# T025 — Springdale Flock Zero-Misidentification Reporting The same department, the same statute, the same reporting officer, two vendors, and two irreconcilable error profiles. ## Statement A Springdale's § 12-12-1805 reports record, for its **Axon** fleet ALPR: | Period | Axon scanned | NCIC misidentifications | |---|---:|---:| | Jan–Jun 2025 | 815,795 | 932 (274 unique plates) | | Jul–Dec 2025 | 4,277,946 | 3,418 | | Jan–Jun 2026 | 4,532,629 | 2,454 | The 2025 second-half figures appear in `LPR Records July-Dec 2025.pdf`, PDF p. 1 (“Number of license plates scanned: 4,277,946”; “did not Correlate to an alert: 3418”). The 2026 first-half figures appear in `LPR Records Jan-June 2026.pdf`, PDF p. 1 (“Number of license plates scanned: 4,532,629”; “did not Correlate to an alert: 2454”). ## Statement B The same reports, prepared by the same sergeant under the same statute, record for **Flock**: | Period | Flock scanned | NCIC misidentifications | Custom-hotlist misidentifications | |---|---:|---:|---:| | Jul–Dec 2025 | 2,411,668 | **0** | **0** | | Jan–Jun 2026 | 21,429,985 | **1** | **0** | **One misidentification across roughly 23.8 million Flock reads**, against 5,872 for Axon across 8.8 million reads over the same two reporting periods. Across all three Axon periods the totals are 6,804 and 9.6 million. Custom-hotlist misidentifications are reported as zero for both vendors in every period, including the period in which Flock custom hotlists produced 3,133 confirmed matches. The Flock figures appear on PDF p. 2 of the same reports: `LPR Records July-Dec 2025.pdf` states “Number of license plates scanned: 2,411,668” and records zero NCIC and custom-hotlist matches that did not correlate; `LPR Records Jan-June 2026.pdf` states “Number of license plates scanned: 21,429,985,” records one NCIC non-correlation, and zero custom-hotlist non-correlations. ## Why it matters Read literally, Statement B asserts a plate-reading error rate near zero for one vendor and a substantial one for another in the same fleet, city, and reporting cycle. The corpus has no evidence supporting that as a hardware or software fact, and the [[ASP Statewide ALPR Data Reports 2025-2026]] cut the other way: ASP records 919 misidentifications across 61.8 million reads on a different vendor's equipment. At least two readings remain: the reported vendor difference may reflect a real system or workflow difference, or the systems may populate and count the field differently. The reports do not document the denominator, disposition workflow, deduplication rule, or extraction method needed to choose between those explanations. If a zero reflects an unpopulated field rather than reviewed outcomes, that would raise a statutory-reporting question; the present records do not establish that premise. This matters beyond Springdale. Misidentification counts are the only outcome measure in the Arkansas statutory scheme that captures the cost of the system to people who did nothing wrong ([[ALPR Error Layers and Outcome Measurement]]). If the field is vendor-dependent, statewide comparison across agencies is unsound, and the corpus's own aggregates need that caveat attached. ## Resolution status open — resolvable only by the Department stating how the Flock figure is derived, or by an audit or query-log export showing alert dispositions. Note the reporting asymmetry cuts against a simple explanation. Springdale's [[Policy 10.14 — Springdale License Plate Reader Systems]] § IV.D.2 requires an officer to document after every alert-based stop, expressly including where the "LPR Alert/Hits was not accurate." A departmental mechanism for recording exactly this therefore exists on paper for both vendors. ## Discovery Surfaced 2026-07-24 during ingest of Springdale production `26-1034`. ## Notes The counting method also changed mid-series: the January–June 2025 report gives raw and unique figures ("932 (274 unique license plates)") while the later reports give raw counts only. Period-to-period comparison within Springdale is therefore not like-for-like either, independent of the vendor question.