01 / Evaluation
A score depends on a rule. An agreement depends on what reviewers saw. These studies examine what a result supports and where its interpretation must stop.
What should disqualify an otherwise persuasive answer? CDFI examines scoring rules, source fidelity and domain-specific authority. Its public record compares the written formula with the configuration and calculation.
Evolving. Scoring-contract reconciliation and human theological validation remain open. SAICRED retains its separate project and publication authority.
Can two reviewers reach the same finding from the same evidence? HIT examines how human influence is documented and how consistently a bounded record can be assessed.
Research in progress. The reported independent-application result covers two scorers, one case packet and one scoring contract. Broader reliability remains unestablished.
02 / Assurance
Assigned authority needs a practical route to action. These methods and reference implementations examine what a reviewer could know, when intervention was possible and what record remains.
A person appears in the workflow. Could they understand the situation, act in time and make an intervention take effect? TAE develops a method for examining practical human control through the available record.
Historical method demonstration: three selected cases, assessed by one researcher. Independent reliability and transfer to current AI systems remain open.
An institution acts on a recommendation. GDI proposes a record connecting the decision question, evidence, alternatives, uncertainty, authority and downstream obligations so the action can be reconstructed.
Specification and reference implementation. Synthetic examples and implementation checks do not establish field effectiveness.
Why did a team proceed, stop or wait? RGDS records the evidence gaps, accepted risks, responsibilities and conditions attached to a consequential phase-gate decision.
Six authored illustrative records in the reviewed public account. Institutional benefits remain unmeasured. The related AI Assistance Governance profile has its own scope and version.
Research in development
Can an independent reviewer reconstruct whether an AI control had practical force from the evidence that survives an event? This proposed research would follow individual control opportunities through public records, tracing source relationships, human authority, intervention and downstream effects.
Incident databases would help identify cases; primary records would support examination. The method would preserve missing evidence and unresolved findings, with independent coding and later validation against known control states.
In development. The initial study would examine what the public record makes observable. It would not establish how often AI controls work across the wider population.
What does an institution owe people after an automated decision causes harm? Forthcoming chapter work connects the displacement of human counsel, judgment and command with HIT’s evidentiary questions and Poenitentia Institutionum, a model of institutional repair and reform drawing on contrition, confession and satisfaction.
The contribution is being developed for AI and Catholic Perspective: Intervention from Economic Catholic Front, a forthcoming volume planned for publication by En Route Books and Media.
Coming soon · In editorial review.
