Daylight proxy
A geometric availability score, not a daylight simulation: no sky model, no climate file, no lux. It asks whether each habitable room has a usable window wall, facing a usable direction, with open ground in front of it, in a room shallow enough for light to reach the back.
How a room is scored
- Window run — contiguous exterior wall of at least 1.2 m, measured per segment of each side.
- Depth — room depth from that wall against the rule of thumb that useful daylight reaches about 2.5 times the window head height (about 6 m here). A rule of thumb, not a compliance test.
- Obstruction — open ground in front of the window run, saturating at 6 m.
- Orientation — a solar weight from 0.62 (north) to 1.0 (south). This assumes the northern hemisphere.
A room is scored on its best facade, with a bonus for genuine dual aspect; the design score is the mean over habitable rooms. A habitable room with no usable window run fails constraint C10 outright.
By method
| Method | median score | median worst room | designs |
|---|---|---|---|
| Random search | 0.705 | 0.508 | 178 |
| CP-SAT | 0.686 | 0.418 | 1,684 |
| NSGA-II | 0.674 | 0.323 | 10,758 |
| Simulated annealing | 0.669 | 0.413 | 5,108 |
| Random sketch + repair | 0.637 | 0.274 | 7,952 |
| Heuristic | 0.606 | 0.222 | 2,159 |
| Heuristic-seeded NSGA-II | 0.578 | 0.204 | 11,121 |
Density against daylight
Within a brief, denser plans score lower on daylight: mean Spearman ρ -0.36 (-0.45–-0.27) over 58 briefs. Filling more of the envelope buries more rooms behind other rooms. This is the clearest trade-off in the benchmark, and it holds inside the proxy only.