Compliance Without Accountability: A Longitudinal Analysis of Documentation Quality in Mandatory Autonomous Vehicle Disengagement Reporting (2019–2025)

Preprint of a manuscript submitted to Policy & Internet (Wiley). Submission ID: 951ef8ca-ee0e-4e70-a96b-c42a0784f761. This paper tests the Compliance Minimisation Hypothesis: that in the absence of a measurable quality standard, organisations subject to mandatory AI incident reporting will produce documentation that satisfies the formal requirement while minimising accountability-relevant content. The California DMV autonomous vehicle disengagement corpus (2019–2025, N = 2,500 records, 20 manufacturers) is analysed using a six-field proxy scoring instrument derived from the Boundary Transparency Index (BTI; So, 2025). Documentation quality varies by a factor of three across manufacturers operating under identical reporting obligations (BTI range: 2.56–8.25/12). Corpus-wide quality shows no statistically significant improvement over seven years (slope = 0.047/year, p = .668). One manufacturer declines from BTI 8.06 to 0.59 across four years (Cohen's d = 9.12). A counter-case shows sustained improvement is possible (slope = +0.489/year across six years). Mandatory reporting, without an embedded quality standard, produces compliance but not accountability. Indexed in the project Root Index: https://doi.org/10.5281/zenodo.17992916