covers 0–100 °C with a relative accuracy of ±1 °C — range and compensation accuracy are separate claims [5].
- Output and protocol now decide integration cost. Digital transmitters speaking Modbus RTU/TCP feed historians and analytics directly, which removes analog-signal degradation over long cable runs [6].
Field 1 — Range and Detection Limit
The range line looks self-explanatory until two models meet: one quotes a measurement range, another a display range, a third a configurable span. The comparison basis that matters is the span over which the accuracy figure is actually guaranteed, and where the useful reading ends — a sensor that measures to the top of its display range with no accuracy promise up there is quieter than it looks. Detection limit carries the same discipline in reverse: the lowest concentration the sensor can resolve with confidence, not the lowest number on the graph. Compare both models at your actual operating point, not at their respective best cases, and note where each datasheet switches from “measures” to “indicates.”
Field 2 — Accuracy and Repeatability
This is the field where identical-looking numbers hide the largest differences. The most important distinction in any datasheet is the reference value behind the percentage: accuracy expressed as a percentage of reading shrinks with the signal, while accuracy as a percentage of full scale stays constant — so at the low end of the span, a full-scale figure can be many times worse than a reading-based figure for the same printed percentage [1]. Repeatability is the quieter sibling: the spread of readings on the same unchanged sample, which bounds what calibration can ever fix. And watch the qualifiers. A typical accuracy is a statistical courtesy; a guaranteed accuracy is a contract. The comparison grid should record, for each model: basis (reading or full scale), qualifier (typical or guaranteed), and the reference conditions under which the figure was verified.
Field 3 — Drift and Stability
Drift is the change of indication over a stated period, under reference conditions that remain constant and without adjustment — the international definitions are explicit that both the period and the conditions are part of the claim [3]. A datasheet that says “excellent long-term stability” without a period has said nothing; one that says “≤ x% per month, at reference conditions” has made a promise you can test at the next calibration. The trap here is unspoken cleaning: sensors in real water foul, and a drift figure measured on a clean bench says little about a fouled month in service. Compare drift figures only when both vendors state the period, the conditions, and whether the figure assumes a maintained sensor.
Field 4 — Response Time
Response time is usually defined through two attributes, T63 and T90 — how long the reading takes to reach 63% and 90% of the final value after an instant change in the measured variable [2]. The reason 63% appears at all is mathematics: it corresponds to one time constant of the exponential approach, which makes it the cleaner figure for modeling. For procurement the question is not which number is bigger but which system the number must serve. A sensor with a slow T90 sampled by a fast logger simply stamps a stale value onto a live record — the reading reaches the database before the sensor reaches the water. Compare T90 against your control loop’s sampling interval and alarm latency, and treat any datasheet that omits the T63/T90 definition as an incomplete specification, not a fast one.
Field 5 — Temperature Range and Compensation
Two separate lines hide in this field, and both must be compared separately: the temperature range the sensor survives, and the range over which compensation holds the measurement honest. A worked example makes the distinction concrete — a laboratory-grade automatic temperature compensation probe specifies a 0–100 °C temperature range with a relative accuracy of ±1 °C [5]. The range says nothing about accuracy, and the accuracy says nothing about chemical compatibility. For process instruments the trap is the word “compensated”: compensation corrects the sensor’s own temperature-dependent behavior, it does not convert a hot sample into a 25 °C answer, and beyond the compensated range the printed accuracy simply expires. Ask each vendor for the compensated range, the compensation method (fixed coefficient or algorithmic), and the accuracy that survives at the ends of that range.
Field 6 — Output and Protocol
The output line decides what the sensor costs to own, because integration is where budgets quietly leak. Analog 4–20 mA remains universal but carries one value per cable pair, with signal quality exposed to cable length and electrical noise; digital transmitters speaking Modbus RTU/TCP deliver synchronized multi-parameter frames to SCADA and cloud platforms directly, which is why Shanghai ChiMay builds its online analyzers around digital output as a primary interface — sensors that feed your AI water model, not just your dashboard. When comparing models, record the protocol, the register map availability, and whether diagnostic values (temperature, electrode state) ride along with the measurement. A sensor whose protocol documentation requires a phone call is telling you what its integration will cost.
