groundSpring — The Dirty Differences

Date: March 24, 2026 | License: AGPL-3.0-or-later | MSRV: Rust 1.87 (2024 edition) Status: V124 — 41 modules, 35 experiments, 1050+ tests, 395/395 validation checks (340 core + 55 NUCLEUS) + 140 metalForge checks + 29 validator exit-code integration tests, 110 active barraCuda delegations (67 CPU + 43 GPU) — synced against barraCuda v0.3.7, toadStool S158+, coralReef Iteration 55+. Three-tier parity proven: 29/29 validation binaries PASS at all three tiers. cargo deny check PASS, cargo clippy --workspace --all-features zero warnings (pedantic+nursery), ≥92% library line coverage, cargo doc -D warnings clean. V124: Deep debt resolution + tolerance hardening — eps module promoted to pub with new SAFE_DIV_STRICT constant, all bare float literals replaced with named tol::/eps:: constants throughout library and validation code, NDJSON sink hardened against JSON injection (RFC 8259 escaping), validate_all exit-code disambiguation (hardware-unavailable vs real failure), NestGate capability-based discovery (no hardcoded addresses), UID discovery chain extended (4-tier, no hardcoded fallback), 29 validator exit-code integration tests, CI validate-all job, 5 pre-existing clippy fixes (cast_possible_truncation stale expects, unnecessary casts). V123: Cross-ecosystem absorption + provenance. V122: Cast evolution + module extraction. V121: Deep debt + ecosystem absorption.

The gap between what models predict and what instruments measure.

groundSpring is the reality layer in the ecoPrimals ecosystem. Where other springs validate clean science — hotSpring (nuclear math), airSpring (FAO-56 equations), wetSpring (taxonomy pipelines) — groundSpring lives in the space where those models meet the physical world.

Core question: "How do things actually look, and why is it different from what we expected?"

Clean models (other springs) → Noisy measurements (groundSpring) → Adapted models (neuralSpring)

The Five Pillars

1. Signal vs Noise — Distinguishing real phenomena from measurement artifacts. Sensor drift, calibration error, environmental interference. When airSpring's soil moisture sensor reads 0.32 instead of 0.30, is that real heterogeneity or instrument error?

2. Inverse Problems — From observations back to causes. Where did an earthquake start, from its emanations? What is a star's composition, from its light frequencies? What contaminant entered the watershed, from downstream sensor readings?

3. Sensing Systems — The physics of measurement itself. How do different instruments see the same phenomenon differently? A thermometer, a satellite, and a reanalysis model all "measure" temperature differently. Color and size have different meaning depending on the detector.

4. Temporal Dynamics — How systems drift over time. Sensor degradation. Seasonal baselines. Long-term climate trends vs short-term weather noise. The geological clock vs the agricultural clock vs the astronomical clock.

5. Spatial Propagation — How signals travel through media. Seismic waves through rock. Light through atmosphere (extinction, redshift). Moisture through soil. Contaminants through aquifers. The medium distorts the message.

