Randomized Branch Sampling: Size Software Projects without Wasting Time
summary
Randomized Branch Sampling applies a multi-stage unequal-probability sampler (originally from horticulture) to a backlog tree: count epics, pick epics at random, break sampled epics into stories, pick a story per sampled epic, measure story points/tasks and scale with the Horvitz-Thompson estimator. Applied to 13 anonymized ScrumDo projects with stable teams and coaching, sampling 7–11 samples yields strong agreement with actuals (R^2 ≈ 0.97), Excel tooling included; works only when a cohesive, consistently applied sizing methodology exists.
AI-written summary · about
chapters (8)
- 0:00Start
- 18:34Correlation Coefficient
- 23:58Execution Is More Important than Planning
- 25:36Sizing New Projects
- 26:56Branching Structure
- 27:56Divide the Project Scope into Epics
- 30:17Divide the Project into Epochs
- 31:29Conclusion
related talks
30mSoftware Estimation Made Easy
57mThe Science of not estimating
48mRespect estimates
44mThe Math behind Project Scheduling, Bug Tracking, and Triage
44mAgile metrics for predicting the future
01h 00mEstimation: The Art of the SWAG
46mEstimates and complexity: a tale in 42 and a half minutes
42mEstimates: Is There a Silver Bullet?