TERRA VIVENS SANCTAE MARIAE

Closed-Loop Bioelectrochemical Restoration of an Insular Soil Microbiome

A 36-month, six-work-package R&D programme integrating pedogenesis, electrokinetics, microbial ecology and machine-learning control to restore a degraded island soil without environmental effluent.

1.1 The Remediation Trilemma

Every conventional approach to rehabilitating a chemically compromised soil resolves the remediation problem by transferring it somewhere else. Excavation and land-filling satisfies the contaminant mass balance on paper while exporting several hundred percent of the original soil carbon to landfill, destroying the microbial inoculum and the physical structure that would have enabled in-situ recovery 1. Chemical washing and chelant extraction achieves high removal efficiency but generates a secondary liquid waste stream of comparable or greater volume and toxicity than the original contamination 24. Unmanaged fallowing avoids harm but forfeits the agricultural baseline, leaving the land in a state of slow biological decline governed by legacy metal residence times measured in decades 3.

Removal percentages are indicative ranges drawn from field trials summarised in 1, 2 and 6; the “evidence tier” denotes the maturity of the underlying evidence base (see Reader's Note in the Abstract).

1.2 Why Bioelectrochemical Pedogenesis Is Different

The bioelectrochemical pedogenesis framework departs from the premise that remediation and restoration are opposing activities. In a saturated soil matrix, an applied electric field drives charged species by electromigrationNet transport of ionic species induced by the electric field — the dominant transport mechanism in low-permeability clays where hydraulic diffusion is negligible. toward the electrode of opposite sign. This is a vectorial process, not a volumetric one. The engineering consequence is decisive: contaminants can be steered to a chosen collection geometry rather than merely extracted 67.

The limitation of classical electrokinetics is equally well established. Water electrolysis at the electrodes generates acid fronts at the anode and alkaline fronts at the cathode; if these fronts are allowed to migrate into the root zone they dissolve essential minerals, precipitate metal hydroxides, and — critically — they destroy the microbiological community that any restoration attempt depends upon 78. Electrokinetic remediation is thus, in its conventional form, microbiologically antagonistic.

1.3 The Insular Advantage: A Natural Control

Island systems are scientifically under-used. The frequent methodological weakness of restoration research — pseudoreplication Treating many samples from a single manipulated system as independent replicates, which inflates apparent statistical power; a design flaw explicitly addressed by this programme's blocked randomised design. and uncontrolled confounding from off-site inputs — is substantially mitigated at 1 St. Mary's Island, Alexandria Bay, New York. Hydrologically delimited by the St. Lawrence River, the site receives a bounded and characterisable input flux, which makes both contamination mass-balance accounting and biological response attribution tractable.

The site additionally supports a Before-After-Control-Impact (BACI) design of unusual quality. A matched mainland reference parcel permits the separation of restoration signal from regional climate and river-chemistry signal, which is the single most common failure of published restoration claims 910.

1.4 Research Question

2.1 The Target Research Estate as an Isolated Test Facility

The target research site is 1 St. Mary's Island, Alexandria Bay, Jefferson County, New York (MLS S1648820), a river island within the St. Lawrence River basin (municipal parcel 222201-003-069-0001-005-000). The organisation holds a Letter of Intent to acquire the parcel, dated 30 September 2026.

If acquired, the property would serve as an isolated experimental facility rather than a production surface — a controlled environment in which the AI-enabled soil-revival system can be developed, instrumented and evaluated without the confounding of open-field conditions. The cedar-clad structure on the island would serve as the laboratory, instrumentation hall and secure store; the surrounding acreage would supply source material and, later, a protected deployment surface.

Coordinates are given to two decimal places, appropriate to a site-scale characterisation and deliberately not implying survey-grade precision. Island landform class and NRCS soil series references are to be confirmed against the NRCS Web Soil Survey during WP1; this document treats them as provisional.

