Workflow Hub

Every IINTS-AF workflow should make four things clear: input, command, output, and review rule.

Choose the workflow that matches your question. Do not enable every feature simply because it is available.

Simulation

Question: How does a candidate algorithm behave for a controlled virtual patient and scenario?

Start: Complete First Workflow

Output: time series, report, metadata, manifest, and optional audit/validation files.

Review rule: confirm the experiment settings and raw time series before interpreting summary metrics.

Algorithm Or Scenario Study

Question: Does one controlled change alter outcomes across seeds, patients, or scenarios?

Start: Scientific Workflow

Output: protocol, study runs, aggregate metrics, comparisons, figures, and exclusions.

Review rule: predefine the comparison and preserve failed runs.

Data Certification

Question: Does a CSV satisfy an explicit schema and quality contract?

Start: Certification Quickstart

Output: certification JSON, grade, validation details, optional dashboard, and integrity artifacts.

Review rule: certification reports checks performed; it does not establish clinical fitness or remove bias.

Local AI Review

Question: Can a local model explain or challenge a completed, validated run?

Start: AI Assistant

Output: review notes or a structured explanation linked to prepared evidence.

Review rule: verify every number and causal claim against deterministic artifacts.

Glucose Forecast Research

Question: How accurately and plausibly can a model forecast future glucose at defined horizons?

Start: Glucose Forecast Model

Output: checkpoint, resolved configuration, split manifest, horizon metrics, hypo-detection metrics, violation metrics, and model card.

Review rule: prevent subject leakage and report horizon-specific performance, uncertainty, and physiological violations.

Mechanistic Reference Model

Question: Does an independent published or externally supplied equation model reproduce, challenge, or clarify one IINTS physiological assumption?

Start: Mechanistic Reference Models

Output: SBML structural summary, exact model hash, independent time-course CSV, execution manifest, and review report.

Review rule: verify source, licence, model population, units, variables, and mappings before comparing outputs; never calibrate IINTS silently.

Cross-scale Reference Evidence

Question: Which assumptions remain stable when challenged by sensitivity analysis, independent CellML validation, a device-physics FMU, or measured binding assays?

Start: Cross-scale Reference Labs

Output: model inspections, exact hashes, engine logs, optional time-course CSVs, affinity records, manifests, and reviewer notes.

Review rule: keep structural confidence, assay affinity, dynamic physiology, physical-device behaviour, and clinical outcomes separate; never execute an untrusted FMU.

Results Management

Question: How can many completed runs be indexed and compared without losing provenance?

Start: Study Analysis

Output: run index, aggregate tables, study summaries, and evidence-ready exports.

Review rule: keep immutable source run bundles and derive summaries from them.

Reports And Public Evidence

Question: How can a completed experiment be communicated without hiding assumptions?

Start: Research Evidence Bundle

Output: copied manifests, run cards, model/data context, figures, and a public file index.

Review rule: include limitations, intended use, sources, and failed or excluded results.

Academic Reproducibility Package

Question: Can another researcher identify and verify the exact files, software revision, settings, seed, and references behind one completed run?

Start: Academic Research Workbench

Output: RO-Crate JSON-LD, SHA-256 artifact inventory, reproducibility audit, source-registry snapshot, and review guide.

Review rule: resolve audit findings and complete privacy, licence, and claim-level source review before sharing.

Desktop Workflow

Question: Can the same SDK operations be used through a graphical workbench?

Start: Desktop App

Output: normal SDK run bundles. The app is a shell; it does not define separate scientific behavior.

Review rule: keep the output folder and command-equivalent settings available for reproducibility.

Hardware Workflow

Question: How does SDK logic behave on a bench device or edge computer?

Start: Hardware Hub

Output: device diagnostics, transport logs, timing measurements, comparison artifacts, or bench reports.

Review rule: no real patient connection or insulin delivery; document hardware, firmware, transport, and timing.

Quick Command Map

iints demo quick --output-dir results/demo
iints run --dry-run --preset baseline_t1d
iints validation-profiles
iints data certify --help
iints ai local-check
iints run-study --help
iints evidence build --help
iints research academic-bundle --help
iints research mechanistic inspect --help
iints edge doctor

For exact arguments, use Command Cheatsheet or the command's --help output.