airSpring — Ecological & Agricultural Sciences

Sovereign compute for precision agriculture, irrigation science, and environmental systems. Date: March 24, 2026 Version: 0.10.0 License: AGPL-3.0-or-later

airSpring is the ecological sciences validation study in the ecoPrimals ecosystem. Where hotSpring validates nuclear physics (clean math, f64) and wetSpring validates points in a system (microbiome, mass spectra, PFAS), airSpring validates systems themselves — agricultural fields, soil-plant-atmosphere continua, irrigation networks, and land-water-energy interactions.

Paper benchmarks → Python/R baselines → Real open data → Rust (BarraCuda CPU)
     → GPU (ToadStool shaders, Titan V live) → metalForge (mixed hardware)
     → biomeOS (NUCLEUS atomics, deployment graphs) → Penny Irrigation

Current Status (v0.10.0)

Phase	Status	Key Metric
Phase 0: Paper baselines (Python)	1,284/1,284 PASS	60 papers: FAO-56, soil, IoT, WB, dual Kc, Richards, biochar, yield, CW2D, 8 ET₀ methods, GDD, pedotransfer, ensemble, bias correction, parity, dispatch, Anderson coupling, SCS-CN + Green-Ampt (coupled), VG inverse, full-season WB, MC ET₀ uncertainty, bootstrap/jackknife CI, SPI drought index
Phase 0+: Real data pipeline	15,300 station-days	ET₀ R²=0.97 vs Open-Meteo (100 Michigan stations)
Phase 1: Rust validation	986 lib + 316 integration (1,302 barracuda) + 62 forge = 1,364 total	91 binaries + 146/146 + 32/32 provenance cross-spring benchmarks (NVK zero-output detection: CPU fallback)
Phase 1.5: CPU Benchmark	13,000× atlas-scale	Rust vs Python: 10M ET₀/s, 6.8M field-days/s (34/34 parity)
Phase 2: Cross-validation	75/75 MATCH	Python↔Rust identical (tol=1e-5), Richards + isotherm included
Phase 2.5: Tier B→A GPU	4 ops GPU-first	Hargreaves (op=6), Kc climate (op=7), dual Kc (op=8), sensor cal (op=5) — ToadStool S70+ absorbed
Phase 2.6: Seasonal pipeline	GPU Stages 1-3	ET₀ + Kc + WB GPU dispatch, multi-field gpu_step(), streaming
Phase 2.7: GPU streaming multi-field	57/57 PASS	M fields × N days, Stage 3 GPU per-day, 6.8M field-days/s (Exp 070)
Phase 3: GPU live dispatch	78/78 PASS	Pure GPU workload validation (Exp 055), 0.04% seasonal parity
Phase 3.1: Pure GPU end-to-end	46/46 PASS	All 4 stages on GPU, 19.7× dispatch reduction (Exp 072)
Phase 3.2: Cross-spring rewire	68/68 PASS	BrentGpu VG inverse, RichardsGpu Picard, full provenance (Exp 073)
Phase 3.3: Paper chain validation	79/79 PASS	Full CPU→GPU→metalForge chain for 28 domains (22 GPU, 6 CPU-only) (Exp 074)
Phase 3.4: Upstream GPU lean	20 ops via BatchedElementwiseF64	Ops 0-19 all upstream, local_dispatch retired, PrecisionRoutingAdvice wired (Exp 075+078)
Phase 3.5: NPU edge	AKD1000 live	3 experiments, 95/95 NPU checks, ~48µs inference
Phase 3.7: metalForge live	5 substrates discovered	RTX 4070 + Titan V + AKD1000 + i9-12900K, 27 workloads route
Phase 3.8: Mixed-hardware pipeline	66/66 PASS	7-stage GPU→NPU PCIe bypass, NUCLEUS mesh routing (Exp 076: 60/60)
Phase 3.9: NUCLEUS primal	45 capabilities	airSpring biomeOS primal, 45 capabilities (science + provenance + cross-spring), JSON-RPC
Phase 4.0: Cross-primal pipeline	28/28 PASS	ecology domain, capability.call routing, cross-primal forwarding
Phase 4.1: Full dispatch experiment	51/51 PASS	CPU vs GPU parity across all domains (Exp 064)
