The MMC capstone rubric labels Reduced Inequalities as program option 4. In the official United Nations Sustainable Development Goals, Reduced Inequalities is SDG 10. ControlSift therefore uses both labels together: MMC rubric option 4 / official UN SDG 10.
NIST explicitly notes that smaller businesses often operate with limited cybersecurity resources and budgets and may be unable to afford dedicated cybersecurity staff. ControlSift doesn't claim that this proves a specific shortage of control-evidence reviewers; it uses the broader resource constraint as motivation to study an open, low-cost assurance-support method.
The project uses synthetic data, an inexpensive classical baseline, a small language model, public code, free or commodity GPU infrastructure, and explicit human-review boundaries. The goal is inspectable research access, not a cheap substitute for professional judgment.
Within Gemma v1.0, few-shot is strongest and QLoRA doesn't beat it. The hardened v1.1 classical benchmark has a TF-IDF baseline around 0.53 macro F1. The capstone contribution isn't a model victory; it's an open, low-cost, inspectable research process with honest limits and failure visibility.
Better evidence discipline can support institutional accountability. This is useful context, not the capstone's formal selection.
Efficient fine-tuning and a public learning artifact are adjacent to innovation and education, but they're supporting connections only.