Skip to content

Latest commit

 

History

History

README.md

Replication: the DYNAMICS-WEB conversion result

This kit reproduces the headline behavioural result of the paper, that aligning the sub perceptual character of a page to a reader raises conversion from 19 to 31 per cent (one sided p = 0.0295), from the released data and the pre registered analysis.

Pre registration

The analysis below was fixed in writing before data collection.

Reproduce it

pip install pandas statsmodels
python3 conversion_analysis.py stage1_data.csv

Expected: the sub perceptual alignment term (align_subtle) is positive, odds ratio about 1.35, one sided p = 0.0295; broad alignment (align_full) is null; the loud commercial alignment (align_loud) is null. That is the result reported in the paper.

Files

  • stage1_data.csv : 936 trials, 117 participants, 8 offers. Participant identifiers are anonymised to P001..P117. Columns: participant, offer, run, matched, choice, quality, reading_time, align_full, align_subtle, align_loud, disposition_axis, noticed_fact_diff, passed_attention.
  • conversion_analysis.py : the pre registered analysis, a logistic GEE clustered by participant with an offer fixed effect.
  • PREREGISTRATION.md, PREREGISTRATION_STAGE1.md : the pre registered protocol and hypotheses.
  • stimuli_offer1.json, stimuli_offer1_sharp.json : a sample offer's stimuli, the eight axis variants with the factual core held constant, to show how alignment is constructed.

See the measurement live

This kit verifies the behavioural result. To see the underlying character measurement on any page yourself, the toolbar scores it on your own device: https://toolbar.kronaxis.com

What is not here, and why

The data collection harness, the character scoring model, and the proprietary mapping from a reader's disposition to the content that suits them are deliberately not included. The first two depend on internal infrastructure; the last is withheld as described in the paper. The released data and the fixed analysis are sufficient to verify the reported statistical result, which is the purpose of this kit.

Licence: the analysis code (conversion_analysis.py) is released under the MIT Licence (see LICENSE in this folder); the data files and written materials are released under CC BY 4.0. Copyright 2026 Kronaxis Limited.

The six character axes and the scoring rubric used in the experiment are documented in AXES_AND_RUBRIC.md.