Current Status

Experiment	Domain	Phase 0 (Python)	Phase 1 (Rust)	Key Question
001: Sensor Noise	Agricultural	32/32 PASS	36/36 PASS	Bias vs variance in soil moisture sensors
002: Observation Gap	Meteorological	PASS/SKIP	13/13 PASS	Reanalysis model vs station readings
003: Error Propagation	Agricultural	PASS	15/15 PASS	How sensor noise becomes ET₀ uncertainty
004: Sequencing Noise	Biological	PASS	15/15 PASS	Taxonomic reliability vs sequencing depth
005: Seismic Waves	Geological	PASS	9/9 PASS	Source localization from noisy arrivals
006: Signal Specificity	Biological	12/12 PASS	12/12 PASS	c-di-GMP signal vs noise in enzyme network
007: RAWR Resampling	Statistics	11/11 PASS	11/11 PASS	Bayesian bootstrap vs naive bootstrap
008: Anderson Localization	Mathematics	8/8 PASS	8/8 PASS	Lyapunov exponents in disordered media
009: Almost-Mathieu Quasiperiodic	Mathematics	PASS	8/8 PASS	Aubry-André metal-insulator transition
010: Bistable Phenotypic Switching	Biological	PASS	10/10 PASS	Fernandez 2020 PNAS bifurcation
011: Multi-Signal QS Integration	Biological	PASS	9/9 PASS	Srivastava 2011 dual-signal integration
012: Spin Chain Transport	Mathematics	18/18 PASS	18/18 PASS	Kachkovskiy 2016 wavepacket MSD, transport exponent
013: Resampling Convergence	Statistics	8/8 PASS	8/8 PASS	Lee & Liu 2024 bootstrap convergence
014: Drift vs Selection	Biological	7/7 PASS	7/7 PASS	R. Anderson 2022 Wright-Fisher, Kimura fixation
015: Uncertainty Bridge	Cross-domain	8/8 PASS	8/8 PASS	Sensor noise → Anderson ξ propagation
016: Rare Biosphere	Biological	11/11 PASS	12/12 PASS	Sequencing depth determines rare taxa signal boundary
017: Quasispecies Threshold	Evolutionary	9/9 PASS	6/6 PASS	Eigen's error threshold predicts mutation-driven information collapse
018: Band Edge Structure	Mathematical	8/8 PASS	10/10 PASS	Transfer matrix reproduces tight-binding band-gap structure
019: Jackknife Error Estimation	Inverse Problems & Spectral Reconstruction	9/9 PASS	9/9 PASS	Bazavov 2025 Phys Rev D 111, 094508 — jackknife variance, bias correction
020: Freeze-Out Inverse Problem	Inverse Problems & Spectral Reconstruction	8/8 PASS	8/8 PASS	Bazavov 2016 Phys Rev D 93, 014512 — freeze-out temperature from hadron yields
021: Spectral Function Reconstruction	Inverse Problems & Spectral Reconstruction	8/8 PASS	8/8 PASS	Bazavov 2025 arXiv 2501.12259 — spectral reconstruction from correlators
022: ET₀ → Anderson Propagation	Cross-spring (FAO-56 + Anderson)	7/7 PASS	7/7 PASS	Humidity-dominated ET₀ error → localization length CV
023: No-Till vs Tilled Sampling	Cross-spring (microbiome + soil)	7/7 PASS	7/7 PASS	Saturation depth by soil management regime
024: Aggregate Stability Noise	Cross-spring (soil physics)	8/8 PASS	8/8 PASS	WSA measurement precision vs Anderson regime discrimination
025: f32 vs f64 Precision Drift	WDM MD	7/7 PASS	7/7 PASS	Green-Kubo f32 accumulation bias
026: System-size Convergence	WDM MD	7/7 PASS	7/7 PASS	Transport coefficient finite-size extrapolation
027: GPU Vendor Parity	WDM MD	7/7 PASS	7/7 PASS	Cross-vendor transport coefficient agreement
028: NPU Anderson Regime	Hardware (NPU)	7/7 PASS	9/9 PASS	Anderson regime classification on AKD1000 via int8 DMA
029: Real GHCND ET₀	Cross-spring (NOAA)	—	6/6 PASS	Hargreaves vs Penman-Monteith on real/synthetic weather via NestGate
030: Real NCBI 16S	Biological (NCBI)	—	9/9 PASS	Rare biosphere detection on real/synthetic NCBI 16S metagenomes
031: NUCLEUS Stack	Infrastructure	—	28/28 PASS	Full NUCLEUS primal validation: Tower + Node + Squirrel + Nest
032: IRIS Seismic	Geological (IRIS)	—	12/12 PASS	IRIS FDSN station geometry + travel times via NestGate
033: Tissue Anderson	Immunological (Paper 12)	—	29/29 PASS	Cytokine Anderson lattice + geometry-aware drug scoring
034: ET₀ Methods	Agricultural (FAO-56)	15/15 PASS	19/19 PASS	5-method ET₀ cross-validation: PM, Hargreaves, Makkink, Turc, Hamon

Phase 1 total: 395/395 PASS across 34 validation binaries (340 core + 55 NUCLEUS via --features biomeos). All public APIs return Result — zero panicking entry points.

