Telemetry-verified closure · recurrence

Complete is not the same as fixed.

A technician tapping Complete is a claim about the work. Telemetry returning to baseline is evidence about the equipment.

Loop 2E · rack inlet°C · 48 min
28262422Baseline bandWork reported complete03:1404:02
Closure conditions3 of 4
Δp holds ≥ 0.7 bar34 min elapsed · window satisfied
Rack inlet back in band14 of 14 racks within baseline band
Engineer confirms causeStrainer blocked upstream of CDU-2E
No recurrence in 24 hWindow still open · 6 h 12 m remaining
Incident stays open.Three of four conditions met is not recovery. The work order is complete; the incident is not.
01Complete vs fixed

Two different statements, routinely treated as one.

Almost every maintenance system closes an incident when a person marks the work done. That is a report about an activity, and it is usually accurate.

It is not the same statement as the physical condition having cleared. The gap between them is where recurring faults live: the loop that recovered for four hours, the valve that was replaced while the actual restriction sat upstream, the sensor that was reading wrong the whole time.

02Defined up front

Four things have to be defined before the work starts.

Defined afterwards, these become an argument about whether the job was done. Defined beforehand, they are a test the repair either passes or fails.

Pre-work baseline
What temperature, pressure and flow looked like before the condition began.
Expected recovery signal
Which specific measurements must move, in which direction, for this repair.
Observation window
How long the recovery must hold before it counts as recovery.
Recurrence rule
What reappearance, within what period, reopens the same incident rather than starting a new one.
03Two sources

Neither the sensor nor the technician is treated as infallible.

Telemetry can be wrong: a drifting sensor, a failed transmitter, a point mapped to the wrong asset. The technician standing at the unit is often the only one who knows.

The technician can also be wrong, or can be reporting a symptom that resolved for reasons unrelated to the work. Requiring both, and recording where they disagree, is the point — not automating one away.

04Disagreement

Disagreement is a defined path, not an exception.

Design-partner scope
Telemetry still abnormal
Work reported complete, measurements have not returned. Incident stays open and escalates.
Sensor untrusted
Technician findings contradict the point. The instrument becomes the next work item.
Temporary recovery
Condition cleared, then returned inside the recurrence window. Reopens as the same incident, with the history attached.
Conflicting field finding
What was found does not match what was inferred. The inference is recorded as wrong — that is the useful part.
05Learning

The outcome changes the next decision.

Vision

Over time the record of what was inferred, what was found and what actually resolved is the most valuable asset in the system. It is what lets the next interpretation of the same signal pattern be better than the last one.

We are explicit that this is direction, not a shipped capability. A system that claims to learn before it has closed a meaningful number of real incidents is claiming something it cannot have.

06Answered

Why not close the incident when the technician says it is done?

Because the two statements answer different questions. A completed work order says the activity happened; recovery says the physical condition cleared and stayed cleared. Most of the time they agree. The value is entirely in the cases where they do not, because those are the faults that come back.

What if the sensor is the thing that is wrong?

Then the technician's findings win, and the instrument becomes the next work item. Telemetry is treated as evidence, not as truth. A verification model that cannot conclude the sensor was wrong is not a verification model.

How long does the observation window need to be?

It depends on the equipment and the failure mode, so it is agreed per intervention type rather than set globally. The requirement is that it is defined before the work starts. A window chosen after the fact is not evidence.