Evidence-governed intelligence systems
_{Serious systems. Deliberately humble names.}

Standard · Portfolio · Architecture · Evidence

ObtuseAI builds AI systems for complex, consequential work—systems that bind claims and outputs to inspectable evidence, preserve uncertainty and negative results, and stop before machine authority outruns the evidence.

Our portfolio spans engineering orchestration, market research, prediction markets, metacognitive optimization, cyber-defense research, creative production, and sports simulation. The domains differ. The operating doctrine does not:

Trace what the system saw. Replay what it did. Inspect why it stopped. Keep consequential authority human-owned.

We are engineering toward frontier capability through a less fashionable discipline: making powerful systems legible, falsifiable, and bounded.

The repositories are separate products, not disconnected experiments. They implement a shared evidence doctrine across three portfolio layers.

flowchart TB
    EC["Obtuse Evidence Contract<br/>Provenance · Replay · Authority · Limits"]

    subgraph CP["Engineering control plane"]
        BL["Blunder<br/>Agent missions and validation"]
    end

    subgraph IL["Intelligence laboratories"]
        DP["DOPEY<br/>Market research"]
        DU["Dummy<br/>Prediction markets"]
        DF["Doofus<br/>Evolution and metacognition"]
        NI["Nimrod<br/>Cyber-defense assurance"]
    end

    subgraph DE["Domain evidence engines"]
        DI["Dimwit<br/>Multi-DCC production"]
        WA["Waterboy<br/>Sports simulation"]
    end

    EC --> BL
    EC --> DP
    EC --> DU
    EC --> DF
    EC --> NI
    EC --> DI
    EC --> WA

    BL -. "governed engineering" .-> IL
    BL -. "governed engineering" .-> DE

    IL --> EV["Versioned evidence<br/>Results · Limits · Receipts"]
    DE --> EV
    EV --> HA["Human-owned consequential authority"]

“Frontier” should describe the quality of the evidence, not the volume of the claim. ObtuseAI systems are designed to make ambitious work inspectable:

• capability claims must name the evaluated scope;
• green automation must not be confused with external validation;
• uncertainty must survive the interface;
• authority must be narrower than the evidence;
• independent evaluation should be welcomed, reproducible, and version-bound.

The portfolio is frontier-engineered and is building toward externally substantiated frontier leadership. We publish explicit maturity and authority labels because credibility compounds when the boundary is visible.

The current organization snapshot is anchored to seven public product repositories and specific commits audited on 2026-07-26.

Passing configured checks is a current engineering signal, not a claim of equal assurance depth, complete coverage, security certification, or external validation. The exact commits, boundaries, and audit caveats are preserved in the versioned portfolio snapshot.

• Builders: inspect the systems, reproduce the evidence, and challenge the boundaries.
• Evaluators: test the claims against explicit commits and declared authority ceilings.
• Partners: bring consequential workflows that deserve more than a plausible answer.

AI systems that bind outputs to evidence—and stop where evidence ends.