“The difficulty lies not so much in developing new ideas as in escaping from old ones.”
John Maynard Keynes
Precedent is a shield, not a weapon. In science, strategy, and law, precedent protects you from risk, but it mathematically prevents outperformance. If you rely entirely on consensus intelligence, your absolute ceiling is the industry average. True competitive advantage requires rigorous divergence: extracting the exact, verified insights that the standard models filter out.
Some prompt techniques can elicit very limited novelty from LLMs, but always at the cost of coherence, a.k.a. the “utility tax”. Novamine currently holds the highest novelty score on Carnegie Mellon's NoveltyBench by a wide margin.
Whilst some agentic approaches have achieved high rigour & hallucination suppression, combining state-of-the-art rigour with truly innovative output has eluded all of them.
Novamine currently holds the single highest utility score on Carnegie Mellon's NoveltyBench.
Two horizontal platforms. Infinite vertical applications.
Requires the highest novelty and the lowest utility.
Requires high novelty and medium utility (e.g. a prior-art check).
Requires maximum novelty and maximum utility, but also speed: Novacore.
Requires maximum novelty and maximum utility; speed is no issue, so suited to a full Novamine run.
Insurance and Private Equity require strict evidentiary utility and targeted non-consensus risk discovery.
Divergent Test-Time Compute API & Chat
Full agentic adversarial deep discovery network
Pinch to zoom
The most novel answers on the leading independent benchmark, and the most rigorous, from one system.
The full agentic adversarial network, run multiple times over 24 to 72 hours, as directed by both your experts and ours working collaboratively.
A drop-in step (3 to 60 minutes, Rigour 0 to 3) at the point where an existing workflow needs non-consensus breadth. An API (Application Programming Interface) your team plugs directly into your existing systems.
Independently reproduced and merged by the Carnegie Mellon NoveltyBench maintainers (10 September 2026). Every benchmark figure on this page was scored by the benchmark's own scoring.
In high-stakes science & strategy, professionals confuse consensus with safety. But if every agency, firm, and lab is looking at the exact same data and generating the exact same median response, consensus is a vulnerability.
Novacore leads on fast Divergent Test-Time Compute (3 to 60 minutes). A full Novamine engagement runs our Agentic Adversarial Deep-Discovery Network (24 to 72 hours, 9+/10 Novelty and 9+/10 Utility), where novelty and rigour are forged directly from first-principles constraints, adversarial falsification, and specialist calibration under NDA. Breakthroughs do not live in the historical record. To gain an asymmetric advantage, you must cross the threshold into measurable, mathematically hardened novelty.