Field 7 — Materials and Ingress Protection
Wetted materials decide whether the sensor survives the water; ingress protection decides whether the electronics survive the plant. On protection, read the code precisely: under the IEC 60529 framework, the first digit 6 means dust-tight and the second digit 8 means protected against continuous immersion in water — under conditions specified by the manufacturer, which is why two IP68 housings are not automatically equivalent [4]. The trap is the unspecified condition: a rating tested to one meter for thirty minutes and one rated for permanent submersion at depth both print “IP68.” For submersible installations, demand the depth and duration behind the digit; for splash-zone installations, verify the cable gland sealing separately, since that is where field ingress actually enters. Wetted materials — glass, polymer, stainless, titanium — belong in the same row of your grid, with your sample chemistry written beside them.
Field 8 — Calibration and Ownership Cost
The last field is the one vendors omit and auditors request first: what does it take to keep the first seven fields true? Compare the calibration interval, the required standards and their shelf lives, the consumed parts, and the documentation each model generates — calibration certificates, compensation records, and diagnostic logs. Audit-ready documentation is a design feature, not a courtesy; a sensor that logs its own temperature compensation and calibration history shortens every future data dispute. And price the interval, not just the probe: a cheap sensor on a monthly calibration contract is not a cheap sensor. Buy the sensor, own the outcome — the five-year arithmetic of standards, labor, and downtime belongs in the comparison grid beside the sticker, which is the accounting logic behind Shanghai ChiMay’s practice of quoting five-year cost alongside unit price.
The Eight Fields on One Page
| Field | Compare on | The trap to defuse |
|---|---|---|
| Range & detection limit | Span over which accuracy is guaranteed; resolution floor at your operating point | Display range quoted where a guaranteed range belongs |
| Accuracy & repeatability | Basis (% of reading vs % of full scale); typical vs guaranteed; reference conditions [1] | Identical percentages on different bases [1] |
| Drift & stability | Stated period + stated reference conditions, per unit time [3] | “Excellent stability” with no period or conditions |
| Response time | T63 and T90 with definitions, vs your sampling interval [2] | A bare time value with no definition attached [2] |
| Temperature range & compensation | Survivable range vs compensated range vs accuracy at range ends [5] | “Compensated” read as “accurate everywhere” [5] |
| Output & protocol | Protocol, register map, diagnostics; digital vs analog integration cost | One-line “4–20 mA / RS485” hiding missing documentation |
| Materials & ingress protection | Wetted materials vs sample chemistry; IP code with stated depth/duration [4] | Two different “IP68” conditions compared as equal [4] |
| Calibration & ownership cost | Interval, standards, consumables, records; 5-year arithmetic | Sticker price compared while calibration burden is ignored |
How to Run the Comparison
Put the two models in one spreadsheet, one row per field, and fill each cell with three things: the figure, its qualifier, and its stated conditions. Where a datasheet is silent, write “not stated” — silence is data too, and it is negotiable before the purchase order and expensive after it. Normalize every percentage to the same basis, every time figure to the same definition, and every environmental claim to the same reference conditions [1][3]. What survives that normalization is the real difference between the models, and it is usually smaller than the brochures imply and more decisive than the price gap suggests.
References
- smartGAS — Selecting a Gas Sensor: Why Two Datasheets Are Not Comparable (accuracy as % of reading vs % of full scale; reference conditions hidden in the small print): https://www.smartgas.eu/en/news/single-view/detail/selecting-a-gas-sensor-why-two-datasheets-are-not-comparable
- Vaisala — Response Time in Humidity Measurement, Technical Note B211803EN (T63 and T90 definitions): https://www.vaisala.com/sites/default/files/documents/Response-time-in-humidity-measurement-Technical-Note-B211803EN.pdf
- IEC 60746-1:2003 — Expressions of Performance of Electrochemical Analyzers, Part 1 (drift defined over a stated period under constant reference conditions; output fluctuation): https://cdn.standards.iteh.ai/samples/5278/84e4749c247645afb1c0351733b6faf2/IEC-60746-1-2003.pdf
- Stork Solutions — What Does IP Rating Mean? Complete Guide (IP68 = dust-tight plus continuous immersion under manufacturer-specified conditions; IEC 60529 testing): https://stork.solutions/?p=12640
- Fisher Scientific (Thermo Scientific Orion) — Stainless-Steel ATC Probes (temperature range 0–100 °C, relative accuracy ±1 °C as a published specification example): https://www.fishersci.pt/shop/products/orion-stainless-steel-automatic-temperature-compensation-atc-probe-8-pin-din-connector/10009140
- Shanghai ChiMay — Product Catalog: Softener Valve, Filter Valve, Online Water Quality Analyzer & Smart Sensors (company product range): http://chimaytech.com/
About the author: Written by the Shanghai ChiMay Technical Editorial Team — instrumentation specialists in online water quality analysis, sensor specification and process integration for industrial, municipal and environmental monitoring applications.
