Swarm
A hundred versions of your massing, all scored the same way. Hand Swarm a parametric model from Assayer or Conjure ("Explore in Swarm →"), pick which sliders are free, and it samples the space (or searches it with NSGA-II — the same engine Assayer runs), scoring every variant live against the same metric registry Scoreboard's columns use: same ids, same units, same stated blind spots. The parallel-coordinates plot draws the whole cloud with the non-dominated front picked out; brush an axis to filter, click any line to rebuild that variant in 3D, send a keeper to Scoreboard. The machine ranks; it never chooses. Client-side, no account.
1 · A model to vary
Swarm cannot invent a design space — it varies one you already made. Send a parametric massing over with Explore in Swarm → on Conjure or Assayer, or take the built-in gable.
What this is not
- Not an optimization study. A small-population NSGA-II on a boxed gene space is under-converged by professional standards; the front is "the best of what we happened to look at", and a different seed will move it. Re-run and see.
- Sample mode is a catalogue, not a search. 200 points in 8 dimensions is 2.1 points per dimension — hence the ≤8-gene cap and the warning above 5.
- Only continuous genes. A typology choice (gable vs flat, courtyard vs bar) is not a gene here — run Swarm twice and compare two clouds.
- Every KPI is exactly as honest as the engine behind it — each axis prints its registry blind-spot line, and a Pareto front built on stylized proxies is a Pareto front of stylized proxies.
- Determinism is not correctness. A seeded run is reproducible; that says nothing about whether it found anything good.
- Where the professional tools take over: Wallacei (generational analytics, phenotype clustering), Octopus, Design Explorer, Galapagos; in industry modeFRONTIER and Optimus. Swarm is the lesson before those.
Method: Deb, Pratap, Agarwal & Meyarivan (2002), A Fast and Elitist Multiobjective Genetic Algorithm: NSGA-II, IEEE Trans. Evol. Comp. 6(2) · McKay, Beckman & Conover (1979) — Latin hypercube sampling · Inselberg (1985), The Plane with Parallel Coordinates.