2.1.2 Site-Contingency Scenarios

Because the property is not yet owned, the programme is designed with an explicit contingency ladder. The scientific content is invariant across all three scenarios; only the physical platform changes. This means funding the research does not depend on the property purchase succeeding — a materially different risk profile from a design that requires the real estate in order to begin.

2.1.3 The AI-First Development Sequence

The programme's distinguishing feature is that the artificial intelligence is the experiment, not a reporting layer applied to a finished biological result. The development sequence therefore inverts the conventional order:

1. Characterise the substrate — WP0 establishes pedological baselines across the six varietal classes so the AI has a structured, physically meaningful feature space rather than an unconstrained sensor dump.
2. Instrument to a telemetry contract — every variable the model consumes is specified, calibrated and continuously recorded, so training data and validation data share an identical schema.
3. Train against ground truth — surrogate models are fitted to laboratory-measured endpoints (DTPA-extractable metals, redox state, colonisation) using the high-frequency telemetry as inputs. The AI learns the mapping sensor stream → measured soil property.
4. Close the loop under constraint — the validated surrogate is coupled to the constrained controller, which proposes actuations; every actuation changes the soil, producing new ground truth.
5. Iterate and generalise across varietal classes — the same model is tested on soils it was not trained on, which is the only honest test of whether the AI has learned soil physics or merely memorised one substrate.
6. Deploy outdoors under supervision — only after the system demonstrates out-of-distribution behaviour on held-out varietal classes is it permitted to act on the open island surface, and then only within its certified action set.

This sequence has a consequence worth stating plainly: the AI cannot outrun its instrumentation. The quality of the soil model is bounded by the quality of the ground truth it is trained against, which is why WP0 measurement and the telemetry contract in Section 6 precede every element of the control stack in Section 7.

2.2 Climate and Hydrologic Context

The St. Lawrence lowland is a humid continental regime with cold winters, warm summers, and precipitation distributed relatively evenly through the year. Two hydrological features dominate the island's pedology. First, the St. Lawrence River provides a strongly buffered baseflow regime, modering seasonal water stress relative to interior sites. Second, the island's exposure to winter freeze–thaw and spring freshet produces repeated wetting–drying cycles that drive aggregate slakingStructural collapse of soil aggregates on re-wetting after air-drying, driven by the dissolution of transient capillary bridges and the swelling of clays — a primary mechanism of structural degradation in historically tilled soils. and drive the seasonal oxidation–reduction cycling that governs $E_h$ dynamics discussed in Section 2.4.

2.3 Source Soil Collection & Varietal Stratification

The programme does not treat “the soil” as a single medium. It begins from collected samples spanning distinct varietal classes, because the electrokinetic and biological mechanisms under test respond to texture, organic matter and mineralogy in quantitatively different ways. A single-texture experiment would test the mechanism only where it is easiest to demonstrate — and would fail exactly where it is most needed. Varietal stratification is therefore a design variable, not a nuisance to be averaged away.

Source material is collected from the island and from paired reference locations, composited by pedon at the 0–20 cm and 20–40 cm intervals, and archived as air-dried, sieved material with retained moisture subsamples for later reconstitution. Six varietal classes are targeted, spanning the hydraulic and textural range that determines redox behaviour in the field:

$\phi^{\text{tot}}$ denotes total porosity, a primary target variable rather than a class label. Sample numbers per class are set by the power analysis in Section 6.3, not by convenience; minimum 30 kg of archived material per varietal class is required to support the destructive column design.

2.4 WP0 Baseline: Measured and Derived Endpoints

WP0 establishes the pre-treatment state for every varietal class. It operates in two layers. The first is directly measured — quantities a laboratory can determine without recourse to any model. The second is derived by pedotransfer from those measurements, using published functions. The distinction is maintained throughout, because presenting a derived value as though it had been observed is the most common route by which soil models misreport themselves.

