4.0 — Decision Intelligence Framework
The repository whose name matches the endeavor phrase most closely is the one least aligned with it. This rebuild makes the name true: optimization-driven, system-level decision frameworks.
Part of a five-repository portfolio on one theme: **optimization-driven, system-level decision frameworks** — integrating operations research, mathematical optimization and applied AI — for domains where a wrong decision carries systemic consequences. The three pillars are financial stability, healthcare safety and secure digital infrastructure.
| | | |---|---| | Projects with measured results | 3 of 4 | | Tests across the suites | 107 | | Running on real third-party data | 3 of 4 | | Most recent reportable run | 2026-09-07 | | Verdict | Built around its own name — three projects measured |
"Measured" means: original work, a stated method, real data at a stated
scale, a measured result, and a README a reviewer can follow. Counts exclude
duplicates, forks of third-party work, retired projects, and asset-only
folders.
Projects
| Folder | Project | Pillar | Evidence value |
|---|---|---|---|
| 2-optimization-under-uncertainty/ | Optimization-Under-Uncertainty Library | Cross-cutting — the endeavor itself | CORE — the endeavor expressed as code |
| 1-decision-audit-framework/ | Decision-Audit Framework | Cross-cutting | CORE — makes interpretability an artifact |
| 3-icu-triage-optimization/ | Multi-Objective ICU Triage Optimization | Healthcare Safety | CORE — joins the healthcare pillar to the OR framing |
| 4-decision-benchmark-suite/ | Decision-Framework Benchmark Suite | Cross-cutting | CORE — the most citable form of contribution |
What each one is
1. Optimization-Under-Uncertainty Library — 2-optimization-under-uncertainty/
Stochastic and robust optimization applied to portfolio allocation and hospital resource allocation.
Why it earns its place: The usual comparison reports each formulation's objective value under its own distribution, which flatters whichever one generated the scenarios. The informative test is cross-evaluation: score the stochastic solution under the robust uncertainty set and the robust solution under the sampled distribution, and report both.
Target scale: Portfolio allocation and hospital resource allocation instances
2. Decision-Audit Framework — 1-decision-audit-framework/
Decision logging, counterfactual replay, and attribution for AI-assisted decisions.
Why it earns its place: Makes "auditable, interpretable decision frameworks" an artifact rather than a phrase.
Target scale: Decision traces from the other projects in this repository
3. Multi-Objective ICU Triage Optimization — 3-icu-triage-optimization/
Multi-objective ICU triage and alert-threshold optimization.
Why it earns its place: Triage policy and alert threshold are normally optimised separately, but the threshold determines the arrival process the triage policy then faces. Optimising either one alone yields a policy that degrades as soon as the other moves, so the two have to be solved together to mean anything.
Target scale: MIMIC-IV / eICU cohorts, shared with repo 2.0
4. Decision-Framework Benchmark Suite — 4-decision-benchmark-suite/
Regret, calibration and robustness benchmarks across the frameworks in this repository.
Why it earns its place: A benchmark is the most citable form of contribution and the easiest for others to adopt.
Target scale: All frameworks in this repository
Repository layout
4.0-Decision-Intelligence-Framework/
├── 2-optimization-under-uncertainty/
├── 1-decision-audit-framework/
├── 3-icu-triage-optimization/
└── 4-decision-benchmark-suite/Every rebuilt project carries the same skeleton: README.md, src/, data/, results/, tests/, website/.
Earlier and off-theme prior work (dashboards, consumer apps, templates, and other non-endeavor folders) has been moved to a separate archive-prior-work repository to keep this repository focused on the endeavor. It remains recoverable from this repository's git history.
Ground rules
- No number without a run log. Any figure quoted anywhere in this
results/README.md with a run
date behind it.
- No simulated data under a real claim. Sample data lives in
data/sample/, labelled, and is never the source of a cited figure.
- Adoption must be documentable — named institutions, dated
- Third-party and forked code stays labelled and is never counted.
Structure only — no results are claimed here unless a project's
results/README.md records a dated run.
Licence
Apache License 2.0. The code, benchmarks and documentation in this repository may be reused, modified and redistributed under its terms, including by projects that need a patent grant to incorporate them.
Third-party data and vendored code kept under */data/ and
historical-archive/ carry their own terms and are labelled where they appear.