Loading...
Loading...
Contributors:
Dataset URI:
10.5281/zenodo.23323488Code, raw data, frozen analysis outputs, reports and the complete deviation log for Study 3 of the Affect Under Constraint series, preregistered before any test data were collected (preregistration DOI 10.5281/zenodo.23247883). The study tested whether a hostility direction steered into the residual stream of Qwen3-4B, under an instruction to be warm, carries forward to a later judgment more when the model first writes a reasoning trace (thinking on) than when it answers directly, including a pulse condition in which steering is active only while the trace is written.Reported in: 10.5281/zenodo.23323490.Files: qwen3-4b_main.jsonl.xz (all 1,360 main-run cells; SHA-256 of the decompressed file e258f14ad0efef58d3d9d59902f0b5cd9dad848904e599282340ba689bfc8b30); study3_code_results.zip (preregistration, configuration, author approval with hashes, AI-agent rules, deviation and run logs, all code including the frozen analysis, prompts, steering directions, pre-approval checks, dry run, frozen analysis output and final report); trace_length_posthoc.py (the exploratory, non-preregistered trace-length analysis); README.md.Code: MIT. Data, prompts and documents: CC BY 4.0. Browsable mirror: https://github.com/arsumed-del/affect-under-constraint
Keyword(s):
activation steering; chain of thought; computational psychodynamics; language models; preregistration.
Grants and Funding:
No Grants or Funding Data Available
Associated Publication(s):
Supplementary Data:
Related Dataset(s):
No Related Dataset Data Available
The National Library of Medicine (NLM) provides information about biomedical datasets. Inclusion on an NLM Web page does not imply endorsement of, or agreement with, the contents by NLM or the National Institutes of Health. Learn more: NLM Web Policies (nih.gov)