Design and analyze controlled A/B experiments with a testable hypothesis, one primary metric, a sample size based on the smallest useful effect, and a fixed stopping rule. Interpret results using confidence intervals and significance, check assignment problems, and apply CUPED to reduce variance when pre-test data is useful.
- Planning sample size and duration before an experiment
- Checking significance, confidence intervals, and practical impact
- Diagnosing a broken test split before trusting results
ericrisco/rsc-harness☆ 613Updated Nov 16, 2026
Improve a web interface toward WCAG 2.2 Level AA by fixing semantics, keyboard access, focus behavior, ARIA names and states, contrast, and target sizes. Combine automated scans with manual checks, since tools cannot assess every accessibility issue.
- Reviewing a web UI for WCAG 2.2 AA issues
- Making custom controls usable by keyboard and screen readers
- Checking focus behavior in dialogs and menus
ericrisco/rsc-harness☆ 613Updated Nov 16, 2026
Manage or troubleshoot paid campaigns on Google and Meta. Choose campaign types, shape creative for each ad surface, set budgets and scaling rules, calculate break-even return on ad spend, and identify consent or conversion-tracking gaps.
- Choosing a Google or Meta campaign structure
- Checking whether ad returns cover gross margin
- Reviewing conversion tracking before scaling spend
ericrisco/rsc-harness☆ 613Updated Nov 16, 2026
Measure changes in LLM and agent quality with a versioned set of real test cases, calibrated scoring, and regression checks. Evaluate RAG answers and agent tool-use paths, then create a pre-merge gate that flags meaningful drops against a baseline.
- Creating a labeled evaluation set from real failures
- Calibrating an LLM judge against human ratings
- Scoring RAG answers and retrieval quality
ericrisco/rsc-harness☆ 613Updated Nov 16, 2026
Review the boundaries around a running AI agent: limit its task scope and tools, treat web, email, and retrieved text as untrusted, gate risky actions for human approval, and set runtime and cost caps. It focuses on controlling agent autonomy.
- Limiting an agent to a defined task and minimum required tools
- Protecting tool use from instructions hidden in web pages or emails
- Requiring approval for irreversible actions such as sending or deleting
ericrisco/rsc-harness☆ 613Updated Nov 16, 2026
Plan a media workflow, choose generative models for voice, image-to-video clips, and music, then assemble the results with ffmpeg. Use it to produce finished files such as narrated explainers, product teasers, and short videos.
- Creating a narrated explainer video
- Making a product teaser from a still image
- Producing a short video with captions
ericrisco/rsc-harness☆ 613Updated Nov 16, 2026