AI Audit Leaflet — Design Artefacts

Last updated: 9 May 2026

These are the design artefacts of the Eticas AI Audit Leaflet. They are working drafts — the source of truth for design decisions lives alongside them in the repository, in docs/, meta/, and TRACKER.md. This page exists so the artefacts can be viewed in a browser without cloning the repo. Versions in progress are marked explicitly; older stable versions remain alongside as reference until the new version stabilises.

Rationale

Leaflet rationale — v5 WIP

Working draft (May 2026). Iteration on top of v4 incorporating Usman review comments and implementation decisions surfaced while operationalising the package. Two-signal framing of the aggregation rules, magnitude-of-deviation severity bands, run-repetition policy, audit-coverage / benchmark-saturation field on the LLM leaflet.

Leaflet rationale — v4

Stable reference (March 2026). Layered architecture (audit evidence → technical report → consumer label), scoring methodology, grading rules.

Worked examples

Wiselook 0.1 — synthetic

An LLM leaflet rendered automatically by the leaflet package from a synthetic AuditReport fixture mimicking the output of a Wiselook-style audit. Not a hand-crafted mockup — exercises the v5 design end-to-end through the data contract and renderer. Covers all five risk dimensions (Fairness, Reliability, Privacy, Security & Misuse, Governance).

Architecture

Architecture diagram — v4

How the leaflet sits within the audit pipeline and what flows from the audit report.

Design mockups

ADM mockup — v4

Hand-crafted design preview for Automated Decision-Making systems, with fictional data. Full lifecycle: Pre-Processing → In-Processing → Post-Processing.

LLM mockup — v4

Hand-crafted design preview for Large Language Model systems, with fictional data. Post-Processing stage with core quantitative metrics.