Sovereign Human Health

Precision Medicine x Pharmacometrics — sovereign PK/PD modeling, biosignal analysis, drug discovery. healthSpring.

Spring: healthSpring (V35) Domain: Human Health × Pharmacology × Microbiome × Biosignal × Endocrinology × NLME × Clinical Translation × Comparative Medicine × Drug Discovery × biomeOS Niche Deployment License: AGPL-3.0-or-later Date: March 17, 2026 Reproduces work by: Andrea J. Gonzales (MSU Pharmacology & Toxicology), Charles Mok (clinical endocrinology) Status: IPC Resilience + Sovereign Dispatch — V35: 7 tracks complete, 613 tests, 113/113 cross-validation checks, 6 WGSL shaders, 79 JSON-RPC capabilities. Sovereign GPU dispatch via CoralReefDevice; IPC resilience patterns: CircuitBreaker, RetryPolicy, DispatchOutcome. Composition guidance: GROUNDSPRING_V114_PRIMAL_COMPOSITION_GUIDANCE_MAR17_2026.md. V34: Deep debt evolution. V33: IpcError::is_recoverable(), ipc::protocol module, centralized cast module. V32: Structured tracing, health.liveness/health.readiness probes, resilient provenance trio IPC. V31: OrExit<T>, IpcError, enriched capability.list, #![forbid(unsafe_code)]. Zero clippy warnings workspace-wide (pedantic + nursery), zero unsafe, zero TODO/FIXME, zero #[allow()], all files under 1000 LOC. AGPL-3.0-or-later across all files.

Thesis

Sovereign scientific computing can replace Python/NONMEM/proprietary tool chains for human health applications — pharmacokinetics, microbiome analytics, real-time biosignal processing, and endocrine outcome modeling — using pure Rust validated against published data, with live GPU acceleration via barraCuda WGSL shaders and heterogeneous dispatch via toadStool/metalForge. V22: healthSpring now operates as a biomeOS niche — a composed set of primals and workflow graphs orchestrated by the Neural API, with all science capabilities exposed via JSON-RPC 2.0 and discoverable via capability routing.

Population-level validation means nothing without per-person translation. The pipeline closes this loop: a PatientTrtProfile generates a patient-specific clinical scenario — parameterized by age, weight, testosterone level, comorbidities — rendered in petalTongue’s clinical mode via the SAME DAVE motor command channel. The clinician sees the patient, not the infrastructure.

GPU-native execution is validated: a single consumer GPU handles population-scale health computations (10M elements, 207 M/s throughput) with the same WGSL shaders portable to edge devices, TPU, and NPU. Mixed hardware dispatch via NUCLEUS topology routes workloads to CPU, GPU, or NPU based on scale and substrate availability.

V14 adds sovereign replacements for three commercial pharmacometric tools: NONMEM (FOCE estimation), Monolix (SAEM estimation), and WinNonlin (NCA metrics). NLME diagnostics (CWRES, VPC, GOF) complete the population PK pipeline. A WFDB parser enables direct PhysioNet biosignal ingestion. Kokkos-equivalent benchmarks validate GPU-portable patterns. The full petalTongue pipeline now spans 5 tracks with 28 nodes, 121 channels, and 14 scenarios — making the complete healthSpring pipeline human-visible and actionable.

The springs validate science. healthSpring makes the drug.

Tracks

Track 1: Pharmacokinetic / Pharmacodynamic Modeling (Exp001–006, 077)

Pure Rust PK/PD tools extending neuralSpring nS-601–605 (veterinary) to human therapeutics via allometric scaling.

Exp	Title	Key Result
001	Hill dose-response (4 JAK inhibitors + canine reference)	4-parameter Hill equation, IC50/EC50 validated
002	One-compartment PK (IV bolus + oral Bateman + multiple dosing)	AUC trapezoidal, steady-state accumulation ratio
003	Two-compartment PK (biexponential α/β phases)	Distribution/elimination phase separation
004	mAb PK cross-species transfer (lokivetmab → human)	Allometric scaling (BW^0.75 CL, BW^1.0 Vd)
005	Population PK Monte Carlo (1,000 virtual patients)	Lognormal IIV, CL-AUC correlation r = -0.92
006	PBPK 5-tissue physiological compartments	Mass conservation, hepatic clearance, tissue Kp
077	Michaelis-Menten nonlinear PK (phenytoin)	Capacity-limited elimination, dose-dependent half-life, supralinear AUC

Lineage: Paper 12 veterinary→human bridge (Gonzales iPSC, Neubig drug discovery) → healthSpring human therapeutics.

