star-populations

What makes a star INTERESTING, and how the galaxy is biased toward it. Pure — no Babylon, no DOM.

The voxel galaxy's dim stars split in two (GALAXY-DESIGN.md → "Interesting stars"):

• interesting: a system with a planet of HI ≤ 2, meaning a breathable or filterable atmosphere, under 2 g, not an extreme temperature. Generated GLOBALLY, so a habitability search sees all of them.
• boring: everything else, generated only LOCALLY as sky texture.

HI 3 is deliberately NOT interesting. It has no atmosphere requirement (the Moon and Mars are HI 3), and nearly every system with planets has one. Tonio: "Almost any star system with planets … is going to have a ball of rock someone can plant a flag on."

How the bias works

Spectral class sets the odds. Within a class, whether a system scores well is the system's own draw (it does not track planet count), so no mix can GUARANTEE a pass. So:

1. A harness (bin/tune-populations.ts) scores a corpus per spectral cell with the REAL rules and writes the pass rates to population-table.ts (generated).
2. interestingMix weights each cell by its pass rate and boringMix by its failure rate, so candidates are drawn where they are likely to succeed.
3. The galaxy VERIFIES each candidate with bestHIOf and keeps the first that passes, so the populations are exact.

When the rules change

Change planet generation or HI (geological age, weather, radiation…), rerun bun bin/tune-populations.ts, and nothing else is rebuilt. The test fingerprints HI on a fixed probe set and fails until you do. Interesting addresses may then point at different stars. That is accepted (Tonio: "If you change the rules for star system generation expecting things not to change is bizarre.").