Phase 4.2: biomeOS graph experiment	35/35 PASS	Offline ecology pipeline, deployment graph validated (Exp 065)
Phase 4.3: Paper 12 immunological Anderson	4 experiments	Tissue diversity, CytokineBrain, barrier state, cross-species (Exp 066-069)
Phase 4.5: biomeOS composition	4 graphs	Provenance trio, NestGate routing, niche deploy, cross-spring data exchange
Phase 4.6: neuralAPI integration	3 enhancements	Structured metrics, operation dependencies, cost estimates — Pathway Learner ready
Phase 4.7: Penny Irrigation	Vision	Sovereign, consumer hardware
Phase 5.6: Deep debt resolution	All findings resolved	v0.8.4: cast lint evolution, eprintln→tracing, hardcoded primal names→constants, deny.toml wildcards=deny, Blaney-Criddle p→benchmark JSON, primal binary refactored (4 modules), Python provenance standardized (20 scripts), CI expanded (cross-compile, metalForge deny, 11 more validation binaries)
Phase 5.7: Cross-spring absorption	Zero C deps achieved	v0.8.5: ureq→Songbird IPC (Tower Atomic, zero C deps), IpcError+DispatchOutcome biomeOS alignment, #[expect(reason)] migration, zero-panic validation (top 9 binaries), named physical constants (FAO-56, SCS-CN, AMC), dual-format capability discovery, 58→58 warnings (0 new)
Phase 5.8: Deep execution	All 47 binaries zero-panic	v0.8.6: Zero-panic validation expanded from 9→47 binaries, typed compute_dispatch client (toadStool compute.dispatch.submit/result/capabilities), centralized extract_rpc_error(), Python tolerance mirror complete (60 constants), 866 lib tests
Phase 5.9: Ecosystem absorption	Zero hardcoded primals	v0.8.7: primal_names::BIOMEOS constant, swapped TOADSTOOL/BEARDOG docs, discovery.rs hardcoding eliminated, #[allow]→#[expect] migration complete, parse_capabilities 4-format support (S156+), collapsible-if (Edition 2024 let-chains), JSON-RPC proptest fuzz (7 properties), PRIMAL_REGISTRY v0.8.7, 880 lib + 22 property tests
Phase 5.10: Cross-ecosystem absorption	Health probes, circuit breaker	v0.8.8: health probes, circuit breaker, OrExit, thiserror, socket_env_var, structured tracing, IpcError::is_recoverable
Phase 5.11: Cross-ecosystem evolution	All absorbed	v0.8.9: Canonical PRIMAL_NAME/PRIMAL_DOMAIN, OnceLock GPU probe cache, cast module (safe numeric casts), DispatchOutcome<T> library type, coralReef/Squirrel discovery, mul_add() FMA (18 sites), smart refactors (4 monoliths → 19 modules: evapotranspiration, dual_kc, biomeos, validation), composition guidance
Phase 5.12: Audit execution	All findings resolved	v0.9.0: #![deny(cast_*)] library strict, 3 new cast helpers (u32_usize, u64_usize, u64_f64), soil_moisture refactored (672→4 submodules), petalTongue 3-tier discovery, primal_dispatch integration tests (14), benchmark JSON provenance for all hardcoded values, CI lint config via Cargo.toml only, hardcoded socket paths removed
Phase 5.13: Deep audit execution	All debt resolved	v0.10.0: Provenance registry 11→63 baselines, OrExit zero-panic across all 91 binaries, centralized tolerances (R², RMSE, IA, bio-diversity), #[allow]→#[expect] Rust 2024, smart refactors (data/provider→4 modules, evolution_gaps→resolved_issues), hardcoded primal names→primal_names::* constants, data accession IDs, cast helpers (f64_i32, usize_i32), determinism contract documented