V16: Exp077 adds Michaelis-Menten (capacity-limited) PK — the first nonlinear elimination model. Phenytoin reference parameters from Ludden 1977. GPU-ready via michaelis_menten_batch_f64.wgsl (per-patient parallel Euler ODE).

Track 2: Gut Microbiome and Colonization Resistance (Exp010–013, 078–080)

Extends wetSpring’s Anderson localization framework from soil to gut.

Exp	Title	Key Result
010	Shannon/Simpson/Pielou/Chao1 diversity indices	Validated against SRA published data
011	Anderson localization in gut lattice (1D ξ)	Localization length predicts colonization resistance
012	C. difficile colonization resistance score	Composite: diversity + ξ + Firmicutes ratio
013	FMT microbiota transplant for rCDI	Engraftment fraction → diversity restoration via Bray-Curtis
078	Antibiotic perturbation model (ciprofloxacin)	Exponential kill + recovery dynamics, Dethlefsen reference
079	SCFA production (acetate, propionate, butyrate)	Michaelis-Menten fermentation kinetics, fiber-to-SCFA validation
080	Gut-brain serotonin axis	5-HT production from tryptophan via gut microbiota, Yano 2015 reference

Lineage: Paper 01/06 Anderson QS framework → Paper 12 immunological Anderson → healthSpring gut colonization.

V13 fix: Anderson/IPR computation now uses true eigenvectors from QL diagonalization, not Hamiltonian diagonal.

V16: Exp078 models antibiotic disruption and recovery of the gut microbiome. Exp079 validates SCFA production via Michaelis-Menten kinetics (GPU-ready via scfa_batch_f64.wgsl). Exp080 adds the gut-brain serotonin axis (tryptophan → 5-HT cross-track hypothesis D5 linking microbiome to neurochemistry).

Track 3: Biosignal Processing (Exp020–023, 081–082)

Real-time physiological signal analysis on sovereign hardware.

Exp	Title	Key Result
020	Pan-Tompkins QRS detection (ECG R-peak)	Bandpass + derivative + MWI + threshold
021	HRV metrics (SDNN, RMSSD, pNN50)	Time-domain HRV from R-peak intervals
022	PPG SpO2 R-value calibration	Beer-Lambert AC/DC ratio → SpO2
023	Multi-channel fusion (ECG + PPG + EDA)	FusedHealthAssessment: HR + SpO2 + stress index
081	EDA electrodermal stress detection	Tonic/phasic decomposition, skin conductance response peaks
082	Arrhythmia beat classification (template matching)	Cross-correlation beat typing: Normal, PVC, PAC, BBB

NPU target: Pan-Tompkins is a streaming signal pipeline ideal for Akida AKD1000 (<1ms, microwatt).

V16: Exp081 validates sovereign EDA analysis (tonic/phasic decomposition + SCR detection). Exp082 adds template-matching beat classification — GPU-ready via beat_classify_batch_f64.wgsl (per-beat normalized cross-correlation).

Track 4: Endocrinology — Testosterone PK and TRT Outcomes (Exp030–038)

Clinical claim verification pipeline: extracting quantifiable claims from Dr. Charles Mok’s clinical reference and validating against published registry data.

Exp	Title	Key Result
030	Testosterone PK: IM injection steady-state	Weekly vs biweekly, trough analysis
031	Testosterone PK: pellet depot (5-month)	Zero-order release, 10mg/lb dosing
032	Age-related testosterone decline	Harman 2001 BLSA: -1.6%/yr after 30
033	TRT metabolic response: weight/BMI/waist	Saad 2013 registry (n=411)
034	TRT cardiovascular: lipids + CRP + BP	Sharma 2015 (VA, n=83,010)
035	TRT diabetes: HbA1c + insulin sensitivity	Kapoor 2006 RCT
036	Population TRT Monte Carlo (10K patients)	Lognormal IIV + age-adjusted decline
037	Testosterone-gut axis (cross-track 2×4)	Pielou evenness → Anderson ξ → metabolic response
038	HRV × TRT cardiovascular (cross-track D3)	SDNN improvement → composite cardiac risk

Novel contribution: Exp037 bridges gut microbiome diversity with TRT metabolic outcomes via Anderson localization — a testable hypothesis for clinical investigation. Exp038 validates HRV as a surrogate endpoint for cardiovascular benefit of TRT.

