About Conclavik

Conclavik is a multi-model adversarial engine for institutional decisions. A user submits a thesis, memo, or board paper (an acquisition case, a portfolio position, a capital deployment), and four independent frontier models analyse it separately, then challenge one another through structured phases: roundtable, cross-examination, steelman, objection. The output is a single PDF with scenario probabilities, load-bearing assumptions, a disagreement map showing where conviction is real and where it splits, and monitoring triggers that flip the case. Eight to twelve minutes, built to be read inside a deal team or an investment committee.

The design principle is independence. One model asked for “four perspectives” is one set of weights playing four roles. Four frontier models from four different labs (Anthropic, OpenAI, Google, xAI) have genuinely different training data, different alignment regimes, different blind spots. They do not see each other's output until the cross-examination round. There is no social pressure to converge, no sycophancy, no incentive to agree. Disagreement is the signal, not the failure mode.

Conclavik is developed and operated by Altanest SAS. It is not a research firm, not a chatbot wrapper, and not a replacement for the deal team. It produces no recommendations, holds no view on the underlying asset, and treats every query as the property of the account that submitted it. It is a structured challenge function: a way to stress-test the case before committing capital, time, or reputation.