Code Quality

Check	Status
cargo test --lib (barracuda)	986 passed, 0 failures
cargo test --tests --all-features (barracuda)	316 passed (integration + doc tests)
cargo test --lib (metalForge)	62 passed, 0 failures
cargo llvm-cov --lib --fail-under-lines 90	90.56% line coverage
cargo clippy (pedantic + nursery, -D warnings)	0 warnings (both crates)
cargo fmt --check	Clean
cargo doc --no-deps	Clean (both crates)
cargo-deny check	Clean (AGPL-3.0-or-later, cargo-deny 0.19)
bench_cross_spring_evolution	146/146 PASS (release, S87 sync)
validate_cross_spring_provenance	32/32 PASS — CPU↔GPU benchmark, 5-spring shader provenance
validate_dispatch_experiment	51/51 PASS — CPU/GPU/batch/absorption/pipeline
validate_biome_graph	35/35 PASS — graph topology, capabilities, offline pipeline

Hardware Validated

Component	Specification	Status
CPU	Intel i9-12900K (16C/24T, AVX2)	Live — all CPU paths
GPU #1	NVIDIA RTX 4070 (12 GB, Vulkan, f64)	Live — wgpu adapter 0
GPU #2	NVIDIA TITAN V (GV100, NVK/Mesa, f64)	Live — 24/24 PASS, BARRACUDA_GPU_ADAPTER=titan
NPU	BrainChip AKD1000 (/dev/akida0)	Live — 95/95 NPU checks
RAM	64 GB DDR5-4800
OS	Pop!_OS 22.04 (kernel 6.17.9)

Evolution Architecture: Write → Absorb → Lean

  airSpring eco:: (CPU, validated against FAO-56 papers)
       │
       ▼
  airSpring gpu:: wrappers (domain-specific batched API)
       │
       ▼
  barracuda::ops/linalg/stats/pde/optimize (GPU dispatch + CPU fallback)
       │
       ▼
  BarraCuda WGSL shaders (f64 canonical, precision per silicon — 767 shaders)
       │
       ▼
  metalForge (mixed CPU + GPU + NPU)
       │
       ▼
  bingoCube/nautilus (evolutionary reservoir, drift detection, NPU export)

airSpring domain code (eco::) is validated against papers, then wrapped by GPU orchestrators (gpu::) that bridge to barracuda primitives. The primitives dispatch to ToadStool WGSL shaders for GPU or fall back to CPU. metalForge stages upstream absorption candidates following the "Write → Absorb → Lean" cycle. bingoCube/nautilus provides evolutionary reservoir computing for time-series prediction (ET₀ forecasting, drift detection) with AKD1000 NPU export.

Cross-Spring Shader Evolution

BarraCuda contains 767+ WGSL shaders (pure math + precision per silicon). airSpring uses 6 shared shader families, contributed 3 upstream fixes, and had all local GPU ops absorbed upstream (Write→Absorb→Lean complete):

Spring	Shaders	What airSpring Gets	What airSpring Gave Back
hotSpring	56	df64 core, pow/exp/log/trig f64, df64_transcendentals — VG, atmospheric	TS-001: pow_f64 fractional exponent fix
wetSpring	25	kriging_f64, fused_map_reduce, moving_window, ridge, diversity metrics	TS-004: reduce buffer N≥1024 fix
neuralSpring	20	nelder_mead, multi_start, ValidationHarness	TS-003: acos precision boundary fix
groundSpring	—	MC ET₀ uncertainty propagation shader	—
airSpring	—	Domain consumer + all 6 local ops absorbed upstream (ops 14-19)	Richards PDE (S40), stats metrics (S64), SCS-CN/Stewart/Makkink/Turc/Hamon/Blaney-Criddle (ops 14-19)

50+ cross-spring absorptions. All metalForge and local GPU ops absorbed. DF64 transcendentals complete (15 functions). PrecisionRoutingAdvice routes f64 dispatch per-hardware (F64Native, F64NativeNoSharedMem, Df64Only, F32Only). Upstream provenance registry (barracuda::shaders::provenance) provides cross-spring evolution tracking. See specs/CROSS_SPRING_EVOLUTION.md.

