An incident database can contain many mentions of drones without being able to answer a precise question about drone threats. Search retrieves text; classification determines what the text means. Confusing those steps can turn an administrative workload estimate into an alarming but unsupported statistic about activity around critical infrastructure.
West Yorkshire Police's October 2026 response to FOI 3215644/26 illustrates the distinction. A request concerned confirmed or suspected drone activity affecting specified infrastructure between September 1, 2025 and August 31, 2026. The force identified 3,092 incident logs containing DRONE or UAV in any context and estimated that minimum manual review would exceed 103 hours. It declined under the cost-limit provision and said it could not establish whether the requested information was held. The page gives a month, not a verified day of publication. Read the primary source account.
A keyword match is a candidate record
The request concerned a narrow set of locations and circumstances. A log mentioning a drone might fall within that scope, concern an authorized operation or refer to an unrelated situation. The response does not classify the returned logs into those groups. Its total therefore cannot be reported as the number of confirmed incursions, crimes or disrupted operations at the specified sites.
The same boundary prevents the opposite conclusion. An inability to establish the requested information within the stated workload does not mean no relevant incidents occurred. Nor does the refusal establish that the force lacks every underlying record. It describes the cost of determining which records meet the request, leaving the requested count unresolved.
The estimate assumes two minutes per log for an initial manual review. The force says further work would be needed to identify additional details such as status, response and outcomes. Those are workload assumptions attached to a search, not measured durations for every actual investigation. The stated financial threshold is the limit cited in the refusal, rather than an invoice or a new cost charged to a drone programme.
Information architecture determines what can be answered
The case shows a distinction between retaining narratives and retaining queryable categories. Free text can preserve important context while making a later cross-cutting request expensive. A structured label can make retrieval easier, but only if it is consistently defined and applied. Neither format automatically establishes whether a report was confirmed or what ultimately happened.
For an organisation introducing additional sensors or automated alerts, that distinction has practical consequences. A new source can increase the volume of observations without improving the ability to distinguish a sighting, a verified aircraft and a consequential incident. The relevant analytical question is whether the resulting records support the decisions and public explanations the organisation expects to make. The FOI response does not assess any particular detector.
Status can also change. An initially suspicious observation may later be explained, while an incomplete report may become relevant after further information arrives. A record designed only around the first label can obscure that development. Retaining the connection between the initial report and its eventual assessment would make later counts more interpretable. This is an implication for record design, not a claim that West Yorkshire uses or omits a specific field.
Comparable totals require comparable definitions
The force explicitly cautions that police organisations do not use identical systems or capture procedures and that its response should not be compared directly with other forces' returns. A league table of keyword totals would erase those differences. Even equal counts could represent different mixtures of public reports, operational references and verified events.
A defensible comparison would first align the period, qualifying sites, event definitions and treatment of duplicates or linked records. It would also state which uncertainties remain unresolved. That does not mean every comparison is impossible; it means the response's unclassified search count cannot supply one on its own. A common word in several databases is not a common measurement method.
The applicant sought records without personal identifying information and asked for an existing electronic format where feasible. The refusal leaves that requested incident-level dataset unavailable. Public readers should therefore avoid reconstructing supposed locations, motives or suspects from the aggregate search total. None of those conclusions follows from the administrative figure published by the force.
The immediate significance is an official account of a retrieval limitation, with its assumptions visible. Its wider relevance to security autonomy is that detection and accountability both depend on usable classifications and traceable decisions. More observations do not automatically produce better statistics. The next informative disclosure would distinguish candidate reports from verified qualifying incidents under clear definitions, while preserving the unresolved status of records that have not been assessed.
Source organizations: West Yorkshire Police.
References
Questions, news tips or corrections? Contact [email protected].