Track 5: NLME Population Pharmacokinetics (Exp075–076)

Sovereign replacement for NONMEM (FOCE), Monolix (SAEM), and WinNonlin (NCA).

Exp	Title	Key Result
075	NLME cross-validation (FOCE/SAEM, NCA, diagnostics)	FOCE 30% theta recovery, SAEM 50%, NCA λz/AUC∞ 5%, CWRES <2.0, GOF R²≥0
076	Full pipeline petalTongue validation (5 tracks, 28 nodes)	197/197 structural checks, 121 channels, all 7 DataChannel types

Novel contribution: First sovereign pure-Rust NLME stack. FOCE + SAEM estimation with NCA and full diagnostics (CWRES, VPC, GOF) — no NONMEM, no Monolix, no WinNonlin, no Fortran, no Python. Deterministic reproducibility (same seed → identical results). VPC Monte Carlo simulation is a GPU promotion candidate (embarrassingly parallel).

WFDB parser: ecoPrimal/src/wfdb.rs — streaming PhysioNet Format 212/16 decoder with beat annotation parsing. Enables direct ingestion from MIT-BIH, MIMIC-III, and other PhysioNet databases.

Kokkos-equivalent benchmarks: ecoPrimal/benches/kokkos_parity.rs — reduction, scatter, Monte Carlo, ODE batch, NLME iteration. Validates GPU-portable patterns ahead of shader promotion.

Industry benchmark mapping: SnapGene, Chromeleon, NONMEM, Monolix, WinNonlin profiled. Sovereign replacements mapped to ecoPrimals stack. See healthSpring/specs/PAPER_REVIEW_QUEUE.md.

Validation Track (Exp040)

Exp	Title	Key Result
040	barraCuda CPU parity (15 analytical contracts)	Hill, Bateman, allometric, Shannon, Simpson, population PK

Integrated Diagnostics (Exp050–052)

Exp	Title	Key Result
050	Integrated 4-track patient diagnostic	Cross-track composite risk score
051	Population diagnostic Monte Carlo (1,000 patients)	Population-level diagnostic distribution
052	petalTongue scenario schema validation	DataChannel, ClinicalRange conformance

CPU vs GPU Parity & Mixed Dispatch (Exp060–062)

Exp	Title	Key Result
060	CPU vs GPU parity matrix (3 kernels × 3 scales)	27/27 parity checks through toadStool Pipeline
061	Mixed hardware dispatch ( NUCLEUS topology)	22/22 dispatch route checks (CPU+GPU+NPU)
062	PCIe P2P transfer validation (Gen3/4/5)	26/26 bandwidth and overhead checks

Clinical Translation (Exp063–065)

Exp	Title	Key Result
063	Patient-parameterized TRT scenarios (5 archetypes)	Per-person: 8 nodes + 8 edges, clinical mode preset
064	IPC push to petalTongue	Unix socket JSON-RPC, live scenario update
065	Live streaming dashboard	ECG, HRV, PK via StreamSession with backpressure

Compute & Benchmark (Exp066–072)

Exp	Title	Key Result
066	barraCuda CPU benchmark	Hill, PopPK, Diversity timing vs Python
067	GPU parity extended	Additional kernel validation
068	GPU benchmark	Throughput at scale
069	toadStool dispatch matrix	Stage assignment validation
070	PCIe P2P bypass	NPU→GPU direct transfer
071	Mixed system pipeline	CPU+GPU+NPU coordinated execution
072	Compute dashboard	toadStool streaming → petalTongue live gauges

Paper Queue Validation (Exp077–082)

Exp	Title	Key Result
077	Michaelis-Menten nonlinear PK (phenytoin)	Capacity-limited elimination, dose-dependent t½, supralinear AUC
078	Antibiotic perturbation model (ciprofloxacin)	Exponential kill + recovery, Dethlefsen reference
079	SCFA production (acetate/propionate/butyrate)	MM fermentation kinetics, Cummings 1987 validation
080	Gut-brain serotonin axis	Tryptophan → 5-HT, microbiota-mediated, Yano 2015
081	EDA electrodermal stress detection	Tonic/phasic decomposition, SCR peak detection
082	Arrhythmia beat classification	Template-matching: Normal/PVC/PAC/BBB, MIT-BIH reference

