Agentic Workflow Forensics Experiment Status Research Status
Preparation is active. Empirical claims remain closed.
What changed in the measurement model?
An earlier numeric Agent Workflow Reconstructability Score was retired. Aggregate scores could conceal the specific missing evidence that controls a decision and risked turning incomplete reconstruction into a favorable-looking total.
The current Agent Workflow Reconstructability Determination v2.0 uses material questions and criterion-level findings. A determination may report a claim as established, limited, contradicted, or not established; identify blocking gaps; and state the decision consequence. It does not issue an aggregate readiness score, maturity band, compliance certification, or prediction of future behavior.
What remains before confirmatory execution?
Required preparation includes current material-question, ground-truth, analyst-record, manifest, and protocol schemas; requalified engineering controls; analyst pilot evidence; ethics review; preregistration; and explicit execution authorization.
Previous engineering qualification, legacy score-bound manifests, or pilot artifacts cannot substitute for those gates.
What may be published now?
The Foundation may publish the versioned field definition, control-surface methodology, public incident demonstrations, open-source implementation boundaries, audit templates, and transparent research status. These are definitional, methodological, or operational artifacts.
The Foundation may not publish expected findings as results, describe the current determination as validated, generalize prevalence, or claim intervention effectiveness from an experiment that has not run.
What happens if the result is null or mixed?
A positive result is not a publication gate. Null, mixed, inconclusive, or adverse findings remain publishable when the analysis, deviations, limitations, and evidence boundaries are reported faithfully. The method’s purpose is to improve the defensibility of decisions, including the decision that evidence does not support a hoped-for conclusion.