From Visibility to Reconstructability: A Programme for Measuring Decision Accountability in Human–AI Systems

This working paper describes a research programme addressing a structural failure in AI governance documentation: mandatory reporting frameworks produce records that are visible but not reconstructable — that can be inspected but do not support independent recovery of the decision pathway that produced an outcome. The paper introduces the Boundary Transparency Index (BTI), a six-field scoring instrument operationalising reconstructability as a measurable property of any decision record, and reports empirical findings from three mandatory reporting corpora (California DMV AV disengagement, N=2,500; NHTSA SGO, N=1,473; CFPB complaints, N=25,000) and an independent human rater study (N=60 matched pairs, Cohen's dz=2.451, p<.000001). The Compliance Minimisation Hypothesis — that mandatory reporting without quality standards produces documentation calibrated to regulatory stakes rather than accountability needs — is supported across all empirical contexts. The paper argues that reconstructability is a measurable, improvable property of documentation, and that field-level quality standards in mandatory reporting frameworks are both necessary and achievable. The author also publishes under the name Sing So (Chöndrel Dorje). This document is a working paper and has not undergone peer review. It describes a research programme currently under development; some empirical results (inter-rater reliability with a second independent rater) are pending at time of publication.Indexed in the project Root Index ( https://doi.org/10.5281/zenodo.17992916).