A good investigation begins with a statement that another person could verify: “the measured inlet flow remained above its comparable overnight pattern,” rather than “a new leak appeared.” The first describes evidence. The second proposes a cause. Keeping them separate makes it easier to discover an operating change, a faulty instrument or a genuine defect without defending an early assumption.
Check whether the record means what it appears to mean
Confirm the asset identity, units, sampling interval and timestamps. Distinguish the time a measurement was taken from the time it arrived at the server. A communications outage can produce a batch of delayed readings that resembles a sudden event on a poorly designed dashboard. Check clock drift and daylight-saving handling before comparing acoustic recordings or the sequence of pressure changes.
Inspect gaps, flat lines, clipping and unusual steps. A constant value may represent a stable process, a stuck sensor or repeated last-known data. A sudden scale change may follow a unit conversion or configuration update. Review sensor health, battery state and maintenance notes before changing an operational alarm. Preserve the raw record and document any cleaning or interpolation so the investigation remains reproducible.
Compare conditions, not just dates
A useful baseline represents a comparable operating state. Weekday demand may differ from weekend demand; industrial schedules and seasonal irrigation may change the night profile. A hospital or other continuous user can contribute legitimate consumption when most households are quiet. Minimum night flow is therefore a starting point for analysis, not a direct measurement of leakage.
Pressure matters because leakage and consumption can both respond to it. A pressure change may alter flow even without a new defect. Reservoir operation, pump schedules and pressure-reducing valve settings should be considered alongside weather and customer activity. Met Office observations can provide environmental context in the United Kingdom, but weather correlation alone does not establish the cause of a network event.
Keep baseline selection visible. If a reference period excludes unusual events, explain why. Avoid repeatedly moving the baseline until an alert disappears. A slowly developing leak can become part of an adaptive “normal” if the method is not reviewed. Conversely, a fixed historical threshold can produce persistent nuisance alerts after an authorised network reconfiguration.
Build competing explanations
For each event, list plausible causes and the evidence that would distinguish them. Increased flow might result from leakage, legitimate use, a boundary change or meter error. A pressure dip could reflect a burst, pump transition, valve movement or local demand. Persistent acoustic noise could come from escaping water, a pump or another mechanical source. A short list of alternatives keeps the next check focused.
| Observation | Possible explanation | Useful next check |
|---|---|---|
| Higher overnight flow | Leak, new demand or supply transfer | Boundary status and customer-use context |
| Brief pressure dip | Operational transition or rapid event | Time-aligned pump and valve records |
| Persistent noise | Leak or continuous mechanical source | Independent listening and nearby equipment review |
| No recent records | Power, communications or storage problem | Device health and original capture times |
Independent evidence has particular value. Flow and acoustic observations from separate instruments may support a shared hypothesis, but check whether they actually cover the same hydraulic area. Multiple software alerts derived from one channel should not be counted as multiple confirmations. An algorithmic confidence score describes its model and training assumptions; it is not automatically the probability that a physical leak exists at a specified location.
Choose a proportionate next action
Priority should consider consequence as well as apparent certainty. A moderately supported concern near a critical crossing or a vulnerable service may warrant faster review than a stronger signal with limited immediate impact. This is a decision for the responsible utility, using its incident and asset-risk procedures. ISO 31000 offers general risk-management principles, while ISO 55001 addresses asset-management systems; neither supplies a universal leak-alarm threshold.
Write a concise handover: the observation, the time window, the affected measurement points, the checks completed, remaining explanations and the proposed field test. Include uncertainty rather than hiding it in a colour-coded status. For readers in Leeds, Manchester or elsewhere, the location label alone is less useful than an accurate asset reference and a current map available to the authorised crew.
Review outcomes after the field visit. Record confirmed leakage, another identified cause and unresolved findings as distinct results. A crew finding no leak does not necessarily prove that the alert was false; conditions may have changed or the search may have been limited. Over time, this outcome record can improve thresholds and deployment choices without exaggerating detection performance.
Context for further review
- Met Office — weather context, not proof of a network cause.
- ISO — ISO 31000 and ISO 55001.
- Ofwat — water-sector reporting and performance context in England and Wales.