FLaME - Anonymized Research Evidence Package

FLaME research evidence package This package accompanies Collaborative Learning Across Data Boundaries: Designing for Heterogeneous Participation in Industrial AI. It documents the industry-grounded Design Science Research study behind FLaME: problem and requirements elicitation, five Attribute-Driven Design (ADD) iterations, 16 Architecture Decision Records (ADRs), and an adapted Architecture Tradeoff Analysis Method (ATAM) evaluation of the resulting C4 container-level architecture. The files support inspection of the study’s evidence and reconstruction of its requirements, design rationale and architecture-level evaluation. They do not constitute an executable implementation or demonstrate operational performance, prediction accuracy or achieved privacy guarantees. Contents Folder Contents Workshops A.1–A.3 existing-system and ML analysis, C.1 user journeys, D.1 Mini-QAW, and E.1 ATAM. Users Two proto-personas and the two user story maps. Quality The quality-attribute web, 81 refined scenarios with votes, and 13 selected quality scenarios alongside 6 quality attribute categories. Requirements ASR derivation workbooks and the consolidated set of 19 selected ASRs. Design Five ADD iterations with architecture diagrams and 16 linked ADRs. Reading the evidence The workshop records describe the activities and identify their inputs and outputs. The source-specific requirement workbooks link observations to candidate ASRs; Consolidated_FLaME_ASRs.ods links those candidates to the selected requirements. For open-question sources, O2 refers to the register in A.2 and O3 to the register in A.1. ADD_Iterations.md shows how the ASRs and quality scenarios informed the architecture. Architecture_Decisions.md records the resulting choices. E.1 evaluates the final container view against the 13 selected scenarios and consolidates the findings into four risk themes. These records support Sections 3–6 of the accompanying paper. Scope and confidentiality We use pseudonyms for people and organizations. Detailed views of the partner’s running service, source code, access paths and implementation-specific model information are withheld for confidentiality and operational security. The public A.1–A.3 artifacts retain the functional descriptions and source IDs used in ASR derivation; detailed source mappings are held separately. Additionally, the proto-personas express stakeholder-informed customer assumptions. The architecture and ATAM findings concern structural support and recorded assumptions at C4 container level, rather than measured system performance or achieved privacy guarantees. File formats PDF files can be opened with any compatible PDF viewer, while ODS spreadsheets require an OpenDocument-compatible spreadsheet application. Markdown files provide explanations and relative links to the evidence.