Research Evidence Bundle¶
Use this page when you need a public, inspectable folder that explains what an IINTS run proves.
The evidence bundle is designed for demos, EUCYS-style project review, engineering handoff, and reproducibility checks. It is not clinical evidence and must not be presented as medical-device validation.
What It Builds¶
iints evidence build \
--run normal=results/live_demo/results/01_normal_run \
--run stress=results/live_demo/results/02_meal_stress \
--run supervisor=results/live_demo/results/03_supervisor_override \
--output-dir results/live_demo/evidence_bundle
The output folder contains:
| File | Meaning |
|---|---|
README.md |
Human-readable summary of the attached runs |
MODEL_CARD.md |
Research-only AI/model framing and safety gate summary |
evidence_summary.json |
Machine-readable evidence index |
run_index.csv |
Table with rows, mean glucose, TIR, and hypo percentages |
runs/*/RUN_CARD.json |
Per-run artifact and metric card |
runs/*/artifacts/ |
Copied CSVs, manifests, reports, and safety JSON where available |
Live Demo Integration¶
The easiest path is still:
iints demo --audience jury --output-dir results/live_demo
By default this now builds:
results/live_demo/
PRESENTER_GUIDE.md
DEMO_CUE_CARD.md
DEMO_ARTIFACTS.md
DEMO_RUN_LOG.txt
evidence_bundle/
If you need a faster rehearsal without the evidence folder:
iints demo --audience jury --output-dir results/live_demo --no-evidence
Add Local AI Evidence¶
After training a local AI lab:
iints research train-local-ai \
--run day1=results/jetson_research_day \
--run day2=results/jetson_research_day_2 \
--output-dir results/local_ai_lab
Attach that evidence to the public bundle:
iints evidence build \
--run day1=results/jetson_research_day \
--run day2=results/jetson_research_day_2 \
--local-ai-dir results/local_ai_lab \
--output-dir results/public_evidence
The model card will include the local AI safety gate status when LOCAL_AI_RESEARCH_SUMMARY.json is present.
Add Pico Pump Bench Evidence¶
After compiling and bench-testing a Pico bundle:
iints pump compile \
--algorithm algorithms/pico_bench_algorithm.py \
--output-dir bundles/pico_bench_bundle
iints pump bench-test \
--bundle-dir bundles/pico_bench_bundle \
--output-json bundles/pico_bench_bundle/bench_test_report.json
Attach the pump bundle:
iints evidence build \
--run one_day=results/one_day_sim \
--pump-bundle-dir bundles/pico_bench_bundle \
--output-dir results/public_evidence
Required Wording¶
Use this exact framing in public demos:
This is a research and educational evidence bundle. It shows reproducibility, software behavior, and safety-supervisor boundaries. It is not clinical evidence, not a certified medical-device submission, and not dosing advice.
Best Practice¶
- Run
iints run-doctorbefore long simulations. - Keep every run folder with its
run_manifest.json. - Attach local AI summaries only after safety gates run.
- Attach Pico pump bundles only after
iints pump bench-testpasses. - Present the evidence bundle next to the normal docs site or public Obsidian vault, not as a replacement for scientific validation.