# Forkwise

> Forkwise is a free web tool that tests how robust a data finding is. It sends one dataset and one research question to many AI analysts, lets each analyst choose its own method, runs every analysis in Python, and reports whether the answer is robust, fragile or choice dependent. It was built as a student project for the A-ISPC 2026 competition.

## Key facts

- Method: an AI-driven multiverse analysis, also called a specification curve analysis or a "many analysts" study.
- AI models: Llama 3.3 70B, Llama 3.1 8B and Mistral Small 3.1, each with five analyst personalities.
- The AI only chooses the analysis plan (control variables, outlier rule, missing value rule). It never calculates results.
- Python (pandas and statsmodels) runs an OLS regression for every analyst. The effect is the change in Y for one standard deviation of X.
- Verdict rules: robust if at least 90% of analysts agree on the direction and at least 80% find p < 0.05; fragile if at least 85% agree and 30% to 80% find p < 0.05; otherwise choice dependent.
- Safety checks: columns that do not exist, the outcome itself, and controls whose name contains the outcome are removed before any calculation.
- Inspiration: Silberzahn et al. (2018), where 29 research teams analysed the same data and reached different conclusions.

## Pages

- [Home](https://forkwise.online/): what the tool does and an example result
- [How it works](https://forkwise.online/how-it-works): the six steps of the method and its limits
- [About](https://forkwise.online/about): the problem, the solution and the findings
- [FAQ](https://forkwise.online/faq): common questions about results and data
- [Privacy](https://forkwise.online/privacy): what data is stored and what the AI sees
- [Contact](https://forkwise.online/contact): reach the team