2.5 Redox Potential as the Master Diagnostic

Of the measured endpoints, oxidation–reduction potential ($E_h$) occupies a unique methodological position: it is the cheapest to monitor continuously, the most sensitive to biological state, and the most diagnostic of the conditions governing metal bioavailability 1112. Under oxic conditions ($E_h$ above roughly +350 mV) iron and manganese are oxidised and largely insoluble, and metal cations are held by exchange and by precipitation on oxide surfaces. Under reducing conditions ($E_h$ below roughly +100 mV) Fe(III) and Mn(IV) oxides dissolve, releasing sorbed metals back into solution and generating phytotoxicity 13. Restoration, in redox terms, is the recovery of a stable, aerobic, biologically productive $E_h$ regime.

Because $E_h$ governs both the contaminant chemistry and the microbial community, it is the state variable the controller must hold. The next subsection makes it predictable.

2.6 Redox from First Principles: The Oxygen Supply Index

Redox potential is not an independent variable of soil; it is the consequence of a supply–demand imbalance. Oxygen is produced at the surface faster than it can diffuse inward, so the profile is set by the ratio of O₂ diffusion capacity to microbial O₂ demand. Defining that ratio explicitly makes $E_h$ predictable from measurable soil properties rather than something that must be waited for and observed.

The construction runs in three steps.

STEP 1 — EFFECTIVE OXYGEN DIFFUSIVITY

Gas transport in a porous medium is reduced by tortuosity. The Millington–Quirk relation expresses the effective coefficient as a fractional power of the air-filled porosity:

$D_0$ = diffusion coefficient in free air; $\epsilon_a$ = air-filled porosity; $\theta$ = volumetric water content; $\rho_b$ = bulk density; $\rho_s$ = particle density ($\approx 2.65$ g cm⁻³ for mineral soils, $\approx 1.5$ for organic soils).

STEP 2 — OXYGEN SUPPLY INDEX

Demand is taken as proportional to the respireable organic carbon pool, scaled by temperature. The Oxygen Supply Index (OSI) is the dimensionless ratio of supply to demand. An OSI above unity indicates that oxygen arrives faster than it is consumed, and the soil is oxidising; an OSI near zero indicates diffusion limitation and reducing conditions:

STEP 3 — MAPPING THE SUPPLY RATIO TO $E_h$

A soil is not in equilibrium with atmospheric oxygen; it is a diffusion-limited reaction front. The controlling couple therefore shifts with the supply ratio. Fully oxidised soil is governed by the MnO₂/Mn²⁺ boundary, $E_{\text{Mn}} = 1230 - 118.3\,\text{pH}$; an anoxic soil by the Fe(OH)₃/Fe²⁺ boundary, $E_{\text{Fe}} = 771 - 59.16\,\text{pH}$. These are bridged by a logarithmic supply term:

$S$ = 190 mV per decade of OSI. Calibrated so that OSI = 1 yields ≈ +490 mV (well aerated) and OSI = 10⁻⁴ yields ≈ −210 mV (strongly anoxic), matching published field ranges for mineral soils. The functional form is a calibrated transfer function, not a thermodynamic identity: the exact value is re-fitted against WP0 measurements before Gate 1.

3.1 Soil as a Living Medium

Conventional agronomy treats soil as a passive medium — a container in which nutrients dissolve and roots are inserted. The modern critical-zone perspective treats the soil surface layer as the principal regulatory interface of the terrestrial biosphere: the zone in which atmosphere, lithosphere, hydrosphere and biosphere intersect and exchange 1516. Within this framing, soil organic carbon is not merely a fertility amendment. It is the structural matrix that determines aggregate architecture, water retention, cation exchange and the habitat in which microbial and fungal communities persist 1718.

A degraded soil exhibits the simultaneous loss of all three: carbon, structure, and biological community. These losses are mutually reinforcing. Compaction reduces porosity, which reduces oxygen diffusion, which drives $E_h$ downward, which suppresses aerobic decomposition, which slows mineralisation, which further starves the microbial community of substrate. The restoration problem is therefore not the removal of any single negative factor but the re-establishment of a self-sustaining positive feedback loop.

