I’ve been building a browser-based general-purpose Influence Engine tool for mapping beliefs - or any scalar values - and seeing how changes propagate through a network. It’s meant for situations where you want to make your reasoning structure explicit, whether for legal investigation, debate analysis, or structured forecasting, but without needing to manually run Bayesian equations every time you tweak something.

The tool has two modes (and allows for active toggle):

• Bayes Lite – Start with qualitative influence strengths (“weak,” “moderate,” “strong”) and the tool gives you reasonable, but fuzzy, probability estimates. This is good for exploratory work or when precise priors aren’t available.
• Bayes Heavy – Enter explicit baseline and conditional probabilities and get rigorous updates via a Naïve Bayes. This mode assumes independence of inputs, disables structural edits while active, and forces more disciplined modeling.

Other features:

• Automatically differentiates between fact nodes (fixed) and assertion nodes (influenced) based on graph structure.
• Visual robustness indicator to show how well-supported a given node is.
• Handles bidirectional and multi-source influences with diminishing returns logic to prevent runaway amplification.
• Keeps cycles out of the network without losing the ability to model mutual alignment or antagonism.

This is not meant as an academic Bayesian network package. It’s deliberately lightweight, fast to use, and visually intuitive. I’ve been using it for investigative work, but I’m curious how it might translate to epistemic hygiene, rationalist forecasting, or even adversarial reasoning scenarios.

What I’m looking for:

• Stress-testing on reasoning problems from LessWrong readers.
• Feedback on whether the simplifying assumptions (especially in Lite mode) are reasonable or dangerously misleading.
• Thoughts on what would make it more useful for group deliberation rather than just solo reasoning.
• Here’s a demo link! I can only confirm full functionality on Chrome and Edge.