GPU V16 Portability (Exp083)

Exp	Title	Key Result
083	GPU V16 parity (3 shaders + metalForge + toadStool)	25/25: CPU determinism, physiological ranges, scalar parity, routing, shaders

petalTongue Evolution (Exp073–074)

Exp	Title	Key Result
073	Clinical TRT live dashboard	PK trough streaming, HRV improvement, cardiac risk replace
074	Interaction roundtrip	Mock petalTongue: render, append, replace, gauge, capabilities, subscribe — 12/12

Metrics

Metric	Value
Experiments	61
Rust lib tests	365 ( barraCuda)
Rust forge tests	33 ( metalForge)
Rust toadStool tests	36
Doc-tests	4
Total tests	458
Python cross-validation checks	113/113
Criterion benchmarks	14
CPU parity bench cases	14 (Exp084: Rust 84× faster than Python)
petalTongue pipeline	28 nodes, 29 edges, 121 channels, 14 scenarios
NLME validation	FOCE + SAEM + NCA + CWRES + VPC + GOF (Exp075, 19 checks)
GPU parity checks	27/27 (Exp060) + 25/25 (Exp083)
GPU fused pipeline checks	11/11 (Exp054)
Mixed dispatch checks	22/22 (Exp061)
PCIe transfer checks	26/26 (Exp062)
WGSL compute shaders	6 (3 V15 + 3 V17)
Patient archetypes	5 (Exp063)
Unsafe blocks	0
Clippy warnings	0 (`#![deny(clippy::pedantic)]` in all lib crates, `-W clippy::nursery`)
Max file size	819 lines (under 1000-line limit)

GPU Pipeline (Tier 2) — LIVE

WGSL Shaders (f64 precision)

Shader	GpuOp	Pattern	Validated
`hill_dose_response_f64.wgsl`	HillSweep	Element-wise, power via f32 exp/log	Exp053
`population_pk_f64.wgsl`	PopulationPkBatch	Embarrassingly parallel MC (u32 PRNG)	Exp053
`diversity_f64.wgsl`	DiversityBatch	Workgroup-level reduction	Exp053
`michaelis_menten_batch_f64.wgsl`	MichaelisMentenBatch	Per-patient Euler ODE + Wang hash PRNG	Exp083
`scfa_batch_f64.wgsl`	ScfaBatch	Element-wise MM kinetics (3-output)	Exp083
`beat_classify_batch_f64.wgsl`	BeatClassifyBatch	Per-beat normalized cross-correlation	Exp083

Architecture

Component	Purpose
`GpuContext`	Persistent wgpu device/queue, eliminates per-dispatch init
`execute_fused()`	Unidirectional pipeline: upload → N compute passes → readback
`Pipeline::execute_gpu()`	toadStool dispatches stages via `GpuContext`
`Pipeline::execute_auto()`	metalForge routes per stage (GPU if element count > threshold)

Scaling (RTX 4070, release build)

Operation	GPU Crossover	Peak Speedup	Peak Throughput
Hill dose-response	100K	2.0x (5M)	207 M/s
Population PK MC	5M	1.15x (5M)	365 M/s
Fused pipeline (overhead)	—	31.7x vs individual	—

V14 NLME + Full Pipeline Evolution

Change	Impact
NLME population PK (FOCE + SAEM)	Sovereign NONMEM/Monolix replacement in `pkpd/nlme.rs`. 30 subjects, theta/omega/sigma recovery.
NCA	Sovereign WinNonlin replacement in `pkpd/nca.rs`. λz, AUC∞, MRT, CL, Vss.
NLME diagnostics (CWRES, VPC, GOF)	`pkpd/diagnostics.rs`. CWRES ~N(0,1), VPC 50 simulations, GOF scatter.
WFDB parser	`wfdb.rs` — PhysioNet Format 212/16 streaming decode + beat annotations.
Kokkos-equivalent benchmarks	`benches/kokkos_parity.rs` — 5 GPU-portable patterns validated on CPU.
Full petalTongue pipeline	28 nodes (was 22), 29 edges (was 22), 121 channels (was 65), 14 scenarios (was 13).
Exp075	NLME cross-validation: 19 binary checks (FOCE/SAEM/NCA/CWRES/GOF).
Exp076	Full pipeline validation: 197 binary checks across all 5 tracks + full study.
Industry benchmarks	SnapGene, Chromeleon, NONMEM, Monolix, WinNonlin profiled and mapped.