Library Modules

Module	Purpose	GPU Tier
stats::agreement	RMSE, MAE, MBE, NSE, R², IA, hit rate (R²/NSE deduplicated via shared coefficient_of_efficiency)	GPU dispatched (rmse, mbe via FusedMapReduceF64/SumReduceF64) + CPU delegated
stats::metrics	mean, std_dev, sample_std_dev, percentile	GPU dispatched (mean via SumReduceF64, std_dev via VarianceReduceF64) + CPU delegated
stats::correlation	Pearson/Spearman correlation, covariance	GPU dispatched (pearson_r via CorrelationF64, covariance via CovarianceF64) + CPU delegated
stats::distributions	norm_cdf, norm_ppf, χ²	3 CPU delegated
stats::regression	Linear, quadratic, exponential, logarithmic fits	4 CPU delegated
decompose	Bias-variance decomposition, noise floor	CPU-only (scalar)
fao56	FAO-56 Penman-Monteith equation chain	Absorbed (barracuda Op::Fao56Et0) + GPU batch (BatchedElementwiseF64)
prng	Xorshift64 PRNG, Box-Muller normal	B (DefaultRng aligned)
rarefaction	Multinomial sampling, Shannon/Simpson diversity, Bray-Curtis, evenness, analytical rarefaction	C (WGSL production ready)
seismic	Haversine, travel time, grid-search inversion	GPU-ready (V31 dispatch)
gillespie	Gillespie SSA for stochastic chemical kinetics	GPU dispatched (batch via GillespieGpu)
bootstrap	Bootstrap (mean/median/std) + RAWR confidence intervals	A Lean (barracuda::stats)
anderson	Anderson localization, Lyapunov exponents, analytical ξ(W,E), 2D/3D eigenvalues, disorder sweep, spectral diagnostics, empirical spectral density, Marchenko-Pastur bounds, transition detection	A Lean (barracuda::spectral + special + stats::spectral_density + ops::peak_detect_f64)
almost_mathieu	Almost-Mathieu quasiperiodic localization, level spacing	A Lean (barracuda::spectral)
linalg	Tridiag eigensolver (implicit QL with Wilkinson shifts) — shared by transport + band_structure	B (adapt)
transport	Wavepacket MSD, transport exponent (re-exports linalg::tridiag_eigh for compat)	B (adapt)
error	Typed input validation errors (InputError: LengthMismatch, InsufficientData, OutOfRange)	N/A
drift	Wright-Fisher fixation, Kimura fixation probability, neutral diversity trajectory	CPU delegated (kimura_fixation_prob S70+) + GPU batch (WrightFisherGpu)
cast	Centralized numeric casts with documented safety	N/A
kinetics	Hill + Monod kinetics (shared bistable + multi-signal)	A Lean (barracuda::stats::hill, monod)
validate	Generic Write harness (hotSpring pattern)	N/A
rare_biosphere	Chao1, detection power/threshold, abundance-occupancy, singleton fraction	GPU-ready (V31 dispatch)
quasispecies	Eigen error threshold, master frequency, Wright-Fisher mutation simulation	GPU-ready (V31 dispatch)
band_structure	Transfer matrix, band edge detection, count bands, periodic Hamiltonian	GPU-ready (V31 dispatch)
jackknife	Jackknife variance, bias correction, leave-one-out resampling	CPU delegated (jackknife_mean_variance S70+)
freeze_out	Freeze-out temperature inversion, hadron yield fitting	GPU-ready (V31 dispatch)
spectral_recon	Spectral function reconstruction from Euclidean correlators	GPU delegated (tikhonov_solve)
biomeos	biomeOS Neural API client: JSON-RPC 2.0, capability routing, UDS server, NestGate storage (behind biomeos feature)	N/A
dispatch	JSON-RPC method dispatch: measurement.* semantic routing to library functions (behind biomeos feature)	N/A
provenance	Provenance Trio lifecycle: session create, dehydrate, attribute via capability calls (behind biomeos feature)	N/A
nestgate	NestGate data pipeline: NCBI/NOAA providers, provenance key schemas, cache-through (behind biomeos feature)	N/A
esn	Echo State Network regime classification: EsnClassifier (barracuda-gpu), rule-based classify_by_spacing_ratio, spectral_features	GPU dispatched (barracuda-gpu ESN) + CPU rule-based
lanczos	Sparse eigensolver for 2D/3D Anderson: sparse_eigenvalues, eigenvalues_from_csr (barracuda-gpu only)	GPU dispatched (barracuda spectral Lanczos)
npu	NPU integration for Akida neuromorphic inference (behind npu feature)	NPU (AKD1000)
niche	Self-knowledge module: capabilities, dependencies, cost estimates, feature gates, consumed capabilities (airSpring pattern)	N/A
primal_names	Centralized primal name constants for socket paths, discovery, env checks (wetSpring V119 pattern); zero hardcoded strings in production	N/A
groundspring-forge	Hardware discovery, cross-substrate dispatch, PCIe topology, multi-stage pipeline, NUCLEUS atomics, remote NUCLEUS discovery (26 workloads, 5+ substrates, 120 tests)	metalForge crate

