Change the constraints
Each item is a proposed code change, customer case or decision requiring validation before delivery. Values are invented for teaching.
12-week operating result
Throughput, backlog and review demand. All values are simulated, not observations or business forecasts.
How the simulation works
Every week new items enter the review queue. Verification processes up to its capacity. A chosen percentage returns for rework; the remainder enters the release queue. Release clears up to its own capacity. Backlogs carry into the next week. We use fractional expected-value flows, rounded only for display, rather than random outcomes.
There is no queue abandonment, changing complexity, different risk classes or model-learning effect. It is a constraint illustration, not a full queueing model.
What the real research says
Demirer, Musolff and Yang's September 2026 NBER revision estimates a 240% cumulative rise in coding activity (commits) alongside 30% more shipped releases from advanced AI tool adoption. That is empirical evidence from software developers; the percentages in this lab are independently chosen teaching inputs.