A
AIG Sentinel
Demo ยท Western Council
๐Ÿ‡ฆ๐Ÿ‡บ NSW LG ยท v1
โ† Back to marketing site
BM
Bob Millward
Governance Lead

UC-005_road_classification_sample_audit_2026Q2.pdf

EV-2026-0138 ยท Test Results ยท linked to Asset image classification โ€” road defects

โ† Back to vault
Artefact ID
EV-2026-0138
Type
Test Results
Linked use case
Uploaded by
Roads & Transport
Uploaded at
2026-05-19 15:00
Size
344 KB
SHA-256
8ee0โ€ฆ7651 (truncated for display)
Hash chain
Verified ยท chained to previous entry
Summary

Quarterly sample audit of road defect classifications produced by the RoadAI vendor model. 200-image random sample, manually reviewed by maintenance officer.

Chain of custody

Sample drawn randomly via vendor API. Manual classifications by senior maintenance officer. Hash chained at upload.

Key facts

Overall agreement
193 / 200 = 96.5%
False positive rate (Q2)
5.0% (target < 8%)
AIRC condition status
On track
Drift flag (mid-cycle)
FP trending up โ€” see Re-attestation page for monitoring

Sample audit results (Q2 2026, n=200 classifications)

Defect classModel classifiedOfficer confirmedDisagreement
Pothole โ€” major24231 (false positive)
Pothole โ€” minor38362 (downgraded by officer)
Cracking โ€” alligator17170
Cracking โ€” longitudinal41392 (1 FP, 1 underclassified)
Line marking โ€” degraded28271 (FP)
Line marking โ€” missing12120
Surface โ€” wear22211 (officer reclassified)
No defect (true negatives)18180