Quick Start

Rust Phase 1

cargo test --workspace                         # 1020+ tests, all PASS
cargo test --workspace --all-features          # all feature paths tested
cargo test --workspace --features biomeos      # NUCLEUS client active
cargo test --workspace --features barracuda-gpu # GPU dispatch active
cargo clippy --workspace --all-targets -- -D warnings -W clippy::pedantic -W clippy::nursery
cargo fmt --check                              # clean

# Barracuda-delegated mode (validates cross-spring math)
cargo test --workspace --features barracuda
cargo test --workspace --features barracuda-gpu

# NPU mode (BrainChip AKD1000)
cargo test --workspace --features npu          # npu module + Exp 028

# metalForge live hardware binaries
cargo run --bin validate_metalforge_inventory
cargo run --bin validate_metalforge_gpu
cargo run --bin validate_metalforge_cross_substrate
cargo run --bin validate_metalforge_titan_v

# Run all validation binaries at once (meta-binary)
cargo run --bin validate_all

# Individual validation binaries (hotSpring pattern: exit 0 = pass, exit 1 = fail)
cargo run --bin validate_decompose
cargo run --bin validate_rarefaction
cargo run --bin validate_seismic
cargo run --bin validate_weather
cargo run --bin validate_fao56
cargo run --bin validate_signal_specificity
cargo run --bin validate_rawr
cargo run --bin validate_anderson
cargo run --bin validate_quasiperiodic
cargo run --bin validate_bistable
cargo run --bin validate_multisignal
cargo run --bin validate_transport
cargo run --bin validate_resampling_conv
cargo run --bin validate_drift
cargo run --bin validate_uncertainty_bridge
cargo run --bin validate_rare_biosphere
cargo run --bin validate_quasispecies
cargo run --bin validate_band_edge
cargo run --bin validate_jackknife
cargo run --bin validate_freeze_out
cargo run --bin validate_spectral_recon
cargo run --bin validate_npu_anderson
cargo run --bin validate_et0_anderson
cargo run --bin validate_notill_sampling
cargo run --bin validate_aggregate_stability
cargo run --bin validate_precision_drift
cargo run --bin validate_size_convergence
cargo run --bin validate_vendor_parity
cargo run --bin validate_tissue_anderson
cargo run --bin validate_et0_methods