V19 Full-Stack Portability Evolution

Change	Impact
Exp085 GPU scaling bench (47/47)	4 scales (64→4096) × 3 V16 ops: MM PK (linear scaling confirmed, 64→96K µs), SCFA (sub-µs at 100, 13 µs at 10K), Beat classify (1.4→129 µs). Fused 3-op pipeline: 6ms CPU. metalForge routes small→CPU, large→GPU.
Exp086 toadStool V16 dispatch (24/24)	All V16 StageOps through `execute_cpu` + `execute_streaming` with per-stage callbacks. Streaming matches CPU result. All V16 stages map to `GpuOp` via `to_gpu_op()`. Mixed V15+V16 pipeline (Generate→Hill→Reduce).
Exp087 mixed NUCLEUS V16 dispatch (35/35)	Eastgate Tower topology (CPU+GPU+NPU, PCIe Gen4). V16 workload routing at 8 scales. PCIe P2P bypass GPU↔NPU (31.5 GB/s, 5.1 µs for 160KB). NPU→GPU dispatch plan for biosignal→classification pipeline. GPU-only pipeline: 0 transitions. Full 5-stage mixed pipeline: GPU→GPU→GPU→NPU→CPU with 2 transitions.
Python control (10/10)	Cross-validates MM AUC, SCFA ratios, beat correlation, and scaling linearity from Rust timing results.

V20 petalTongue V16 Visualization Evolution

V20 makes healthSpring’s validated science visible. Six V16 primitives and the compute pipeline now have petalTongue scenario builders producing real-data visualizations.

Change	Impact
V16 scenario builder	6 nodes with real math: MM PK dose curves, antibiotic recovery, SCFA saturation, serotonin pathway, EDA decomposition, arrhythmia templates
Compute pipeline scenarios	GPU scaling curves, NUCLEUS topology, mixed dispatch plan — all as petalTongue DataChannels
Full study extended	28 → 34 nodes, 29 → 38 edges, all 7 DataChannel types in unified graph
Exp088 unified dashboard	326 validation checks across all scenarios, JSON dump + IPC push, quick-start guide
Exp089 patient explorer	CLI-parameterized diagnostic + V16 analysis, streams to petalTongue, combined diagnostic+V16 scenario
dump_scenarios extended	14 → 16 scenario JSONs (added healthspring-v16.json, healthspring-compute.json)

V18 CPU Parity Benchmark Evolution

Change	Impact
Exp084 V16 CPU parity bench	14 matching benchmark cases across 6 V16 primitives: Python baseline (17/17 checks) vs Rust CPU (33/33 checks).
Rust 84× overall speedup	Aggregate mean across all 14 bench cases: Python total 52,034 µs vs Rust 622 µs.
SCFA 160×, Antibiotic 233×	Michaelis-Menten math: compiled Rust eliminates interpreter overhead entirely.
Beat classification 149×	Template matching via normalized cross-correlation: 1000 beats in 204 µs (Rust) vs 30,377 µs (Python).
Serotonin 149×, Tryptophan 155×	Sigmoid diversity-factor computation: Rust inlines and optimizes exp/sigmoid chains.
MM PK simulate 33×	Euler ODE integration: 10,000 steps in 110 µs (Rust) vs 3,626 µs (Python).
EDA SCL/phasic — numpy faster	numpy `convolve` uses compiled C/BLAS internally; naive Rust rolling average is 3× slower. Target for SIMD optimization in barraCuda.
`bench_results_v16_rust_cpu.json`	Machine-readable timing results for downstream CI comparison.
`bench_results_v16_python.json`	Python baseline timings with provenance (Python version, numpy version).
`compare_v16_benchmarks.py`	Side-by-side speedup table generator for Rust vs Python timing data.

