ai4he/headroom-decision-procedure

Headroom Decision Procedure Examples

Dataset accompanying the paper Headroom Before Protocol Effects: A Decision Procedure for Human-Agent Evaluation (Human-AI Coevolution Workshop (HAIC) at NeurIPS 2026).

Synthetic worked example for the headroom decision procedure: worked-headroom, planning-headroom, severity-scaled and guard-sensitivity result tables, and the scripts that generate them.

Paper	Headroom Before Protocol Effects: A Decision Procedure for Human-Agent Evaluation
Venue	Human-AI Coevolution Workshop (HAIC) at NeurIPS 2026
Authors	Manuel Delaflor, Carlos Toxtli
License	Creative Commons Attribution 4.0 International (cc-by-4.0); code: MIT License
Dataset DOI	10.5281/zenodo.24179019
Version	1.0.0
Maintainer	Human-AI Empowerment Lab, Clemson University

Paper abstract

A team protocol has no interpretable benefit without a matched no-protocol baseline and decision-relevant headroom. We present a reporting method that tests admissibility, identifies the marginal contrast, evaluates opportunity, and maps paired uncertainty to an operational action. It distinguishes exact headroom on a fixed battery from uncertain population headroom used in planning. A published meta-analysis of human-AI experiments illustrates why the estimand matters: the same evidence supports augmentation relative to humans alone but contradicts synergy relative to the better constituent. Neither comparison identifies a coordination protocol without a matched no-protocol team. An executable synthetic application further shows that a promising point estimate can remain operationally unresolved once paired uncertainty is considered, and that near-perfect pilot performance does not by itself make a modest benefit impossible. Several inherited multi-model summaries fail the method's artifact gates, so they do not validate it. This is a bounded position and reporting contribution rather than an adoption-ready standard. Human-facing claims still require direct human outcomes, and we introduce no new human experiment.

Dataset structure

Path	Format	Size	Records
code/guard_sensitivity.py	py	3.1 KB
code/severity_scaled_case.py	py	3.3 KB
code/worked_headroom.py	py	7.2 KB
data/guard_sensitivity_results.csv	csv	3.2 KB	72
data/planning_headroom_results.csv	csv	848 B	3
data/severity_scaled_results.csv	csv	328 B	2
data/worked_headroom_results.csv	csv	1.1 KB	4
docs/README.md	md	582 B

8 files, 19.7 KB in total (metadata files excluded). Records are rows for CSV/TSV/JSONL files and list length for JSON arrays.

Loading the data

from datasets import load_dataset

ds = load_dataset("ai4he/headroom-decision-procedure", "guard_sensitivity_results")

Available subsets: guard_sensitivity_results, worked_headroom_results, planning_headroom_results, severity_scaled_results. All other files can be downloaded with huggingface_hub.snapshot_download or from the Zenodo record.

Known limitations

• All tables are synthetic illustrations; the five earlier pilots audited in the paper have no retained raw data.

Personal and sensitive information

None.

License

The data and documentation are released under the Creative Commons Attribution 4.0 International. Source code files (.ipynb, .js, .m, .py, .sh) are released under the MIT License.

Citation

If you use this dataset, please cite the paper:

@inproceedings{Delaflor2026Headroom,
 title = {Headroom Before Protocol Effects: A Decision Procedure for Human-Agent Evaluation},
 author = {Delaflor, Manuel and Toxtli, Carlos},
 booktitle = {Human-AI Coevolution: Measuring Human-Agent Teams in the Agentic Era (HAIC) Workshop at NeurIPS 2026},
 address = {Atlanta, GA},
 year = {2026},
 month = dec,
 note = {Workshop paper},
 url = {https://openreview.net/forum?id=odOgsQYleZ}
}

and the dataset:

@dataset{Delaflor2026HeadroomDecisionProcedurData,
  author = {Delaflor, Manuel and Toxtli, Carlos},
  title = {{Headroom Decision Procedure Examples}: Dataset for ``Headroom Before Protocol Effects: A Decision Procedure for Human-Agent Evaluation''},
  year = {2026},
  version = {1.0.0},
  publisher = {Zenodo},
  doi = {10.5281/zenodo.24179019},
  url = {https://doi.org/10.5281/zenodo.24179019}
}

Further documentation: docs/README.md.