# NUCLEUS / biomeOS validation (requires biomeos feature, NUCLEUS optional)
cargo run --features biomeos --bin validate_real_ghcnd_et0
cargo run --features biomeos --bin validate_real_ncbi_16s
cargo run --features biomeos --bin validate_nucleus_stack
cargo run --features biomeos --bin validate_iris_seismic

Python Phase 0

pip install -e ".[dev]"
python3 -m pytest tests/ -v       # 29 experiments (287 tests)
ruff check control/ tests/        # zero errors
mypy control/ tests/              # zero errors

Test Coverage

cargo llvm-cov --workspace --lib    # ≥92% library line coverage (target 90%)

Performance: Rust vs Python

Median of 3 trials across all 28 experiments (Feb 27, 2026). Generate full data: python3 scripts/bench_rust_vs_python.py.

Experiment	Python (s)	Rust (s)	Speedup
Exp 001: Sensor Noise	0.38	0.07	5.3×
Exp 002: Observation Gap	0.27	0.08	3.6×
Exp 003: Error Propagation	0.34	0.08	4.4×
Exp 004: Sequencing Noise	0.14	0.09	1.5×
Exp 005: Seismic Inversion	7.42	0.14	53.5×
Exp 006: Signal Specificity	26.51	0.86	31.0×
Exp 007: RAWR Resampling	4.54	0.64	7.1×
Exp 008: Anderson Localization	21.96	0.77	28.6×
Exp 009: Quasiperiodic	0.65	11.32 *	0.1×
Exp 010: Bistable Switching	3.26	0.18	18.1×
Exp 011: Multi-Signal QS	4.25	0.10	44.7×
Exp 012: Spin Chain Transport	0.92	0.31	3.0×
Exp 013: Resampling Convergence	1.36	0.13	10.4×
Exp 014: Drift vs Selection	0.42	1.14	0.4×
Exp 015: Uncertainty Bridge	1.32	0.12	11.1×
Exp 016: Rare Biosphere	0.38	0.20	1.9×
Exp 017: Quasispecies Threshold	0.12	0.09	1.3×
Exp 018: Band Edge Structure	0.23	0.11	2.1×
Exp 019: Jackknife Estimation	0.12	0.07	1.7×
Exp 020: Freeze-Out Inverse	0.36	0.07	5.1×
Exp 021: Spectral Recon	0.12	0.07	1.7×
Exp 022: ET₀ Anderson	0.87	0.10	8.6×
Exp 023: No-Till Sampling	0.11	0.09	1.3×
Exp 024: Aggregate Stability	0.14	0.09	1.6×
Exp 025: Precision Drift	27.93	3.18	8.8×
Exp 026: Size Convergence	0.12	0.07	1.6×
Exp 027: Vendor Parity	0.14	0.12	1.1×
Exp 028: NPU Anderson	0.12	0.08	1.5×
Total	104.49	20.35	5.1×
Total (excl. LAPACK-bound)	103.84	9.04	11.5×

* Exp 009/014: Rust custom QR/Wright-Fisher vs NumPy LAPACK/SciPy. Barracuda-gpu (Sturm tridiag from hotSpring S26) closes the gap: 47.7× speedup for Exp 009.

Mathematical parity: 29/29 PROVEN — both languages validate against the same shared benchmark JSONs. Generate report: python3 scripts/parity_report.py.

