What question did this study set out to answer?

The aim is to clarify the scoring specification for evaluating the sufficiency of public evidence in Decision-Event Reconstructability for AI systems.

June 12, 2026Open Access

Decision-Event Reconstructability Scoring Spec v0.3: Post-Pilot Clarified Specification (GOV.UK ATRS Pilot P01-P10)

Key Points

The aim is to clarify the scoring specification for evaluating the sufficiency of public evidence in Decision-Event Reconstructability for AI systems.
Establishing an eight-field evidence-sufficiency rubric for decision-event reconstruction assessment.
Clarifying rules post-pilot review, including distinctions and scoring guidance for evidence fields.
Providing guidelines for automated and human-based assessments in scoring processes.
Specification defines eight critical fields for evidence-sufficiency evaluation.
Incorporation of distinct scoring rules improves consistency in assessing decision-events.
New guidelines enhance clarity on committed outcomes and audit pathways.

Abstract

This deposit provides the post-pilot clarified scoring specification for the Decision-Event Reconstructability (DER) pilot study on ten purposively selected GOV.UK Algorithmic Transparency Records (P01-P10). The specification defines an eight-field record-level evidence-sufficiency rubric for assessing whether public AI-enabled system records preserve enough public evidence to support decision-event reconstruction. The eight fields are: F1 decision-event identification, F2 operational context, F3 AI or algorithmic contribution, F4 human contribution, F5 data/scope boundary, F6 committed outcome, F7 review/audit pathway, and F8 reconstruction sufficiency. Version v0.3 incorporates post-pilot clarification rules developed after calibration and post-pilot review. These include the event-class versus event-instance distinction, the F8 holistic diagnostic rule, the strict threshold for F7 review/audit pathways, automated-default versus human-fallback scoring guidance, committed outcome clarification, FOI/redaction scoring guidance, and safe band interpretation. Claim ceiling:This deposit is part of a bounded method-development pilot on ten purposively selected public GOV.UK Algorithmic Transparency Records. It provides the scoring specification for reproducibility and inspection of the pilot method only. The artefact is not a validated instrument. It is not a formal verification method, legal assessment, fairness assessment, safety assessment, compliance finding, accountability proof, or general finding about all ATRS records or public AI governance.

Read Full Paperexternally

Mark Helpful

Bookmark

Relay

View Full Paper