BarraCuda Integration (25 Tier A + 6 GPU-universal + 3 pipeline)

airSpring Module	BarraCuda Primitive	Op/Shader	Status
gpu::et0	batched_elementwise_f64	op=0	GPU-FIRST
gpu::water_balance	batched_elementwise_f64	op=1	GPU-STEP
gpu::sensor_calibration	batched_elementwise_f64	op=5	Integrated
gpu::hargreaves	HargreavesBatchGpu	dedicated	Integrated
gpu::kc_climate	batched_elementwise_f64	op=7	Integrated
gpu::dual_kc	batched_elementwise_f64	op=8	Integrated
gpu::van_genuchten	batched_elementwise_f64	op=9,10	Integrated (S79)
gpu::thornthwaite	batched_elementwise_f64	op=11	Integrated (S79)
gpu::gdd	batched_elementwise_f64	op=12	Integrated (S79)
gpu::pedotransfer	batched_elementwise_f64	op=13	Integrated (S79)
gpu::kriging	kriging_f64::KrigingF64	dedicated	Integrated
gpu::reduce	fused_map_reduce_f64	dedicated	GPU N≥1024
gpu::stream	moving_window_stats	dedicated	Integrated
gpu::richards	pde::richards::solve_richards	dedicated	Integrated
gpu::isotherm	optimize::nelder_mead + multi_start	dedicated	Integrated
gpu::mc_et0	mc_et0_propagate_f64.wgsl	dedicated	Integrated
gpu::jackknife	JackknifeMeanGpu	dedicated	Integrated (S79)
gpu::bootstrap	BootstrapMeanGpu	dedicated	Integrated (S79)
gpu::diversity	DiversityFusionGpu	dedicated	Integrated (S79)
gpu::stats	linear_regression_f64 + matrix_correlation_f64	dedicated	Integrated
gpu::runoff	batched_elementwise_f64	op=17	GPU-FIRST (v0.7.2, absorbed upstream)
gpu::yield_response	batched_elementwise_f64	op=18	GPU-FIRST (v0.7.2, absorbed upstream)
gpu::simple_et0	batched_elementwise_f64	ops 14-16, 19	GPU-FIRST (v0.7.2, absorbed upstream)
eco::correction::fit_ridge	linalg::ridge::ridge_regression	CPU	Integrated
eco::richards::inverse_vg_h	optimize::brent	CPU	Integrated
eco::diversity	stats::diversity	CPU	Leaning
eco::drought_index	special::gamma::ln_gamma	CPU	New (v0.7.4)
gpu::seasonal_pipeline	Chains ops 0→7→1→yield	fused	CPU + GpuPipelined
gpu::atlas_stream	UnidirectionalPipeline	streaming	CPU chained

Also wired: validation::ValidationHarness (neuralSpring), stats::pearson, spearman, bootstrap_ci, stats::metrics re-exports (airSpring→upstream S64).

25 Tier A GPU modules (ops 0-19 all upstream via BatchedElementwiseF64) + jackknife/bootstrap/diversity uncertainty stack. local_dispatch retired (v0.7.2 Write→Absorb→Lean complete). PrecisionRoutingAdvice wired (v0.7.3). Upstream provenance registry integrated. See barracuda/src/gpu/evolution_gaps.rs and barracuda/EVOLUTION_READINESS.md for the full roadmap.

CPU Benchmarks: Rust vs Python (14.3× geometric mean speedup, 24/24 parity)

Run cargo run --release --bin bench_cpu_vs_python. 24 algorithms, same f64 precision, same inputs, same outputs. 14.3× geometric mean includes the seasonal pipeline (ET₀→Kc→WB→Yield chain) and Paper 12 immunological modules. Individual algorithm speedups range from 5× to 124× for pure compute, with 21/21 CPU-GPU parity modules validated via validate_cpu_gpu_comprehensive.

Quick Start

# Python baselines (Phase 0)
pip install -r control/requirements.txt
python scripts/download_open_meteo.py --all-stations --growing-season 2023
bash run_all_baselines.sh

# Rust validation (Phase 1)
cd barracuda && cargo test
cargo run --release --bin validate_et0

# Benchmarks
cargo run --release --bin bench_cpu_vs_python
cargo run --release --bin bench_airspring_gpu

No institutional access required. Zero synthetic data in the default pipeline.