Run benchmarks: python3 scripts/bench_rust_vs_python.py Run parity report: python3 scripts/parity_report.py

BarraCUDA Delegation Performance

Validation Binary Benchmark (29 binaries, release mode, March 8 2026)

Mode	Wall time (s)	Δ vs local
Local (no features)	12.5	baseline
BarraCUDA CPU	18.4	+47% (dispatch overhead on small workloads)
BarraCUDA GPU	9.9	−21% (1.27× faster)

Workspace Test Benchmark (1020+ tests, release mode)

Mode	Wall time (s)	Δ vs local
Local (no features)	50.3	baseline
BarraCUDA CPU	29.0	−42% (1.73× faster)

110 active delegations (67 CPU + 43 GPU). Cross-spring evolution powers this: hotSpring precision shaders (DF64 core, Sturm tridiag — 47.7× for Exp 009), wetSpring bio shaders (Gillespie SSA, diversity fusion, Bray-Curtis), neuralSpring stats (chi-squared, KL divergence, matrix correlation), airSpring hydrology (seasonal pipeline, Hargreaves ET₀), and groundSpring spectral shaders (Anderson Lyapunov, uncertainty propagation). PrecisionRoutingAdvice + runtime f64 smoke test (V97) guards all 21 GPU dispatch paths via get_device_f64_safe() — hardware-aware routing with empirical f64 reduction verification.

Cross-Spring Shader Evolution

Every spring contributes shaders that benefit the entire ecosystem through barraCuda:

hotSpring (precision)  ─────┐
  df64_core, Sturm tridiag, │    All absorbed into barraCuda v0.3.7
  stress_virial, CG kernels ├──► 784 WGSL shaders, f64-canonical
                             │    with f16/f32/f64/Df64 per hardware
wetSpring (bio)  ────────────┤
  smith_waterman, gillespie, │    toadStool S158+ routes hardware
  fused_map_reduce, HMM      │    coralReef compiles to native GPU binary
                             │
neuralSpring (ML)  ──────────┤    groundSpring consumes 110 ops:
  chi_squared, KL_divergence,│    67 CPU delegated, 43 GPU dispatched
  matrix_correlation, ESN    │
                             │
airSpring (hydrology)  ──────┤
  hargreaves, seasonal_pipe, │
  moving_window, Brent root  │
                             │
groundSpring (spectral)  ────┘
  anderson_lyapunov, welford,
  chi_squared → ALL springs
  f64 bug → PrecisionRoutingAdvice

Evolution Path

Phase 0 (Python)  →  Phase 1 (Rust)  →  Phase 2 (GPU)  →  Phase 3 (Hardware)  →  Phase 4 (NUCLEUS)
  NumPy/SciPy         Pure safe Rust     BarraCUDA/ToadStool   metalForge dispatch    biomeOS Neural API
  ✓ Complete          ✓ 395/395 PASS     ◐ 110 active          30 workloads           Tower+Node+Squirrel
  11.5× slower        35/35 experiments    (67+43)              24 GPU + 2 NPU + 2 CPU-only         NestGate data pipes
                      1020+ workspace tests                      PCIe topology          NUCLEUS atomics
                                                                Pipeline dispatch      Sovereign degradation

  Write locally    →  Hand off          →  Lean on upstream   →  Cross-substrate     →  Primal orchestration
  (metalForge)       (wateringHole/)       (barracuda ops)       (metalForge forge)    (biomeOS graphs)

Lean progress: 110 functions delegate to barracuda with graceful sovereign fallback. 67 CPU delegated via #[cfg(feature = "barracuda")], 43 GPU dispatched via #[cfg(feature = "barracuda-gpu")]. V97: runtime f64 reduction smoke test ensures GPU correctness — detects naga/SPIR-V zeros bug, graceful CPU fallback. All local shaders absorbed upstream; only 2 unique anderson_lyapunov*.wgsl reference shaders remain in metalForge. 13-tier tolerance architecture, all gates green.

NUCLEUS progress: biomeOS Neural API integration via #[cfg(feature = "biomeos")]. Tower (BearDog) health + beacon, Node (ToadStool) compute capabilities, Squirrel AI health — all validated live. NestGate data pipelines (NCBI, NOAA GHCND, IRIS FDSN) wired with sovereign fallback to synthetic data.

See specs/BARRACUDA_EVOLUTION.md for GPU promotion mapping. See specs/PRIMAL_INTERACTION_EVOLUTION.md for NUCLEUS evolution. See metalForge/ for absorption-ready shaders and the manifest.

