Q125 - What_is_a_governance_artefact_or_evidence_object_in_RAIDT
Q125 — What is a governance artefact or evidence object in RAIDT?
← RAIDT · Star S3 - Run-Level Evidence Logic · primary item: S3.02 · Evidence object
Appears in sources
integrated_82#Q2.14
Answer
In RAIDT, a governance artefact is a designed information object that makes a governance claim reviewable in practice. The papers are explicit that RAIDT's artefacts are not software components; they are governance artefacts: a run-level evidence pack and a score profile. The evidence object is the bounded part of this pair. It is the auditable record for one configured use, preserved so that a third party can inspect whether an organisational claim about responsible use is warranted.
This distinction matters because RAIDT separates artefacts by role. A model card describes a model in general, a programme audit reviews broader control arrangements, and an audit trail preserves linked traces. By contrast, the RAIDT evidence object binds the minimum records required for one governed use into a single bounded evidentiary unit. The score profile then interprets that unit across the five pillars (Responsibility, Auditability, Interpretability, Dependability, Traceability), keeping trade-offs visible rather than hidden inside narrative assurance. Because the run is the unit of governance, the artefact must preserve configuration and oversight at that level: prompts, adapters, retrieval snapshots, outputs, hashes, reviewer actions, and links to incident or risk processes. RAIDT therefore turns governance from abstract declaration into inspectable proof, and it makes influence methods as governance interventions governable rather than ad hoc.
Practical example
Consider an HR shortlist justification workflow in which GenAI drafts reasons for why certain applicants were advanced. A governance artefact in RAIDT would include the versioned prompt template, the scoring criteria supplied to the system, the model and adapter IDs if PEFT/LoRA was used, the generated justification, and the reviewer's approval or correction. Those records form the evidence object for that one run.
If a candidate later challenges the decision, the organisation does not have to rely on a general hiring policy or a model description alone. It can inspect the bounded record for the contested run, check whether the justification matched approved criteria, and see where traceability or responsibility weakened. That is the practical function of the governance artefact.
Sources in RAIDT papers
13-RAIDT-Evidence-Review_M_v1016-RAIDT-Audit-Accountability_M_v05