Paper with field description and evaluation procedure: https://github.com/facells/fabio-celli-publications/blob/main/docs/2026_perspective_seshat_clicit26.pdf
Code to replicate data generation: https://colab.research.google.com/drive/1_4aUNGjl7_uhLZZKE7mAYHPhWYUZ9jvr?usp=sharing
Seshat-perspective is the first historical databank synthetically annotated with a perspectivist approach by means of 3 different Large Language Models (LLMs): Deepseek (dr1), Llama (l3l) and Mistral (m3m). This dataset expands Seshat (https://seshat-db.com/) to 1023 polities and it is fully annotated. Please refer to the paper for information about data validation (https://github.com/facells/fabio-celli-publications/blob/main/docs/2026_perspective_seshat_clicit26.pdf). The code for data generation is available for replication purposes (https://colab.research.google.com/drive/1_4aUNGjl7_uhLZZKE7mAYHPhWYUZ9jvr?usp=sharing)
The schema is generated with the following prompt by each LLM:
Act as an expert historian and generate a jsonl with information about the <'polityname'> existed in <'region'> between <'start_year'> and <'end_year'>. Consider that negative numbers are years BC. The json must have the following values:
"pop": polity population: 0=around 100 people, 0.1=between 100 a…
