Published September 2, 2026 | Version 1.0-t1
Software Open

Silicon Sample Benchmark — Tier 1 (individual simulation) submission (team team_11)

Authors/Creators

  • 1. Harvard University
  • 2. Transluce
  • 3. Stanford University

Description

Per-respondent simulation using the per-item independent-sampling method of Hewitt, Ashokkumar, Ghezae & Willer (Nature 2026; treatmenteffect.app). Each synthetic respondent is a real TESS-respondent demographic profile rendered through the paper's randomized prompt-template bank; every scored item is answered with a single temperature-1 draw by one of five LLMs (respondents rotate round-robin across GPT-5.6-luna, Gemma-3-27B, DeepSeek-V3-0324, Qwen3.5-27B, GPT-OSS-120B).

Notes

Team team_11. Disclosure class A.

Files

insperatum/silicon-sample-team11-v1.0.zip

Files (21.7 MB)

Name Size
md5:e21f503ded49ba57f70b56d6b31b5cb9
21.7 MB Preview Download

Additional details

Related works

Is compiled by
Software: https://github.com/insperatum/silicon-sample-team11 (URL)
Is part of
Other: https://janpfander.github.io/llm_predictions_megastudy/ (URL)
References
Preprint: https://janpfander.github.io/llm_predictions_megastudy/amendment_preregistration.html (URL)

Software

Repository URL
https://github.com/insperatum/silicon-sample-team11