nova-worldforge

KO | EN

gitlite — search

#ai-agents #claude #react-native #cli #tensorflow #css #open-source #ai-agent #ios #artificial-intelligence

nova-worldforge

(★ 9)

Terrain generation as simulated geology, validated against real-Earth surveys. Cited process laws, deterministic seed-to-world bakes, pre-registered morphometric gates.

파일 탐색기

ZIP 다운로드

- aster.json
- sindri.json
- TARGET_RANGES.md
- checksums.json
- fetch_dems.py
- README.md
- sites.json
- - coastline_demo.png
  - generated_drainage.png
  - generated_tile.npy
  - metrics.json
  - scenario_exploration.md
  - side_by_side.png
  - water_annotated.png
- - beauty_render.png
  - final_ice_thickness.npy
  - implicit_solver_audit.md
  - manifest.json
  - metrics.json
  - metrics_iter2_overlong_interglacials.json
  - metrics_iter3_weak_ice.json
  - metrics_iter4_single_coast.json
  - nordic_tile.npy
  - scenario_exploration.md
  - side_by_side.png
  - timelapse.mp4
  - water_overlay.png
  - xsection_debug.png
  - xsections_v2_ig2.png
- - coastd_trajectory.md
  - creep_probe.md
  - final_ice_thickness.npy
  - formfactor_probe.md
  - manifest.json
  - metrics.json
  - nordic_regate.md
  - nordic_tile.npy
  - scenario.md
  - side_by_side.png
  - tahiti_scenario.md
  - volcanic_probe.md
  - water_overlay.png
- - coastd_dose_probe.md
  - dispersion_pins.md
  - gpu_cg_criteria.md
  - gpu_groundwork.md
  - hetero_coast_probe.md
  - l12_probe.md
  - l13_fields.png
  - uform_probe.md
- README.md
- - img_01_lauterbrunnen_trough.jpg
  - img_02_trough_shoulders_aerial.jpg
  - img_03_staubbach_hanging_fall.jpg
  - img_04_eiger_nordwand.jpg
  - img_05_cirques_and_aretes.jpg
  - REFPACK.md
- - img_01_dune_crest_slipface.jpg
  - img_02_dune45_shadow_asymmetry.jpg
  - img_03_deadvlei_clay_pan.jpg
  - img_04_dune_network_aerial.jpg
  - img_05_ripple_hierarchy.jpg
  - img_06_pan_mudcracks.jpg
  - REFPACK.md
- - img_01_hornbjarg_ridge.jpg
  - img_02_sea_cliff_stack.jpg
  - img_03_truncated_spurs.jpg
  - img_04_hornvik_trough_bay.jpg
  - img_05_basalt_benches_talus.jpg
  - img_06_turf_slope_fjord.jpg
  - REFPACK.md
- - img_01_petermann_calving_front.jpg
  - img_02_ice_sheet_margin.jpg
  - img_03_melt_pond_speckle.jpg
  - img_04_ice_wedge_polygons.jpg
  - img_05_sorted_stone_circles.jpg
  - img_06_ice_shelf_edge_cliff.jpg
  - REFPACK.md
- - img_01_fringing_reef_band.jpg
  - img_02_twin_bays_radial_valleys.jpg
  - img_03_vegetated_basalt_ridge.jpg
  - img_04_knife_ridges_spires.jpg
  - img_05_barrier_reef_zonation.jpg
  - img_06_reef_flat_motus.jpg
  - REFPACK.md
- - img_01_pahoehoe_sheet_ropes.jpg
  - img_02_pahoehoe_aa_contact.jpg
  - img_03_channel_levees_overflow.jpg
  - img_04_fountain_crust_quench.jpg
  - img_05_cone_flow_field.jpg
  - img_06_flow_lobe_planform.jpg
  - REFPACK.md
- fetch_refpacks.py
- README.md
- planet_datasheet.schema.json
- test_climate.py
- test_coastal.py
- test_coastal_collapse.py
- test_datasheet.py
- test_dispersion.py
- test_drainage.py
- test_fetch.py
- test_glacial.py
- test_glacial_formfactor.py
- test_glacial_implicit.py
- test_glacial_selectivity.py
- test_glacial_speedcoupling.py
- test_heterogeneity.py
- test_heterogeneity_pipeline.py
- test_hillslope.py
- test_hillslope_creep.py
- test_landforms.py
- test_laws.py
- test_mass_wasting.py
- test_mass_wasting_pipeline.py
- test_pipeline_resume.py
- test_pipeline_tahiti.py
- test_raster.py
- test_refpacks.py
- test_snapshots.py
- test_stratigraphy.py
- test_terrain_stats.py
- test_volcanic.py
- measure_dispersion.py
- render_movie.py
- render_relief.py
- render_slice2.py
- run_harness.py
- run_slice1.py
- run_slice2.py
- run_tahiti.py
- - __init__.py
  - climate.py
  - coastal.py
  - fluvial.py
  - glacial.py
  - heterogeneity.py
  - hillslope.py
  - mass_wasting.py
  - pipeline.py
  - pipeline_tahiti.py
  - sim.py
  - stratigraphy.py
  - uplift.py
  - volcanic.py
- - __init__.py
  - coastline.py
  - craters.py
  - dispersion.py
  - drainage.py
  - dunes.py
  - hypsometry.py
  - slope.py
  - spectra.py
  - valleys.py
- __init__.py
- datasheet.py
- raster.py
- synth.py
.gitignore
CONVENTIONS.md
LICENSE
NOTICE.md
pyproject.toml
README.md
REPORT.md
requirements.txt

