“Test small before you go big”

Canary deployments route a small percentage of traffic to new code while the rest continues on the old version. Automated metrics (error rates, latency, resource usage) are compared between canary and control. If the canary degrades, traffic is automatically shifted back. Feature flags add another dimension — code is deployed to everyone but activated only for a subset. Netflix's deployment system can roll out to 0.01% of users and roll back in seconds.

Drug development is the most formalized graduated rollout in existence. Phase I tests safety on 20-100 healthy volunteers. Phase II tests efficacy on 100-300 patients. Phase III tests superiority on 1,000-3,000 patients against existing treatments. Each phase is a gate: failure at any stage stops the rollout, protecting the wider population from an unsafe or ineffective drug. The system is slow but has prevented thousands of disasters.

Governments test policies in limited areas before national rollout. Finland's universal basic income experiment (2017-2018) tested unconditional payments on 2,000 unemployed citizens before considering broader implementation. Oregon's Drug Decriminalization (Measure 110) was a state-level pilot for a national policy debate. The best pilot programs have explicit success metrics, control groups, and predetermined decision criteria — the worst are political theater with no evaluation plan.

Farmers test new crop varieties, fertilizers, and techniques on small plots before committing their full acreage. Agricultural extension services have formalized this into demonstration farms and test plot networks. The cost of a failed experiment on one acre is manageable; the cost of a failed crop on a thousand acres is devastating.

Education reforms are routinely deployed at scale without graduated rollout. Common Core was adopted by 45 states before anyone had rigorous evidence about its effects. A graduated approach would test new curricula in a few classrooms, measure outcomes against controls, expand to schools, then districts, then states — with rollback at every stage. The technology for this exists (randomized controlled trials in education are well-established); the institutional will to use it is what's missing.

Most urban changes are permanent and expensive — a new bike lane means ripping up concrete. Tactical urbanism tests changes with temporary materials: paint instead of concrete, planters instead of curbs, pop-up parks instead of permanent ones. If the temporary change works, it becomes permanent. If it doesn't, cleanup costs a fraction of what demolition would. Cities like Barcelona (superblocks) and New York (Times Square pedestrianization) have used tactical urbanism to test ideas that would have been politically impossible to implement permanently without evidence.

Financial regulation traditionally requires full compliance before any product can launch — the equivalent of requiring FDA approval before Phase I trials. Regulatory sandboxes (pioneered by the UK's FCA in 2016) let fintech companies operate under relaxed rules with a limited customer base, generating real-world evidence of risks and benefits. If the product works, regulations are adapted; if it doesn't, the damage is contained. The sandbox IS a graduated rollout for regulation itself.