What this is

3Dogs Nexus is a working product, not a concept. It takes a question, puts it in front of a panel of independent AI models, and makes them argue before anyone gets an answer. The confidence number it reports comes from how much disagreement was still standing at the end. It is not a model telling you it feels sure.

What I want to find out

Whether the panel is actually better calibrated than one model, and to publish the answer either way. The work: assemble 50 to 100 high-stakes questions where the real outcome is known or can be adjudicated. Run each through the panel, and through a single model as a control. Score both with Brier and expected calibration error. Publish the corpus, the method, and the numbers.

Where the money goes

Compute for both arms of the test, plus paid time to label and adjudicate the corpus. That is the whole budget.

Who I am

Alan Finney, founder. 35 years working, five of them selling enterprise software at IBM. I built 3Dogs myself and paid for it myself. The platform is running today at app.3dogs.ai. I have already published one blind head-to-head against a leading single-model assistant.

How this could fail

Two ways it comes up empty. The corpus is too small, or the adjudicators disagree too much to read anything from it. Or the panel just is not better calibrated than one model. I will publish either outcome. A null result here is still worth knowing.

Money raised so far

None. Self-funded, no outside capital. I have applications in with several non-dilutive grant and credit programs. Nothing has paid out yet.

What comes out of this, funded or not

The corpus, the scoring code and the results all get published openly, including the negative result if that is what we get. 3Dogs Pointer, the other half of the platform, measures how AI systems actually find, read and describe an organization and reports what is verifiable rather than what merely exists. It is free to use and it stays free for any nonprofit or safety org that wants it. That is not a reward for funding this project, it is already free today. I mention it because the same measurement discipline is what this corpus is meant to test.