How groundSpring Relates to Other Springs

Spring	What It Validates	What groundSpring Adds
hotSpring	Clean nuclear math (f64, GPU)	How AME2020 mass uncertainties propagate to model predictions
airSpring	FAO-56 ET₀, soil calibration	The REAL sensor noise — quantifying factory vs field calibration
wetSpring	Microbiome taxonomy, PFAS detection	Sequencing error rates, mass spec noise floors
neuralSpring (future)	ML surrogates, transfer learning	groundSpring provides labeled dirty data; NPU dispatch via metalForge
biomeOS / NUCLEUS	Primal orchestration, data acquisition	groundSpring validates Tower+Node+Squirrel+Nest through Neural API

Directory Structure

groundSpring/
├── control/                         # Phase 0 Python experiments
│   ├── common.py                    # Shared statistical primitives
│   ├── sensor_noise/                # Exp 001: bias-variance decomposition
│   ├── observation_gap/             # Exp 002: model vs station
│   ├── error_propagation/           # Exp 003: Monte Carlo through FAO-56
│   ├── sequencing_noise/            # Exp 004: taxonomic noise floor
│   ├── seismic/                     # Exp 005: wave propagation + source inversion
│   ├── signal_specificity/          # Exp 006: c-di-GMP Gillespie SSA
│   ├── rawr_resampling/             # Exp 007: RAWR vs bootstrap
│   ├── anderson_localization/       # Exp 008: Anderson localization Lyapunov
│   ├── quasiperiodic/               # Exp 009: Almost-Mathieu Quasiperiodic
│   ├── bistable_switching/          # Exp 010: Bistable phenotypic switching
│   ├── multisignal_qs/             # Exp 011: Multi-signal QS integration
│   ├── spin_transport/             # Exp 012: Spin chain transport (Kachkovskiy 2016)
│   ├── resampling_convergence/     # Exp 013: Resampling convergence (Lee & Liu 2024)
│   ├── drift_selection/            # Exp 014: Drift vs selection (R. Anderson 2022)
│   ├── uncertainty_bridge/         # Exp 015: Sensor noise → Anderson ξ uncertainty
│   ├── rare_biosphere/            # Exp 016: Rare biosphere signal detection
│   ├── quasispecies_threshold/    # Exp 017: Eco-evolutionary noise threshold
│   ├── band_edge/                 # Exp 018: Band edge structure
│   ├── jackknife_estimation/      # Exp 019: Jackknife error estimation (Bazavov 2025)
│   ├── freeze_out_inverse/        # Exp 020: Freeze-out inverse problem (Bazavov 2016)
│   ├── spectral_recon/            # Exp 021: Spectral function reconstruction (Bazavov 2025)
│   ├── et0_anderson_propagation/   # Exp 022: ET₀ → Anderson uncertainty
│   ├── notill_sampling/            # Exp 023: No-till vs tilled 16S sampling
│   ├── aggregate_stability/        # Exp 024: Aggregate stability noise
│   ├── precision_drift/            # Exp 025: f32 vs f64 precision drift
│   ├── size_convergence/           # Exp 026: System-size convergence
│   ├── vendor_parity/              # Exp 027: GPU vendor parity
│   ├── npu_anderson/               # Exp 028: NPU Anderson regime classification
│   └── et0_methods/                # Exp 034: Multi-method ET₀ cross-validation
├── crates/
│   ├── groundspring/               # Phase 1 Rust library (41 modules incl. rawr, esn, lanczos, tissue_anderson, biomeos, nestgate, npu, primal_names)
│   └── groundspring-validate/      # 34 validation binaries (hotSpring pattern)
├── metalForge/                     # Write → Absorb → Lean artifacts
│   ├── forge/                      # groundspring-forge crate: hardware discovery, dispatch, topology, pipeline, atomics, remote
│   ├── npu/akida/                  # AKD1000 NPU integration, HARDWARE.md
│   ├── ABSORPTION_MANIFEST.md      # Module-by-module absorption inventory
│   └── shaders/                    # Production WGSL shaders for ToadStool absorption
├── graphs/                         # biomeOS pipeline graphs (deploy, Tower, Node, cross-substrate, validation)
├── niches/                         # BYOB niche YAML definitions (groundspring-measurement)
├── .github/workflows/ci.yml        # GitHub Actions CI
├── wateringHole/                   # Handoff directory (V123 current)
├── specs/
│   ├── BARRACUDA_EVOLUTION.md      # Module → GPU promotion mapping + PRNG roadmap
│   ├── BARRACUDA_REQUIREMENTS.md   # GPU kernel gap analysis
│   ├── CROSS_SPRING_EVOLUTION.md   # Cross-spring shader provenance
│   ├── PRIMAL_INTERACTION_EVOLUTION.md # NUCLEUS Neural API evolution (V0–V6)
│   ├── LAN_DEPLOYMENT_READINESS.md # LAN HPC readiness assessment
│   └── PAPER_REVIEW_QUEUE.md       # 30 papers, three-tier control matrix, open data audit
├── whitePaper/                     # Study, methodology, baseCamp, experiments
│   ├── baseCamp/                   # Per-faculty research briefings (7 faculty)
│   ├── experiments/                # Per-experiment summaries (001-035)
├── tests/                          # Python test suite (29 experiments + parity)
├── Cargo.toml                      # Rust workspace (barracuda feature gate)
├── CONTRIBUTING.md
├── CHANGELOG.md
└── LICENSE                         # AGPL-3.0-or-later

