Research on whether claims of human control over AI-assisted decisions can be checked.
The question this program asks. When an AI-assisted decision is said to remain under human control, what evidence shows that a person could intervene, and that the evaluation would detect a control failure?
Node & Norm’s active research is focused on revocation in delegated AI workflows: when permission is withdrawn, does further protected action stop? Further research and development on the other projects is paused. Their published methods, code and evidence remain available, with findings and limitations preserved.
| You want to | Go to |
|---|---|
| Understand the current study | Revocation in delegated AI workflows: public research overview. The study repository remains private; no public replication package is offered here. |
| Inspect a method or implementation | Repository directory, with each project's retained evidence and limits |
| Read the argument first | Essays and program purpose |
| Check how AI contributes | Research practice and each repository's disclosure |
The eight-project research overview distinguishes five kinds of evidence. Its recorded counts, checked 11 November 2026, are:
| Kind of evidence | Held | Open |
|---|---|---|
| Method or specification written | 8 of 8 projects | 0 |
| Applied to cases or records | 6 of 7 | 1 |
| Agreement or software checks | 4 of 7 | 3 |
| Independent replication | 0 of 7 | 7 |
| Effect in a deployed setting | 0 of 8 | 8 |
These counts cover the projects designed to admit each kind of evidence, not the number of GitHub repositories. They include authored demonstrations and bounded software checks where labeled. A held cell does not establish evidence in another column.
HIT retains an earlier two-scorer exercise with 7/7 agreement under earlier rules. That is not independent replication under the current rules or evidence of general reliability. Its current candidate protocol prohibits scoring while its release gates remain unresolved.
Each entry preserves its own methods, versions and research boundaries. This presentation changes no study approval, recruitment permission or maintenance commitment.
| Work | Retained contribution | What remains open |
|---|---|---|
| Trust, Autonomy, and Evidence | Retrospective reconstruction method; three historical cases, all conditionally Unresolved | Independent reliability, validity and transfer to contemporary AI systems |
| Human Influence Telemetry | Documentary scoring method; one preserved scoring exercise under earlier rules; maintenance | Current-protocol approval and independent replication. Candidate-protocol scoring remains prohibited. |
| Control Evidence Corpus | Development record structure; three authored synthetic demonstrations | Empirical pilot, independent coding, validity and reliability |
| Governed Decision Intelligence | Specification and reference implementation; structure, routing and receipt-integrity checks | Factual truth, reviewer understanding and deployment effect. Upstream interoperability driver remains in review. |
| Regulated Gate Decision Support | Conditional-approval record format and validation tools; six illustrative records | Empirical evaluation, regulatory acceptance and workflow effect |
| Catholic Doctrinal Fidelity Index | Provisional scoring rules and review procedures for SAICRED | Rule reconciliation and qualified theological validation |
| Applied AI Research Translator | Templates, bounded tasks and retained demonstrations | Independent evaluation and institutional benefit |
| RGDS Independent Study | Exploratory study of decision records in regulated drug development | Supporting records for retained quantitative claims and substantive review |
| AI Assistance Governance | Working method for recording AI participation and the human basis for a decision | Empirical field evaluation |
The current revocation study and Poenitentia Institutionum chapter are described in the public research overview; their private repositories are not linked here.
Read the repository's limits, protocol status and source records before attempting a study. A public repository or research question is not permission to begin participant recruitment, scoring or a new experiment. Code checks, documentary review and independent empirical validation answer different questions.
For a reproducible software check, record the commit, environment, command and expected output specified by that repository. Preserve unexpected results and unresolved findings. Report a bounded documentation or software issue in the relevant repository without uploading confidential records.
Cite the specific work and version you used. Later commits can contain work outside a release.
| Work | Retained version | Citation record |
|---|---|---|
| Trust, Autonomy, and Evidence | Software v0.18.0 | CITATION.cff |
| Human Influence Telemetry | Software v0.6.7 | CITATION.cff |
| Governed Decision Intelligence | v3.0 | CITATION.cff |
| Control Evidence Corpus | Development v0.3.0-dev | CITATION.cff |
| Regulated Gate Decision Support | Implementation v2.0.1 | CITATION.cff |
| Catholic Doctrinal Fidelity Index | Archived v1.5 | CITATION.cff |
CEC also retains a separate Catholic-pilot v0.4.1-dev prerelease. That disclosure/readability update does not replace the infrastructure version above or establish a new experiment.
Mark Julius Banasihan works with AI agents on the program and is accountable for what is published. Read the AI-use statement alongside each repository's own disclosure and review records. An AI-assisted replay or review within the program is not independent external validation.