Research

Autonomous agents, studied as behavioral entities.

Four programmes organize behavior under controlled variation. None has yet established agent-behavior findings or selected its first experiment.

What is established

Research has four programmes: Agent Behaviour & Emergence, Agentic Systems & Engineering, Evaluation & Measurement, and Multi-Agent & Human-Agent Systems.

What is required

Bounded, reproducible claims; discriminating interventions; and replication where the claim requires it.

What remains open

Measurements, experiment families, the first study and any question-derived apparatus remain undecided. The programmes do not activate them.

The work

Named here. Linked when the signal is public.

Some work is still private. It belongs in the index anyway.

12 entries
  1. 01
    ProgrammeOpen shelf

    Agent Behaviour & Emergence

    How observable agent behavior changes as conditions, feedback, memory, information and repeated interaction vary.

  2. 02
    ProgrammeOpen shelf

    Agentic Systems & Engineering

    What behavior belongs to a model and what belongs to its tools, authority, memory, orchestration and recovery system.

  3. 03
    ProgrammeOpen shelf

    Evaluation & Measurement

    How behavioral measures earn reliability, validity, comparability and stated limits.

  4. 04
    ProgrammeOpen shelf

    Multi-Agent & Human-Agent Systems

    How roles, communication, authority and institutions shape collective and human-agent behavior.

  5. 05
    StudyPublic

    Clock Dataset

    What was actually knowable at the moment of a decision?

  6. 06
    StudyPublic

    Decision Assurance

    When does a score stop being enough for a finite-capacity decision?

  7. 07
    StudyPublic

    Degradation Diagnosis

    How much can a model’s trouble be diagnosed before its labels arrive?

  8. 08
    StudyPublic

    Release Equivalence

    Is the system after release still the one you tested?

  9. 09
    StudyPublic

    DeepLOB

    How an experiment can manufacture confidence from a market with no signal.

  10. 10
    StudyPublic

    Spectral Recovery

    A controlled look at when correlation-matrix denoising helps — and when it does not.

  11. 11
    StudyPublic

    Tail Estimator Audit

    Known-tail experiments for finding where familiar estimators bend or break.

  12. 12
    FindingPublicFlagged

    Evidence that exists and cannot be retrieved

    A successful empty audit query did not establish an empty record.