# CDN으로 사용하기

jsDelivr

jsDelivr는 공개 GitHub 리포지토리를 별도 설정 없이 CDN으로 즉시 서빙합니다. 버전과 파일을 고르면 웹페이지에 바로 붙일 수 있는 링크와 예시 코드가 만들어집니다.

버전

파일

링크

예시

// repository documentation

Was this content helpful?

★ 0 (0 ratings)

**Terrain generation as simulated geology, validated against real-Earth survey data.**

![WorldForge — rendered fjord landscape, raw simulation output](https://raw.githubusercontent.com/Nylem-Labs/nova-worldforge/main/Receipts/slice2/beauty_render.png)

*Raw simulation output (Cycles render, no artist pass): a 26 km fjord peninsula
produced by 6.2 million simulated years of coupled geological process models,
from a single seed.*

---

## Abstract

Most procedural terrain starts from spectral noise and is sculpted until it
looks plausible. WorldForge takes the approach of academic landscape-evolution
models instead: the landscape is the output of coupled, literature-cited
process equations — stream-power river incision, shallow-ice glacier flow,
hillslope talus transport, rocky-coast retreat — integrated over a simulated
geological history. Generation is deterministic (seed → bit-identical world)
and planet-portable: all world facts (gravity, ocean level, stratigraphy)
arrive from a validated **planet datasheet**, so the same law library targets
an earthlike or an icy moon by swapping one JSON file.

The distinguishing discipline is **pre-registered validation**: before the
first generator run, thirteen morphometric target bands were measured from
real survey DEMs (Copernicus GLO-30 — the Hornstrandir peninsula, Iceland,
and volcanic Tahiti) and pinned under version control. Every bake is scored
against them, failures published alongside passes. A negative control —
seeded fractal noise, the family most game terrain is built from — fails
6 of 8 applicable gates, passing only network-topology measures (consistent
with Kirchner 1993 on the statistical weakness of Horton ratios as
discriminators).

**Current state: 7 of 13 gates pass** on the nordic tile (Slice 2, seed 1094),
including first-of-program passes for glacial lake cover and stream-network
structure. Each failing row carries a receipted diagnosis and a scoped, cited
path (see [Limitations](#limitations--open-problems)).

## The recorded history

The pipeline records itself through an observation-only snapshot hook (proven
by test to leave the integration bit-identical): 1,497 frames across the full
history, assembled into a time-lapse — rivers dissect the uplifting plateau,
three glacial pulses carve it, the sea floods the result.

**▶ [`Receipts/slice2/timelapse.mp4`](Receipts/slice2/timelapse.mp4)** — 6.2 Myr in 26 seconds.

![Matched-sun comparison: real Hornstrandir survey, WorldForge full pipeline, rivers-only ablation](https://raw.githubusercontent.com/Nylem-Labs/nova-worldforge/main/Receipts/slice2/side_by_side.png)

*Matched-sun hillshades, 31 m/px. Left: real Hornstrandir (Copernicus GLO-30).
Center: WorldForge, full pipeline. Right: the same seed, fluvial stage only —
what the glacial laws added is the difference between the panels.*

