Research note
Moppe aliveness campaign: readout before simulation
Moppe’s aliveness campaign (RFC-0003) and the techniques selected for it.
Finding that set the strategy: the engine computes far more than it shows, so the readout comes before more simulation. Evidence: terrain material bands compared a metre elevation against normalized 0..1 thresholds, so on seed 123 (land 50 m to 490 m) the snow band saturated at the shoreline and every world was a snowfield with the grass and dirt layers unreachable; the merge tree #MTDKDE -style structure) is complete and called only by its test; Waterline contours are extracted and dropped; stream-power channel tangents are produced and never fed back; D-infinity direction and area-flux columns are write-only; WebGPU discards the water-sheet texture entirely.
Selected techniques, by movement.
Ecology from hydrology. Moisture currently blends proximity-to-water with drainage accumulation and has no slope term, so a valley floor and a gully of equal catchment read equally wet: add TWI per FastFlow #NV2YRW noting its SFD/MFD split – channels want single-receiver, vegetation wants diffuse. Forest cover multiplies habitat by a noise mosaic, making noise an equal author; invert so habitat sets the threshold #GBXEP3 TreeStand already runs Deussen’s circle model minus the competition – adding a per-species wet/dry preference and vigor makes riparian galleries and dry-ridge conifer emerge from the loop that already runs.
Motion, the cheapest aliveness in the corpus. HZD #ABD2B8 wind as a three-level hierarchy keyed to object height, distance-to-trunk, and distance-to-branch, driven by a field sampled at object centre; per-object phase; camera tilt; ground hugging; distance-faded animation at the LOD handoff. Moppe has one scalar, one world-space phase, and one spare vertex byte. Responsive grass #PQ68ZH the mechanism worth stealing is persistence – collision strength decaying and suppressing the recovery force – not per-blade physics. Portal 2 #A2QB8L the noise fetch that decorrelates phase is what stops a flow-advected surface pulsing.
Water kinds. The confluence overlap is a property of having junction geometry at all; #MVUJ8Z ’s per-pixel Bezier distance field over a bounding quad removes the geometry and is expected to be cheaper. #AK7NGE puts the lip-versus-cascade distinction in the refined longitudinal profile with inserted basins, which makes it a named fiat threshold in program data – the granularity policy of #AHZGJ5 applied to a rendering decision. The curtain cannot come from the height field #AK7NGE says so); #8SERGP ’s particle sheets advected from a detected line give it, and the split matches the SNAP/SPAN observation in #AHZGJ5 exactly: FallLip as a per-world continuant domain beside Waterline, the sheet as per-frame occurrent geometry. Halftone foam #869NHK at under 3% drives shore swash, river churn, and cascade foot from one clustered mask.
Warnings taken as constraints: sums of sines do not read as ocean #C4AY2M Eulerian texture advection stretches or loses motion either way #92XRH7 analytical erosion’s multigrid can relocate valleys as the age parameter moves #DWXKYQ which would break deep-time scrubbing before it exists.