Problem
The 2024 report mixed three goals and treated cell means as observations (ANOVA residual DF = 0). The live lab asks one ranking question: does field structure change which searcher is statistically dominant on node expansions, and at what occupancy does A* lose that advantage?
What it is / is not
It is a TypeScript port of the 2024 notebooks: occupancy generators, A*/Weighted A*/Dijkstra/BFS/DFS/Greedy, factorial and LHS designs, two-way ANOVA with replicates. It is not a robotics stack and it does not keep the 2024 CNN-as-result.
Architecture
structure family × occupancy × algorithm × grid n
↓
search on stochastic field
↓
responses: path length L, expansions O, optimality gap
↓
two-way ANOVA with genuine replicates
Evidence
Re-analysis of the 2024 archive (900 rows, residual DF = 891): algorithm main effect is large; structure main effect is larger; algorithm × structure interaction is significant.