## Method at a glance

| | |
| --- | --- |
| **Domain** | 832 × 832 cells at 31 m/px (≈26 km peninsula tile), float32 state |
| **Simulated span** | 6.2 Myr: 6 Myr fluvial dissection, 3 × 4 kyr glacial pulses, 45 kyr coastal |
| **Fluvial core** | implicit O(n) stream-power incision (Braun & Willett 2013), priority-flood depression routing (Barnes et al. 2014), dynamic emergent shoreline |
| **Glacial solver** | implicit backward-Euler shallow-ice (Hindmarsh & Payne 1996 family), Strang-split exact mass-balance integration, damped Picard iteration, sequential sparse LU — the explicit kernel is retained as a small-scale equivalence oracle |
| **Mass balance** | realistic nourishment (1 m/yr accumulation cap, 5 mm/yr·m gradient, Oerlemans 2001); the mass ledger closes to 4×10⁻¹³ on the committed bake (~10⁻¹⁴ at test-oracle scale) |
| **Glacial erosion** | E = K·(u/u₀)² — velocity-squared selectivity (Herman et al. 2015, *Science*), K from the Hallet et al. 1996 compilation |
| **Determinism** | same (datasheet, seed) → bit-identical tile, across checkpoint/resume, verified byte-level across BLAS thread environments |
| **Verification** | 199-test pytest suite; solver equivalence oracle; independent adversarial audit (findings + fixes: [`Receipts/slice2/implicit_solver_audit.md`](Receipts/slice2/implicit_solver_audit.md)) |
| **Hardware** | one desktop workstation: Intel Core Ultra 9 285K, 64 GB RAM; simulation is CPU-only at background priority; an RTX 3090 serves renders only |
| **Compute cost** | final bake ≈ 11 h wall; full Slice-2 campaign ≈ 60 CPU·h of bakes and instrumented probes over 4 days; beauty render 9 s (OptiX); time-lapse assembly ~25 min |

## Results — the gate table (bake v3, seed 1094)

Bands were pinned from survey data **before** generation and are not adjusted
to fit results; changing one requires a signed decision recorded in
[`Docs/TARGET_RANGES.md`](Docs/TARGET_RANGES.md).

| Morphometric target | This bake | Survey band | | |
| --- | ---: | :---: | :---: | --- |
| Slope, median | 10.3° | 6–13° | ✅ | hillslope steepness distribution |
| Slope, 95th pct | 37.8° | 30–42° | ✅ | capped by lithology's angle of repose |
| Area > 30° | 13.0% | 5–15% | ✅ | share of cliff-grade terrain |
| Horton R_b | 5.25 | 3.5–5.5 | ✅ | stream-network branching |
| Lake area | 0.10% | 0.1–2% | ✅ | glacial overdeepenings hold water |
| Pit fill volume | 0.000 | ≤ 0.005 | ✅ | hydrologic integrity |
| Drainage density | 0.64 | 0.35–0.65 | ✅ | channel length per area |
| Hypsometric integral | 0.237 | 0.30–0.45 | ❌ | elevation mix plateau-weighted |
| Hack exponent | 0.580 | 0.35–0.55 | ❌ | basins slightly elongated |
| Spectral slope β | 3.30 | 3.5–4.3 | ❌ | excess high-frequency relief |
| Coastline dimension | 1.28 | 1.05–1.20 | ❌ | wave stage re-roughens the coast |
| Valley exponent b | 0.42 | 1.15–1.60 | ❌ | V-profiles persist (see below) |
| Glacial–fluvial separation | −0.57 | ≥ 0.25 | ❌ | derived from the row above |

## Limitations & open problems

Stated plainly, with the receipts that established them:

1. **Valley cross-sections remain V-shaped.** The erosion proxy is the
   shallow-ice deformation speed at the local surface gradient; when a valley
   fills brim-deep, the cross-valley gradient signal vanishes and incision
   concentrates at the trough floor. The full V→U transformation requires the
   cross-sectional basal-velocity distribution (Harbor 1992) — a scoped
   follow-up law. Five instrumented experiments ruled out every scenario-level
   alternative first ([`Receipts/slice2/scenario_exploration.md`](Receipts/slice2/scenario_exploration.md)).
2. **Spectral texture too fine.** Largely inherited from the fluvial era;
   the selectivity law moved network structure into band but one glacial era
   does not erase 6 Myr of fine dissection.
3. **Post-glacial coastline re-roughening.** Mid-simulation the coast measures
   in-band (1.05–1.10); the wave/rebound stage undoes some of it — a
   documented interaction between two laws.
4. **Chronology is compressed.** 12 kyr of simulated full-glacial pulses stand
   in for the Quaternary's ~50 glaciations, with the cited erosion-rate
   coefficient carrying the dose linearly (argument recorded in the datasheet
   provenance). The time-lapse is stamped accordingly.
5. **Single-tile domain.** Two hypsometry-family misses are consistent with
   simulating one 26 km tile of a landform class whose statistics span a
   region; multi-tile domains are roadmapped.

