Operative-State Contestability
Agent-Mediated Action and the Institutional Conversion of Computer Evidence
Working paper · 7 September 2026
Agent systems can increasingly produce technically reconstructable records of model calls, tool use, credentials, infrastructure events, and external actions. This paper asks what remains after that progress is conceded: whether a relevant challenge can alter the status or consequences of a machine-mediated state while its evidentiary basis is assembled and reviewed.
Working Paper
Download the working paper (PDF)
This working paper has not been peer reviewed. Comments and corrections are welcome.
Core Contribution
The paper identifies operative-state contestability: an institutional property whereby a relevant challenge can trigger a process capable of altering the status or consequences of an operative state while the evidentiary basis for that state is assembled and reviewed.
It distinguishes three questions that are often collapsed. Reconstruction asks whether someone can later determine what happened. Contestation asks whether a relevant actor can challenge the record, interpretation, or state. Status control asks whether that challenge changes what the disputed state may do while review is pending.
Approach
The paper uses the Post Office Horizon litigation as a discriminating non-AI case, US public-benefits litigation as a comparative administrative line, and the July-August 2026 OpenAI, Hugging Face, and METR record as an agent-side contrast.
A capability-duty matrix tests what documented technical systems, evidence-management components, legal instruments, and procurement guidance already supply. A prototype Runtime Agent Deployment Governance Stress Test translates the concept into questions for prospective deployment. The paper treats its proposition about agent-mediated deployments as a hypothesis awaiting prospective testing.
Related Methods
- Governance Stress-TestsThe broader public method for testing institutional readiness before deployment.
- Governance Stress-Test ChecklistA reusable diagnostic for evidence, authority, recourse, control, and learning capacity.
- Governance Stress-Test Protocol NoteThe evidence and workflow protocol behind the method.
Version History
7 September 2026. Current working-paper version.
16 August 2026. Earlier public working paper, From Traces to Evidence, version 0.1.
The current paper substantially revises and retitles the earlier working paper while preserving the development record.
Suggested Citation
Nadella, Venkat. 2026. “Operative-State Contestability: Agent-Mediated Action and the Institutional Conversion of Computer Evidence.” Working paper, 7 September 2026.