A sensor observation and the calibration evidence used to interpret it can arrive at different times. If a calibration record appears after a reading has already informed a dashboard or work order, replacing the old value in place hides what operators actually saw. This guide shows a narrower pattern: keep the raw observation, attach time-bounded calibration evidence, issue a new interpretation, and retain any earlier decision. The example is fictional and illustrates a data contract, not a validated laboratory method or customer deployment.
A Late Calibration Record: Four Events to Preserve
Suppose a fictional vibration sensor S-7 sends observation O-104. The receiving system has no approved calibration record for the observation time, so it records the raw signal and marks its first interpretation uncertain. A record later arrives claiming to cover that time. The system must verify the asset, channel, interval, approval, method, and evidence before it uses the record; a date range alone does not establish a trustworthy calibration.
| Event time | Fictional event | Durable record |
|---|---|---|
| 09:00 | S-7 reports raw signal O-104 of 1.80 V. | Store raw value and unit, device/channel IDs, observation time, receipt time, and quality state. |
| 09:05 | A dashboard creates interpretation I-1 with calibration evidence missing. | Keep I-1 as uncertain; record any decision that relied on it. |
| 13:00 | Calibration record C-17 arrives, claiming an 08:00–12:00 effective interval. | Store its effective interval and separate recorded-at time; route evidence and approval for review. |
| 13:15 | After review, the system creates interpretation I-2 from O-104 and approved C-17. | Link I-2 to O-104 and C-17, mark it as superseding I-1, and retain I-1 and the original decision trail. |
The 13:15 interpretation does not mean a 09:05 maintenance decision was automatically correct or reversible. A responsible owner must decide whether a past action needs review. If C-17 fails approval, conflicts with another record, or lacks uncertainty evidence needed for this use, the system should leave the interpretation unresolved and surface the reason.
The accompanying calibration-bench photograph is a synthetic illustration. Its plotted values, status lights, and instrument labels are fictional; they are not evidence of calibrated precision or a real Edilec laboratory result.
The Minimum Record Contract
A useful contract separates the event being measured from evidence recorded later and from the software interpretation. These fields are an implementation checklist, not a universal metrology standard. Keep the original sensor signal even if a corrected engineering value is derived later.
| Record | Fields to retain | Why they matter |
|---|---|---|
| Observation | Observation ID, device and channel IDs, raw value and unit, observation time, receipt time, original quality flags. | Identifies what the sensor sent and when the system knew it. |
| Calibration evidence | Calibration ID/version, asset and channel scope, claimed effective interval, recorded-at time, method/reference, uncertainty statement where applicable, approver and approval state. | Makes a late or contested calibration visible without silently rewriting history. |
| Interpretation | Interpretation ID/version, source observation and calibration IDs, conversion rule/version, derived value and unit, quality state, superseded interpretation ID. | Shows how a later result was obtained and which earlier view it replaced. |
| Decision | Decision ID, interpretation version seen, user or rule, action time, outcome, review status. | Preserves what a person or automation actually used before evidence changed. |
Model observation time, calibration effective time, and the time each record entered the system separately. The OGC SensorThings API offers a structured observation model; this late-record reconciliation pattern is a product-design choice layered on top of observation data. Use stable IDs and immutable versions so a reviewer can reproduce both the earlier and later interpretation.
Reconcile Without Rewriting the Measurement
- Match the calibration record to the exact device, channel, unit, and observation time; reject an interval or scope mismatch.
- Check evidence and approval separately from the claimed effective date. A received certificate is not blanket proof that every later reading is fit for use.
- Compute a new interpretation under a versioned conversion rule. Do not replace the raw observation or erase the prior interpretation.
- Compare the new interpretation with the application decision rule, and queue affected past decisions for an accountable human review where needed.
- Record who approved the reconciliation, which inputs were used, and whether the new interpretation was published, held, or rejected.
NIST's metrological traceability policy treats traceability as a property of a particular measurement result and states that traceability by itself does not guarantee fitness for purpose. Keep a measurement's uncertainty evidence distinct from a product's acceptance tolerance or action threshold. The fictional timeline provides no uncertainty calculation or safety decision.
What to Do When the Evidence Is Incomplete
| Condition | System state | Review path |
|---|---|---|
| No calibration covers the observation time | Keep the raw observation; mark derived use as unknown or invalid under the local policy. | Request missing evidence; do not invent a version from the nearest date. |
| Two approved records overlap with different coefficients | Hold automatic reinterpretation. | Resolve the conflicting scope and approval history before publishing a new value. |
| A coefficient or unit changes | Create a new interpretation version with explicit rule and unit. | Compare impacted decisions and alert owners; never silently restate a past dashboard. |
| A correction is disputed | Retain both interpretations and the original decision record. | Record the dispute, reviewer, outcome, and any follow-up action as separate events. |
Pilot the Contract on One Decision
Choose one non-safety-critical workflow with an accountable owner. Replay a normal observation, an expired calibration, a missing record, an overlapping record, and a late approved record. The review screen should show the raw signal, calibration evidence, current and prior interpretation, and any decision affected. Measure reconciliation time, unresolved exceptions, and whether a reviewer can reconstruct the earlier decision from retained records. Do not claim a lower error rate until a real, properly scoped comparison supports it.
Where This Fits in the Larger System
For the broader device and measurement model, read Sensor Calibration Data for Connected Systems. For the decisions to settle before building, use Sensor Calibration Data Decisions That Matter before the First Build. This page owns the narrower late-record correction workflow.
Frequently Asked Questions
Should a late calibration record overwrite old readings?
No. Keep the original observation and the interpretation that was available at the time. If reviewed evidence supports a correction, publish a new interpretation linked to the same observation and identify the version it supersedes.
What if two calibration records overlap?
Hold automatic reinterpretation when the records disagree about scope or coefficients. An owner must resolve which evidence applies, document the reason, and then approve any newly derived value.
Does a valid certificate prove a reading is fit for use?
No. Traceability and calibration evidence support a measurement claim, but the uncertainty and the intended decision still need review. A software status flag cannot replace that assessment.
Conclusion
Late evidence should make sensor history clearer, not cleaner-looking by erasure. Retain the raw event, the calibration record's effective and recorded times, each interpretation version, and the decisions each version supported. That gives operators a reviewable path from measurement to action while leaving scientific validity and safety judgments with the people qualified to make them.