## Reproduce

```
python -m venv .venv
.venv\Scripts\python -m pip install -r requirements.txt
.venv\Scripts\python -m pytest                       # 199 tests
.venv\Scripts\python fetch\fetch_dems.py             # survey DEMs (checksummed; rasters not committed)
.venv\Scripts\python tools\run_harness.py            # re-measure the real-Earth bands
.venv\Scripts\python tools\run_slice2.py --movie     # the full bake (hours; checkpoints + resumes itself)
.venv\Scripts\python tools\render_movie.py           # assemble the time-lapse
.venv\Scripts\python tools\render_slice2.py          # the beauty render (GPU)
```

Determinism contract: the same datasheet + seed reproduces the committed tile
bit-for-bit, including across interruption and resume. All heavy compute runs
at below-normal priority by design.

## Repository map

| Path | What |
| --- | --- |
| `worldforge/laws/` | The law library — every equation cited inline; versions in `LAW_VERSIONS` |
| `worldforge/metrics/` | The measurement harness: hypsometry, slope, spectra, drainage, coastline, valleys |
| `schema/` + `datasheets/` | Planet datasheet schema v1; `aster.json` (earthlike), `sindri.json` (icy-moon stub) |
| `fetch/` | Scripted DEM acquisition + SHA-256 manifest (raw rasters gitignored) |
| `Docs/TARGET_RANGES.md` | The **pinned** validation bands, with measurement conventions |
| `Receipts/` | Per-slice receipts: metrics, plates, audits, probe chains, the time-lapse |
| `REPORT.md` | The honest running report — every slice ends with its exact state |
| `tests/` | 199 tests: law physics, solver equivalence, determinism, resume bit-fidelity |

## Key references

Braun & Willett 2013 (*Geomorphology* 180–181) · Glen 1955 (*Proc. R. Soc. A* 228) ·
Cuffey & Paterson, *The Physics of Glaciers* 4e · Mahaffy 1976 (*JGR* 81) ·
Hindmarsh & Payne 1996 (*Ann. Glaciol.* 23) · Oerlemans 2001, *Glaciers and
Climate Change* · Hallet 1979 (*J. Glaciol.* 23) · Hallet, Hunter & Bogen 1996
(*Glob. Planet. Change* 12) · Herman et al. 2015 (*Science* 350) · Harbor 1992
(*GSA Bull.* 104) · Benn, Warren & Mottram 2007 (*Earth-Sci. Rev.* 82) ·
van der Veen 1996 (*Cold Reg. Sci. Tech.* 24) · Sunamura 1992, *Geomorphology
of Rocky Coasts* · Barnes et al. 2014 (*Comput. Geosci.* 62) · Strang 1968
(*SIAM J. Numer. Anal.* 5) · Kirchner 1993 (*Geology* 21) · Perron et al. 2008
(*JGR* 113)

---

## License & provenance

Code and documentation are **MIT** (see [`LICENSE`](LICENSE)). Two content
classes keep their own terms, detailed in [`NOTICE.md`](NOTICE.md): the
reference photographs in `RefPacks/` (third-party Wikimedia Commons imagery,
per-image licenses recorded in each `REFPACK.md`) and survey-derived receipt
imagery — *produced using Copernicus WorldDEM-30 © DLR e.V. 2010-2014 and ©
Airbus Defence and Space GmbH 2014-2018 provided under COPERNICUS by the
European Union and ESA; all rights reserved.*

**A note on vocabulary:** this lab is one crew of a larger, private game
project (Nova). "The charter" is that project's internal program brief for
this lab, and "acceptance gates" are its review checkpoints, signed by the
project maintainer. Slice reports and receipts are kept as written.

---

*Part of Project Nova — a space game where the coastline you see from orbit
was produced by the same processes that produce coastlines. All rendered
terrain imagery is simulated (under a geologically compressed chronology);
the reference photographs in `RefPacks/` and the survey-comparison panels are
real-world data, credited per their records.*

nova-worldforge

파일 탐색기

# CDN으로 사용하기

Was this content helpful?

Recent Feedback

Download README