Data Strategy

Layer	Source	Purpose	API Key
Benchmark	FAO-56 tables, Dong 2020/2024	Ground truth (digitized)	None
Open Data	Open-Meteo archive	80+ yr Michigan weather	None (free)
Open Data	OpenWeatherMap	Current + 5-day forecast	testing-secrets/
Open Data	NOAA CDO (GHCND)	Historical daily	testing-secrets/
Open Data	USDA Web Soil Survey	Soil properties	None
Fallback	Synthetic generation	Only if API unreachable	N/A

Research Context

Track 1: Precision Irrigation & Soil Science

Younsuk Dong, PhD — Assistant Professor & Irrigation Specialist, Biosystems and Agricultural Engineering, Michigan State University.

The Problem: Agricultural irrigation consumes ~70% of global freshwater withdrawals. Current precision irrigation relies on proprietary sensor systems ($500-$5000/field) and vendor-locked software.

Computational Methods: FAO-56 PM ET₀, Topp equation soil calibration, IoT field monitoring, FAO-56 Ch 8 water balance scheduling, FAO-56 Ch 7 dual Kc with cover crops and no-till mulch.

Track 2: Environmental Systems & Land Treatment

Richards equation (unsaturated flow — open-source alternative to HYDRUS), biochar adsorption isotherms (Langmuir/Freundlich), long-term water balance (60-year reconstruction via ERA5).

Key Publications

• Dong et al. (2020) "Soil moisture sensor evaluation in Michigan soils" Agriculture 10(12), 598
• Dong & Hansen (2023) "Affordable weighing lysimeter design" Smart Ag Tech 4, 100147
• Dong et al. (2024) "In-field IoT for precision irrigation" Frontiers in Water 6, 1353597
• Ali, Dong & Lavely (2024) "Irrigation scheduling vs yield" Ag Water Mgmt 306, 109148
• Dong et al. (2019) "Land-based wastewater modeling using HYDRUS CW2D" J. SWBE 5(4)
• Kumari, Dong & Safferman (2025) "Phosphorus adsorption using biochar" Applied Water Sci 15(7)

Directory Structure

airSpring/
├── control/                     # Phase 0: Python baselines (1284/1284, 47 scripts)
│   ├── fao56/ ... drought_index/  # 47 paper controls
│   ├── coupled_runoff_infiltration/   # Coupled SCS-CN + Green-Ampt (292/292)
│   ├── mc_et0/                  # Monte Carlo ET₀ uncertainty propagation (47/47)
│   ├── bootstrap_jackknife/     # Bootstrap & Jackknife CI (20/20)
│   ├── drought_index/           # SPI drought index (20/20)
│   └── requirements.txt
├── barracuda/                   # Phase 1+3: Rust validation + GPU dispatch (943 lib + 316 integration/doc = 1,259 tests, 91 binaries, barraCuda 0.3.7 / wgpu 28, Edition 2024)
│   ├── src/
│   │   ├── biomeos/                # biomeOS socket resolution + primal discovery (3 sub-modules)
│   │   ├── eco/                 # Domain modules (22 validated, 8 ET₀ + runoff + infiltration + VG + Anderson + tissue + cytokine + drought_index)
│   │   ├── gpu/                 # ToadStool/BarraCuda GPU bridge (25 Tier A, ops 0-19 upstream + BrentGpu + RichardsGpu)
│   │   ├── data/                # Data provider abstraction (HttpProvider, BiomeosProvider, NestGateProvider)
│   │   ├── ipc/                 # Inter-primal communication (provenance trio integration)
│   │   ├── nautilus.rs          # bingoCube/nautilus evolutionary reservoir (NautilusBrain v0.1.0)
│   │   ├── rpc/                 # JSON-RPC 2.0 IPC (error.rs, transport.rs, mod.rs)
│   │   ├── npu.rs               # BrainChip AKD1000 NPU (feature-gated)
│   │   ├── tolerances/          # Domain-specific validation tolerances (58 named, 4 submodules)
│   │   └── bin/                 # validate_*, bench_*, airspring_primal (91 declared)
│   ├── tests/                   # Integration + property tests (15 files + common/)
│   └── Cargo.toml               # v0.10.0 (barraCuda 0.3.7, wgpu 28)
├── niches/                      # BYOB niche definitions (airspring-ecology.yaml)
├── metalForge/                  # Mixed hardware dispatch (CPU+GPU+NPU)
│   ├── deploy/                  # biomeOS deployment graphs (airspring_deploy.toml)
│   └── forge/                   # airspring-forge (62 tests, 6 binaries, live hardware probe)
├── specs/                       # Specifications and requirements
│   ├── PAPER_REVIEW_QUEUE.md    # Paper reproduction queue (87 experiments)
│   ├── BARRACUDA_REQUIREMENTS.md# GPU + NPU kernel requirements
│   └── CROSS_SPRING_EVOLUTION.md # Cross-spring shader provenance (S87)
├── whitePaper/                  # Methodology and study documentation
│   └── baseCamp/                # Per-faculty research briefings + baseCamp extensions
├── experiments/                 # Experiment protocols and results (87 experiments)
├── wateringHole/                # Spring-local handoffs to ToadStool/BarraCuda
│   └── handoffs/                # Versioned handoffs (V010 current)
├── graphs/                      # biomeOS deployment graphs (eco pipeline, provenance pipeline, niche deploy, cross-primal)
├── CHANGELOG.md                 # Keep-a-Changelog versioned history
├── CONTROL_EXPERIMENT_STATUS.md # Detailed experiment log
└── LICENSE                      # AGPL-3.0-or-later

