The Transparency–Reconstruction Gap: Why AI Governance Records Cannot Yet Answer for Decisions

This phenomenon statement introduces the Transparency–Reconstruction Gap: the gap between AI governance records that make systems visible and records that preserve enough decision-event content to reconstruct how consequential decisions became possible, actionable, and accountable. It distinguishes visibility from reconstructability and positions decision-event content as a requirement for accountability, contestation, and post-event review across domains including autonomous vehicle reporting, public-sector algorithmic transparency, defence AI, policing, shelter and welfare allocation, and institutional response records. The statement argues that the problem is not simply missing data. In some cases, the failure is built into the schema itself: records may disclose systems, purposes, datasets, risks, or outcomes while failing to preserve who acted, when, on what basis, under what uncertainty, and through what oversight. The document is intended as a programme-level phenomenon statement and external entry point for future work on decision-event reconstructability, record infrastructures, and AI governance accountability.