/claim #16

Summary

• Adds a distinct model-assumption-diagnostics-assistant/ slice for the AI-Powered Research Assistant Suite.
• Reviews synthetic analysis packets for model-family mismatch, low events per predictor, separation risk, heteroscedasticity, multicollinearity, residual outliers, autocorrelation, few-cluster fragility, missing calibration, and incomplete diagnostic handoff evidence.
• Emits deterministic JSON, Markdown, SVG, GIF, and MP4 demo artifacts from synthetic data.

Non-overlap

This is not another study-power, Bayesian-prior, missing-data, causal-adjustment, uncertainty-calibration, figure/citation claim, external-validity, randomization/blinding, spatial-autocorrelation, or sample chain-of-custody assistant. It focuses specifically on whether the declared statistical/ML model assumptions support manuscript claims before an AI review summary is trusted.

Validation

• npm test -> passed
• npm run demo -> generated reports/model-assumption-report.json, reports/model-assumption-report.md, reports/summary.svg, and reports/demo-script.txt
• npm run demo:video -> generated reports/demo.gif and reports/demo.mp4
• git diff --check -> passed

Synthetic data only. No credentials, private research data, patient data, external APIs, payment systems, live model calls, or network services are used by this module.

AI-assisted with OpenAI Codex; I reviewed and locally verified the diff before submitting.