Relationship to Other Springs

	hotSpring	wetSpring	airSpring
Domain	Nuclear physics	Life/analytical science	Ecological systems
Scale	Nucleus (fm)	Cell/molecule	Field/watershed
Validation	Binding energies	Organism/PFAS ID	ET₀, yield, water balance
Baseline	Python/scipy	Galaxy/QIIME2/asari	FAO-56/Python/R
Data	AME2020 (IAEA)	GenBank, MassBank	Open-Meteo, NOAA, FAO
GPU layer	ToadStool (wgpu)	ToadStool	ToadStool
Success metric	chi² match	Same taxonomy/PFAS	Same ET₀/yield
Ultimate goal	Sovereign nuclear	Penny water monitoring	Penny irrigation

Document Index

Document	Purpose
CHANGELOG.md	Versioned change history
CONTROL_EXPERIMENT_STATUS.md	Detailed experiment results (87 experiments)
barracuda/EVOLUTION_READINESS.md	Tier A/B/C GPU evolution, absorbed/stays-local
metalForge/ABSORPTION_MANIFEST.md	6/6 modules absorbed upstream (S64+S66), 27 workloads
metalForge/forge/	Mixed hardware dispatch: live probe + capability routing
specs/CROSS_SPRING_EVOLUTION.md	Cross-spring shader provenance (S87)
specs/PAPER_REVIEW_QUEUE.md	Paper reproduction queue (87 experiments)
whitePaper/baseCamp/README.md	Faculty research briefings + baseCamp extensions
wateringHole/handoffs/	ToadStool/BarraCuda handoffs (V010 current)

License

AGPL-3.0-or-later

---

March 24, 2026 — v0.10.0. Deep audit execution complete. cargo-deny 0.19 evolved (SPDX AGPL-3.0-or-later, CC0-1.0 allowance, blake3 cc wrapper, version-pinned path deps). Coverage expanded: validation/json, ipc/provenance, ipc/compute_dispatch, niche — 90.56% line coverage (fail-under-lines 90 passes). evolution_gaps.rs synced to v0.10.0 / barraCuda 0.3.7. #[expect]→#[allow] for shared test infra (tests/common/mod.rs). const assert for compile-time tolerance contract. 986 lib + 316 integration/doc = 1,302 barracuda tests; 62 forge; 1,364 total tests. 0 failures. Clippy pedantic+nursery zero warnings. Zero unsafe. Zero C deps. Zero hardcoded primals. AGPL-3.0-or-later.