V17 GPU Portability Evolution

Change	Impact
`michaelis_menten_batch_f64.wgsl`	Per-patient Michaelis-Menten ODE via Euler integration on GPU. Wang hash + xorshift32 PRNG for lognormal Vmax variation.
`scfa_batch_f64.wgsl`	Batch SCFA production (acetate/propionate/butyrate) via element-wise Michaelis-Menten fermentation kinetics.
`beat_classify_batch_f64.wgsl`	Per-beat template-matching classification via normalized cross-correlation. Normal/PVC/PAC/BBB typing.
metalForge 3 new `Workload` variants	`MichaelisMentenBatch`, `ScfaBatch`, `BeatClassifyBatch` — cross-system routing with GPU/CPU threshold selection.
toadStool 3 new `StageOp` variants	Streaming pipeline dispatch for all V16 primitives — CPU fallback + GPU promotion.
Exp083 GPU V16 parity	25/25 validation checks: CPU determinism, physiological ranges, scalar API parity, metalForge routing, shader compilation, memory estimates.
`GpuContext` fused support	All 6 `GpuOp` variants dispatchable through `execute_fused()` unidirectional pipeline.
`bytemuck` param structs	`MmParams`, `ScfaGpuParams`, `BeatClassifyParams` — zero-copy GPU uniform upload.

V16 Paper Queue Complete Evolution

Change	Impact
Exp077 Michaelis-Menten PK	Capacity-limited (nonlinear) elimination model. Phenytoin reference (Ludden 1977). Dose-dependent half-life, supralinear AUC.
Exp078 Antibiotic perturbation	Gut microbiome disruption + recovery dynamics. Exponential kill model, Dethlefsen ciprofloxacin reference.
Exp079 SCFA production	Fiber-to-SCFA fermentation via Michaelis-Menten kinetics. Acetate, propionate, butyrate validated against Cummings 1987.
Exp080 Gut-brain serotonin	Tryptophan → 5-HT production via gut microbiota. Cross-track hypothesis D5 (Yano 2015).
Exp081 EDA stress detection	Tonic/phasic decomposition, skin conductance response peak detection.
Exp082 Arrhythmia classification	Template-matching beat typing: Normal, PVC, PAC, BBB. MIT-BIH annotation reference.
6 Python control scripts	Exp077–082 each have `control_*.py` cross-validation. 167 total Python checks.
14 Criterion benchmarks	`benches/paper_queue.rs` — MM PK, antibiotic, SCFA, serotonin, EDA, arrhythmia at 3 scales.
Paper queue 30/30	All 30 reviewed papers now have validated experiments.

V15 Upstream Rewire Evolution

Change	Impact
`PrecisionRouting`	Mirrors toadStool S128 precision dispatch. CPU f64, GPU split/emulated f64, NPU quantized.
`rng` delegate	LCG/xorshift delegate to canonical `barracuda::rng`.
`eigensolver` delegate	QL diagonalization delegates to `barracuda::special`.
Cross-spring shader docs	WGSL shader evolution path documented across springs.
Upstream parity benchmarks	Kokkos patterns validated against canonical upstream.

V13 Deep Audit Evolution

Change	Impact
Anderson eigensolver	QL algorithm for correct eigenvalue/eigenvector computation from tridiagonal Hamiltonian
Smart clinical.rs refactor	1177 → 374 + 819 lines, domain-coherent split
LCG PRNG centralization	New `rng.rs` module, 4 files updated
Math deduplication	`evenness_to_disorder` and `lognormal_params` delegate to canonical source
Capability-based discovery	Glob-based Songbird socket search replaces hardcoded path
Flaky test fix	AtomicU64 paths + kernel connection queuing replaces Barrier synchronization
Doc-tests	4 added (shannon_index, hill_dose_response, auc_trapezoidal, state_to_f64)

Per-Person Clinical Translation

The critical addition: the pipeline from validated population models to individual patient scenarios.

Pipeline

Published data → Computational model → Population validation
    → PatientTrtProfile (age, weight, T level, comorbidities)
        → trt_clinical_scenario() → 8-node graph + edges + channels + ranges
            → petalTongue clinical mode (motor commands: hide sidebars, skip awakening, fit view)
                → Clinician sees THIS patient's projected trajectory

SAME DAVE Neuroanatomy (petalTongue Integration)

The SAME DAVE model (Sensory Afferent, Motor Efferent) provides self-aware control of the visualization system:

Channel	Direction	healthSpring Use
Scenario load	Afferent	Patient scenario → graph topology
Mode preset	Efferent	`mode: "clinical"` → bundle of motor commands
Panel visibility	Efferent	`show_panels` → SetPanelVisibility motor commands
Awakening control	Efferent	`awakening_enabled: false` → skip startup animation
Zoom control	Efferent	`initial_zoom: "fit"` → FitToView motor command
IPC push	Afferent	JSON-RPC → scenario update without restart

