🌊 Worldgen field study: measure & tune seas, rivers and waterfalls

[marceld23]

What this is

A data-gathering & tuning issue for world generation hydrology. A recent measured audit (probe harness against the real WorldGenerator, seed 7, sparse global grids) found that the water/river system underdelivers by construction — not by bad luck. We'd love help reproducing across many seeds, extending the measurements, and proposing parameter tunes. Numbers people welcome; this is half analysis, half playtest. Fixes are not implemented yet — this issue collects the evidence and tuning proposals that the fix will be designed from.

What the first audit measured

1. Most green worlds have no sea at all. The sea level formula places the sea at BaseHeight − 0.4·Amplitude, but the terrain height function never swings that far on flat terrain styles — a flats-style world cannot produce a single sea column, arithmetically. Measured % of columns below sea level: jungle 0.00%, swamp 0.00%, varied 0.06%, highland 0.13%, savanna 1.03% … ocean 99.99%.
2. The ocean planet has no rivers — 99.99% water leaves zero eligible river source cells.
3. Waterfalls essentially never generate. The min-drop threshold (4) is checked per 1-block step in the field builder — it needs a 5-block cliff within one block, which FBM terrain doesn't produce. Measured waterfall columns: jungle 0, varied 0, lava 0, highland 52.
4. Rivers are 1-wide, 1–2-deep ditches. ~72% of channel cells are 1 block wide (lava worlds: 100%), and depth 1–2 means they're not even swimmable. The headwater cells (47% of the network on jungle) are discarded by the flow threshold, so rivers begin abruptly mid-slope.

How to help

• Reproduce & broaden: run world generation across many seeds/planet types and report the same stats (columns below sea level, river width histogram, waterfall count). A small C# probe against WorldGeneration/ is the way — happy to share pointers on where to hook in.
• Propose tunes: given the formulas above, what sea-offset / width / threshold values produce believable hydrology per planet type? Concrete numbers with reasoning are the ideal contribution.
• Design constraint to keep in mind: worldgen is deterministic, and changing it alters existing player worlds. Part of the discussion is whether a fix should apply to new worlds only (we've done that before for star-system variance).