A Game World That Cannot Lie About Water

SATELLITE TERRAIN — A GAME WORLD THAT CANNOT LIE ABOUT WATER

NASA 30 m elevation → rainfall simulation → a branching river network solved as one system — and the audit that refused our own first attempt · 2026-07 → 2026-08

1 · Background & problem

A strategy game set on real terrain, built in two steps: 1. render the land from public NASA satellite elevation data, 2. earn the water from a rainfall simulation and a branching river network — nothing drawn by hand, and every claim checked by refusal-style audits that stop the build rather than ship a lie.

Problem: those audits caught our own pipeline. Each river was carving its own private copy of the terrain, and merging the copies afterwards left rivers disagreeing with each other — water levels stepping up to 1.4 m where rivers met, sheets standing 2.9 m above the ground that shipped.

2 · Root cause & countermeasures

root cause	countermeasure
each river computed its water level on a private copy of the ground	ONE solver now sets every level for the whole 164-river network in a single in-memory pass — 1.0 second for 25,039 river stations
where a tributary met its parent, two private answers met too (steps up to 1.4 m)	the meeting level is a solver constraint: tributaries sit exactly 1 cm under their parent, by construction, at all 72 junctions
164 separate terrain edits were merged after the fact, deepest cut wins	ONE carve pass cuts the solved network into ONE terrain — 215,726 cells in 34.8 s, zero over-limit cuts
a measured carving bug: river segments cut terrain past their own ends, up to 0.8 m too deep on descending reaches	every terrain cell now takes its depth from the nearest river station and no other — found by the new audit, proven by test before the fix shipped
build order mattered (parents before children), so one failure starved a whole subtree	with levels pre-solved, all 164 rivers build independently in parallel — the first fully complete build the project has produced

BEFORE · every river owned a private ground; the merge reconciled the disagreements after the fact · junction steps to 1.4 m, water hanging above its banks, and no build ever completed all rivers

AFTER · one solve, one carve, one terrain · junction agreement to 1 cm by construction · 164 of 164 rivers built · the strictest water-on-ground audit is now honest enough to argue with a 4 m terrain grid about centimeters

3 · Results — the evidence

In-game world map — the solved river network flowing to the sea — **The world in-game:** the live world map under the one-system build — every channel shown is solved and carved as part of one 164-river network (92 sea mouths), selected from the rainfall simulation, never drawn.

A river junction under the old pipeline — stepped, stacked water sheets — **Before — the same junction, old pipeline:** stacked sheets and stepped edges where three rivers each brought their own private answer for the water level.

The same river junction under the network solve — one continuous confluence — **After — one solved system:** the identical camera. The confluence is one continuous surface; the tributary sits 1 cm under its parent because the solver makes any other answer impossible.

A carved valley that was dry under the old pipeline, now carrying a river — **A dry valley comes alive:** twelve rivers the old per-river build had to abandon (their valleys shipped carved but dry) all solve under the network approach — this one now runs to the sea.

River network map with accepted branching links — **The branching network:** a candidate river split joins only after passing four physical tests (inflow, no cycles, flow share, descent) — from the published GRIT method.

NETSOLVE, GREEN — 164 courses / 25039 stations / 72 junctions solved in 1.0s; max junction step 0.0100m (bar 0.0100m, <= tuck BY CONSTRUCTION); d_max 1.33m (cap 15m)

Verbatim line from the network solver's witness: the whole forest solved in one second, and the largest disagreement between any tributary and its parent is exactly the designed 1 cm tuck — not approximately, structurally.

4A · Resolved: one system, one terrain, complete builds

All 164 rivers solve, carve, and build together · junction agreement is a property of the solver, not a repair pass · a measured sub-meter carving bug was found by the new audit and fixed with a proving test · total water stays governed by the region's real-world cap (projected 3.1% of land against an 8.0% river budget)

4B · Unresolved: the audit is now arguing about centimeters

The strictest check demands water sit within 1 cm of its shipped ground; a 4 m terrain grid can only honestly represent that to a few centimeters, so the check currently refuses the build — the refusal is the system working, and deciding the check's correct form is the next piece of work · a handful of steep coastal sites (66 of 24,528 stations past 1 m) are real defects on a named worklist