Skip to content
View mmjbds's full-sized avatar

Block or report mmjbds

Block user

Prevent this user from interacting with your repositories and sending you notifications. Learn more about blocking users.

You must be logged in to block users.

Maximum 250 characters. Please don’t include any personal information such as legal names or email addresses. Markdown is supported. This note will only be visible to you.
Report abuse

Contact GitHub support about this user’s behavior. Learn more about reporting abuse.

Report abuse
mmjbds/README.md

Mian Zhang

Independent researcher building evidence-gated AI systems that must learn from action, feedback, and failure.

AI systems can produce fluent answers. My work asks the harder operational questions: What changes after the system acts? Does a correction survive the next interaction? What evidence must exist before a consequential action is authorized?

Start Here

Public Projects

Project Public question First action
ReflexBench Does agent reasoning survive when its output changes users, evidence, incentives, or institutions? Inspect 20 released scenarios and four observer-depth levels.
WisdomBench Does feedback produce durable change, or does the system repeat the same failure? Recompute the bundled longitudinal metrics.
Proof-Carrying Action What evidence must accompany a consequential AI action request? Run the no-credit repair demo and inspect the proof schema.
SOVEREIGN public interfaces How can verified failure become a scoped, reviewable, reversible rule? Run the deterministic failure-memory lifecycle fixture.

Research Archive

Contribute

Questions, safe use cases, documentation repairs, public baselines, negative results, reproducible extensions, and narrow interoperability work are welcome.

Boundary

The public repositories expose papers, protocols, schemas, validators, small fixtures, minimal references, and documented limitations. They do not expose production orchestration, exact operational thresholds or weights, private prompts or data, customer systems, deployment automation, or unreleased research. A public benchmark result is not production safety certification.

Full boundary: OPEN_SOURCE_BOUNDARY.md

Popular repositories Loading

  1. reflexive-intelligence-paper reflexive-intelligence-paper Public

    Public paper and artifacts for decision-making in observer-participant environments.

    TeX 2

  2. ouroboros-papers ouroboros-papers Public

    Public manuscript archive for reflexive intelligence, multi-reward learning, and bounded research claims.

    TeX 2

  3. reflexbench reflexbench Public

    Benchmark for observer-participant failure and counterfactual trustworthiness in agentic AI.

    TeX 1

  4. wisdombench wisdombench Public

    Longitudinal benchmark for measuring whether AI agents learn from repeated failure and feedback.

    Python 1

  5. sovereign-os sovereign-os Public

    Minimal public interfaces for cognitive immunity and failure-memory experiments in AI agents.

    Python 1

  6. mmjbds mmjbds Public

    Config files for my GitHub profile.