BRAS-9: Meta-prompts and platform for generating Brazilian synthetic survey respondents
Year: 2026
Type: other
Abstract: BRAS-9 (Brazilian Respondent Architecture for Synthetic samples) is an architecture for generating synthetic survey respondents representative of the Brazilian adult population. This record contains the two components required to reproduce it. 1. Meta-prompts (BRAS-9_ Synthetic Personas (Port).docx). Nine system prompts, one per population segment, each conditioning a generative language model to answer as the modal exemplar of that segment, carrying its documented demographics, values, language register, and internal contradictions. The prompts are written in Brazilian Portuguese and are intended to be used unedited and untranslated: the linguistic register is part of the instrument. 2. Platform source code (BRAS-9.zip). Full codebase of the web platform that operationalizes the architecture, supporting individual interviews with a single persona and comparative focus-group sessions across multiple personas. Stack: React 18, Vite, TypeScript, Tailwind CSS, Supabase. Persona responses are generated at runtime by a generative language model invoked from the chat-persona edge function. Segmentation and population weights are calibrated from Nunes (2025), based on the Quaest / Brasil no Espelho survey (n = 9,994; November-December 2023; 340 municipalities): Christian Conservative (27%), Class DE / State-dependent (23%), Agribusiness (13%), Progressives (11%), Left-wing Activist (7%), Business Owner (6%), Social Liberal (5%), Individual Entrepreneur (5%), Far Right (3%). These materials accompany a manuscript submitted to BAR - Brazilian Administration Review. Project site: https://bras-9.com/ No personal data from human respondents are contained in this record. The personas describe aggregate population segments, not identifiable individuals.
Author: Paladino, Luna
Institution: Universidade Federal do Rio de Janeiro
Language: English
Subfield: Sociology and Political Science
Field: Social Sciences
Domain: Social Sciences
Open Access status: green
OpenAlex ID: W7204131547