Hardware Gate

Same as all ecoPrimals springs:

Component	Specification
CPU	Intel i9-12900K (16C/24T, 5.2 GHz)
RAM	64 GB DDR5-4800
GPU	NVIDIA GeForce RTX 4070 (12 GB VRAM)
GPU	NVIDIA Titan V (12 GB HBM2)
NPU	BrainChip AKD1000 (80 NPs, 10 MB SRAM, PCIe 2.0 x1)
Storage	1 TB NVMe SSD
OS	Pop!_OS 22.04 (Ubuntu-based)

License

AGPL-3.0-or-later — See LICENSE

---

Version Timeline

Version	Date	Milestone
Init	Feb 16	Repository initialized
Phase 1	Feb 25	29 experiments PASS in Rust
V21	Feb 26	Complete barraCuda rewiring
V26	Feb 27	metalForge live hardware (RTX 4070, Titan V, AKD1000)
V39	Feb 27	NUCLEUS integration + NestGate data pipelines
V53	Feb 28	57 active delegations + GPU grid adapters
V68	Mar 2	L-BFGS refinement, 4D Anderson
V76	Mar 5	Structural evolution, deep debt zero
V85	Mar 6	coralReef sovereign compilation (SM70/SM89)
V97	Mar 7	Three-tier parity proven (29/29 binaries × 3 tiers)
V110	Mar 16	Cross-ecosystem absorption, #[expect(reason)] migration
V115	Mar 18	Zero panicking APIs, ecoBin compliance, 930+ tests
V118	Mar 19	RPC expansion (16 caps), 110 delegations, 960+ tests
V119	Mar 22	Deep audit + cross-ecosystem absorption, 990+ tests, ≥92% coverage
V123	Mar 24	Cross-ecosystem absorption + provenance, 1020+ tests, 41 modules
V120	Mar 23	Deep audit execution: dispatch refactored, forbid(unsafe_code) on 50 binaries, DeviceCapabilities
V121	Mar 23	Deep debt + ecosystem absorption: tolerance centralization, provenance hardening, MSRV 1.87
V122	Mar 24	Cast evolution + module extraction: lib.rs 607→182, validate/lib.rs 769→226, 20+ bare casts→named helpers

Part of ecoPrimals · wateringHole · AGPL-3.0-or-later