HALOS Example: Education / Student Learning
Education / Student Learning
Mateo Rivera — Undergraduate biochemistry student
Rejected AI-fabricated citation after verification failure
Examples
Real-world scenarios showing how HALOS provenance records capture human-AI collaboration across professional domains.
Each example includes a detailed collaboration narrative — who the human is, what they’re building, how AI is used, and where human judgment makes the difference. Every scenario demonstrates at least one decision where AI output is rejected or substantially modified with domain-specific rationale.
All scenarios, names, and organizations are fictional. Provenance records use the v0.3 provenance spec with graph model, decision provenance, human-AI interaction semantics, multi-policy governance, and policy evaluation traces.
The machine-readable .halos.json provenance records, additional example types, and a generation skill for contributing new examples are all in the halos-spec examples directory →
Domain Examples
HALOS Example: Enterprise Software Development
Enterprise Software Development
Tomoko Hayashi — Senior platform engineer
Rejected AI-recommended CQRS; rewrote state machine from scratch
HALOS Example: Government / Policy / Legal
Government / Policy / Legal
Fatima Al-Rashid — Senior policy analyst
Chose how to frame politically consequential housing supply finding
HALOS Example: Journalism / News Media
Journalism / News Media
Carlos Medina — Investigative reporter
Corrected AI-generated violation dataset before publication
HALOS Example: Music / Creative Production
Music / Creative Production
Maya Reeves — Film composer
Rejected AI harmonic sketch, kept only structural idea
HALOS Example: Nonprofit / Humanitarian
Nonprofit / Humanitarian
Amara Diallo — Field coordinator
Overrode AI priority ranking based on ground-truth field report
HALOS Example: Real-time, Critical Systems
Real-time Critical Systems
Luis Herrera, PE — Control systems engineer
Reduced AI-proposed dosing reduction, added safety logic AI omitted
HALOS Example: Scientific Research
Scientific Research
Dr. Nkechi Okonkwo — Computational epidemiologist
Excluded confounded feature the AI ranked as highly predictive
Companion Resources
Each narrative has a machine-readable companion .halos.json provenance record in the halos-spec examples directory.
Additional example types in halos-spec:
| Example | Purpose |
|---|---|
| minimal.json | Smallest valid v0.1 provenance record |
| v0.2-graph.json | v0.2 graph model reference example |
| CycloneDX embedding | HALOS inside a CycloneDX SBOM |
| SLSA embedding | HALOS as an in-toto/SLSA predicate |
Contributing
New domain examples can be contributed to halos-spec. Use the GENERATE-EXAMPLE.md agent prompt to generate a new example with an AI assistant.