NaN Is Not a Sensor Reading: Validate Finite Telemetry Before Charting
An integration calculation produces NaN or an infinite result, yet the next stage expects a normal measurement. Letting that result travel as if it were a valid reading can create rejected uploads, missing trends or misleading summaries. Define how invalid numeric results are identified and retained before serialisation, mapping and operational display.

Original conceptual illustration; not a specific product, live software screen or customer installation.
Define accepted field types and rejection records before agreeing a monitoring deployment. Start with the ProSight Software platform overview. Confirm the exact equipment, interface and project scope before selection.
Key takeaways
JSON numeric syntax does not include NaN or Infinity tokens.
Check numeric validity before encoding and again after decoding.
A real zero and an invalid calculation need different quality treatment.
1. Separate a number from a usable measurement
RFC 8259 excludes NaN and Infinity from JSON number syntax. A quoted string containing those characters is text, not a numeric reading. A parser accepting an extension does not establish that the next receiver supports it.
There is also an application question after syntax succeeds. A finite number may still be outside the permitted sensor range, carry the wrong units or come from an invalid source state. Treat numeric representation, quality, units and engineering plausibility as distinct checks.
This is an integration acceptance guide, not a claim that a particular validation function is included in ProSight Software. Confirm the available ingestion method, quality representation and handling of rejected records for the actual deployment.
2. Check the producer as well as the receiver
Inspect where a non-finite result can arise: a calculation with an unavailable operand, an invalid mathematical operation or an overflow in the selected runtime. Retain enough diagnostic context to identify the stage. Do not convert every failure to zero just to make an upload succeed.
Python’s JSON documentation provides a useful interoperability warning: its default encoder and decoder accept non-standard NaN and infinity values. The documented allow_nan and parse_constant options allow stricter handling. A successful local encode-and-decode test therefore needs to be interpreted against the actual settings.
Use the runtime’s documented finite-number check where appropriate. Keep type checking separate so that text, booleans and missing fields do not silently become measurements. Review the complete conversion path, including transformations that occur after the initial parse.
3. Build a hypothetical invalid-result case
Suppose an illustrative calculation estimates a ratio from two collected quantities. One input is unavailable, so the required ratio cannot be established. The expected application result should be an identifiable invalid or unavailable observation under the agreed schema, not a measured ratio of zero.
Create synthetic cases for a valid ratio, a genuine zero, a missing input and a non-finite intermediate result. State the expected acceptance or rejection and quality outcome for each. These cases are test design examples, not descriptions of any live ProSight Software feature.
Check a syntactically valid exponent that exceeds the receiver’s supported numeric range as well. Grammar acceptance is not proof that the decoded value remains finite. The receiver and every subsequent calculation need an explicit supported range.

Original explanatory diagram. Examples are hypothetical; verify the actual equipment, interface contract and project requirements.
4. Choose a documented failure policy
Agree whether an invalid field causes a complete record rejection, a field-level exclusion or an accepted quality-marked observation. Base that decision on the actual interface contract and operational requirement. Avoid inventing a null-field meaning that the receiving system has never documented.
Preserve raw input and rejection reasons under the project’s access and retention rules. A troubleshooting record should identify the asset, field, timestamp and failed check without exposing credentials or unnecessary customer data. Decide who investigates repeated failures.
Do not let a replacement value feed totals, averages or alarms as if it were measured. If an estimate is authorised, keep its method and estimated quality explicit. A visually continuous trend is not evidence that the underlying measurement exists.
List numeric types, supported ranges and permitted quality states.
Create valid, zero, missing and non-finite synthetic cases.
Check producer encoding and receiver conversion settings.
Verify stored quality, trends and summaries against the agreed outcomes.
5. Verify the full reporting path
Run the synthetic cases through the actual producer, transport, parser, mapping and display path in an approved test environment. Inspect stored values and quality, not just a success response from the upload interface. Check derived summaries over the same interval.
Repeat after a runtime or decoder change because permissive defaults can differ between implementations. Record software versions, relevant options, expected outcomes and observed results. Keep a small reproducible fixture for future commissioning work.
For the next ProSight Software discussion, bring a sanitised payload, the calculation context, producer settings and the intended quality behaviour. Confirm deployment-specific facilities before choosing the implementation. The useful acceptance result is a traceable distinction between valid measurement, true zero and unavailable information.
Related guide and project resources
Application selection and next steps
Recommended equipment / related solutions
For this project, review ProSight Software platform overview. Use the article’s checks as acceptance questions; the listing alone does not establish the behaviour of every connected device or integration.
For an alternative project scope, review ProSight Custom project scope. Ask ProSense to confirm the required inputs, data handling and deployment options. This is a selection starting point, not a compatibility guarantee.
Need help with your application?
Send the relevant equipment documentation, field definitions and a sanitised example of the behaviour you need to verify. Confirm the available functions and integration requirements for the exact equipment and deployment.

.png)
Comments