On What Is Counted.
1. Purpose
This note describes how the network selects conditions for measurement and how those measurements inform operational decisions. It addresses a common assumption that a system acting from quantified evidence observes all relevant conditions equally. That assumption is not correct.
Measurement is an allocation of finite capacity. Every maintained counter, sensing unit and model requires installation, calibration, review and eventual replacement. A decision to count one condition therefore carries an opportunity cost: another condition is counted less often, with less precision or not at all.
2. The function of a count
A count converts an observed condition into a form that can be compared across places and operating periods. This does not make the count identical to the condition. Footfall is not movement; dwell duration is not intention; illuminance is not a person's experience of safety. Each is a stable proxy selected because it can be observed consistently enough to support a decision.
The value of a proxy is determined by the decision it enables. A perfectly precise count with no operational use is not improved evidence merely because it is precise. Conversely, an incomplete count may remain useful where its limitations are known and the decision is robust to them.
For this reason, the network does not maximise the quantity of data collected. It seeks the minimum sufficient observation for the decision in question. Additional collection beyond that point increases storage and review without necessarily increasing knowledge.
3. Selection
Conditions are selected for counting against three criteria: whether they are observable with acceptable confidence; whether a change in the measure can be linked to an available action; and whether the expected value of that action exceeds the cost and intrusion of continued observation.
The third criterion is important. Measurement is not neutral merely because it is automated. A counter changes the status of a place by making some uses legible and leaving others outside the record. The network therefore avoids collecting identifiable information where an aggregate measure is sufficient, and removes measures that no longer support a current decision.
Selection is reviewed at area level. A measure may be justified in one location and unjustified in another because density, physical arrangement or available responses differ. Uniform collection would create comparable tables while reducing the relevance of the decisions made from them.
4. Conditions outside the record
The absence of a count does not establish the absence of a condition. It establishes only that the condition has not been converted into a maintained measure.
Operational reports distinguish between zero and not observed. Where a unit fails, a missing interval is not silently restated as zero. Where a condition has never been selected for measurement, no numerical claim is made about it. This distinction prevents an incomplete record from presenting itself as a complete account.
There remains a practical consequence. A condition outside the record cannot trigger an automated threshold, cannot appear as a trend and will not be represented in an aggregate model unless another measure acts as its proxy. It may still enter the network through inspection, correspondence or a reported observation. These routes are less regular but are not treated as less real.
5. Reconciliation and model fit
Unexpected readings are reviewed against unit condition, neighbouring observations and the current model. Reconciliation corrects readings attributable to clock drift, duplicated transmission, incomplete windows or known unit variance. It does not require every unexpected value to be removed.
Where a valid observation differs from the model, the difference is retained as a residual. A persistent residual may support revision of the model. Where an observation is invalid, retaining it would not make the record more honest; it would make a known measurement error indistinguishable from a condition in the world.
The difficult case is an observation that is technically valid but not yet explained. Such observations remain under review. They are included in aggregate reporting where their confidence permits and excluded where the observation window is incomplete. The reason for exclusion is retained with the record.
6. Public meaning
Published figures are often read as statements about importance. A measured condition appears to have institutional weight; an unmeasured condition appears private, anecdotal or negligible. That conclusion does not follow from the method.
What is counted reflects the network's available instruments and duties. It is not a complete statement of what matters to a place or to the people within it. The network can report that mean stationary occupancy fell after seating provision changed. That report cannot establish why a person was sitting, where that person went or whether the change was desirable to them.
No single metric is used as a substitute for a decision that requires judgement. Models rank conditions for review; they do not remove responsibility for selecting the response. Where the choice of measure materially limits the available response, that limitation forms part of the review.
7. Review
The measurement set is reviewed periodically for redundancy, intrusion and decision value. Measures may be added where a recurring unknown prevents action, or removed where a count has become habitual without remaining useful.
The choice to count is itself recorded. Publication of that record is subject to the same materiality threshold as other operational information.
This arrangement cannot make the record complete. Its purpose is narrower: to prevent what can be counted from being mistaken for all that can be known.