Cross-Paper Dependencies

Paper 01 (Anderson-QS) ──→ gut lattice model (Exp011, 037)
Paper 06 (No-Till)     ──→ Anderson framework for biological substrates
Paper 12 (Immunological Anderson) ──→ veterinary→human PK bridge (Exp004), allometric scaling
Paper 08 (NPU Edge IoT) ──→ biosignal NPU target (Exp020-023)
Paper 07 (Sovereign WDM) ──→ GPU dispatch methodology, metalForge architecture

Reading Order

Standalone: Paper 13 is self-contained for anyone interested in computational pharmacology or clinical decision support.

In context: 12 (immunological Anderson, veterinary PK lineage) → 13 (human health applications) → 01 (Anderson framework)

Biosignal focus: 13 §Track 3 (sovereign biosignal) → 08 (NPU edge IoT) → 04 (sentinels)

Microbiome focus: 13 §Track 2 (gut Anderson) → 01 (Anderson-QS) → 06 (no-till Anderson) → 03 (bioag microbiome)

Data Sources

All validation data derives from published, open-access sources:

Harman 2001 (BLSA testosterone decline)
Saad 2013/2016, Traish 2014 (TRT registries)
Sharma 2015 (VA cardiovascular cohort)
Kapoor 2006 (RCT diabetes)
Kabashima 2020, Silverberg 2021 (mAb Phase III)
Gabrielsson & Weiner (PBPK reference model)
Kleiger 1987 (HRV mortality landmark study)
SRA public 16S datasets for microbiome baselines

No proprietary clinical data is used. Mok clinical reference provides hypotheses; registry data provides validation.

V22 — biomeOS Niche Deployment

V22 transforms healthSpring from experiment binaries into a biomeOS BYOB niche:

Component	File	Purpose
Primal binary	`ecoPrimal/src/bin/healthspring_primal/`	79 capabilities via JSON-RPC 2.0 over Unix socket
IPC dispatch	`ecoPrimal/src/ipc/dispatch/`	Method → science function routing for 6 domains
Niche manifest	`graphs/healthspring_niche.toml`	Declares the niche: primals + workflow graphs
Patient assessment	`graphs/healthspring_patient_assessment.toml`	ConditionalDag: 4 parallel science tracks → composite
TRT scenario	`graphs/healthspring_trt_scenario.toml`	Sequential TRT clinical workflow
Microbiome analysis	`graphs/healthspring_microbiome_analysis.toml`	Sequential diversity → Anderson → SCFA pipeline
Biosignal monitor	`graphs/healthspring_biosignal_monitor.toml`	Continuous 250 Hz real-time monitoring

healthSpring is now a niche, not a node. The primal provides capabilities; the graphs define composition. biomeOS’s Neural API orchestrates and optimizes via the Pathway Learner. See wateringHole/SPRING_NICHE_SETUP_GUIDE.md for how other springs can follow this pattern.

V32 — Cross-Spring Ecosystem Convergence

V32 absorbs proven patterns from all ecoPrimals components:

Feature	Source	Impact
Structured `tracing`	All 6 sibling springs	`eprintln!` → `tracing::info!/warn!/error!` with env-filter (`RUST_LOG`)
`health.liveness` + `health.readiness`	coralReef Iter 51	Lightweight probes for orchestrator health monitoring
Resilient provenance trio IPC	sweetGrass v0.7.18	Circuit breaker (5s cooldown) + exponential backoff retry
`IpcError` ecosystem type	biomeOS/airSpring/groundSpring	Structured error enum with `RpcError` and `Timeout` variants
`OrExit<T>` trait	wetSpring V123	Panic-free error handling for validation/utility binaries
Enriched `capability.list`	biomeOS Pathway Learner	`operation_dependencies` + `cost_estimates` for execution graph planning

613 tests, 79 JSON-RPC capabilities, 6 WGSL shaders, zero clippy warnings, zero unsafe. Sovereign GPU dispatch via CoralReefDevice; IPC resilience: CircuitBreaker, RetryPolicy, DispatchOutcome. See GROUNDSPRING_V114_PRIMAL_COMPOSITION_GUIDANCE_MAR17_2026.md for composition guidance.

License: AGPL-3.0-or-later