3.2 Four Contested Claims, Graded Honestly

A proposal that conflates established science with speculation invites the loss of credibility that attaches to the whole programme when the speculative components fail. The four mechanistic pillars of this work are therefore stated with their evidential status attached.

3.3 Governing Transport Model

Solute transport in a saturated porous medium under an imposed potential follows the classical electrokinetic form of the advection–dispersion equation, with electromigration added as a dominant sink term 627:

$c$ = dissolved concentration; $\mathbf{u}$ = pore-water velocity; $D_{e}$ = effective dispersion coefficient; $z$ = ionic charge number; $F$ = Faraday constant; $R$ = gas constant; $T$ = absolute temperature; $\mathbf{E}$ = electric field; $\lambda$ = decay/sorption rate constant. The fourth term is the electromigration flux that makes directed steering possible; the last term represents sorptive loss to the reactive matrix.

Each hypothesis below is stated in a form that permits disproof. For each, the programme specifies (i) the null hypothesis, (ii) the effect size that would constitute a meaningful result, and (iii) the disconfirming observation — the measurement that, if made, would cause the programme to abandon or redesign that component. Hypotheses are grouped into restoration (H1–H3), control-system (H4–H6) and AI-generalisation (H7–H8) families.

The MSD architecture is a permeable reactive barrierA barrier through which pore water and ions pass freely, but which immobilises dissolved contaminants; here realised as a replaceable cassette rather than a fixed excavation. configured as a modular, sacrificial cassette positioned between the treatment electrodes and the vegetative root zone. The design resolves a deceptively simple problem: how to confine the three things electrokinetic remediation generates — migrating contaminant ions, electrode-generated pH fronts, and the heat of electrolysis — away from the biological zone that restoration depends upon.

5.1 Cassette Formulation

Layered double hydroxides are specified specifically because standard silica sand and biochar fail to capture anionic species; a reactive barrier that targets only cations is ineffective against oxyanion contamination 28.

5.2 The Four-Stage Telemetry Mesh

Field nodes use ESP32-S3 / nRF52840 solar micro-nodes communicating over LoRaWAN (915 MHz) or RS-485 Modbus RTU, with an edge controller (Raspberry Pi 5 / Jetson Orin Nano) running containerised MQTT ingestion, a time-series store, and local inference 29.

6.1 Source Material Characterisation

Characterisation of the six varietal classes follows the USDA-NRCS Soil Survey Manual. Material is sampled by depth increment to 40 cm using a soil auger, with undisturbed cores taken for bulk density, porosity and aggregate determination. Samples are composited within varietal class and never across classes, so the substrate assigned to a treatment arm is homogeneous and reproducible across replicate vessels.

Because the design runs in an isolated facility rather than open ground, three classes of variance that normally dominate field experiments are removed by construction: weather events, uncontrolled hydrological flux, and off-site deposition. What remains is the variance the programme actually cares about — substrate class, treatment, and their interaction — and each is separately blocked and estimable.

6.2 Controlled-Environment Experimental Design

The design is a randomised complete block factorial, replicated across vessels, run in controlled-environment growth space. The vessel is the experimental unit — the unit randomised, the unit treated, and the unit of inference. The varietal class is the blocking factor, which converts substrate heterogeneity from a confound into a tested effect.

Each vessel is a replicated column packed with a single varietal class at a specified bulk density, containing an instrumented MSD cassette and a planted root zone. Moisture content and temperature are actively held at the class-specific set points derived in Section 2.6, so that redox state is a controlled variable rather than a weather-dependent outcome.

The PEMF sub-factor is applied by splitting the treatment vessel longitudinally, so both PEMF states sit within the same substrate, moisture history and microbial inoculum. This is a stronger control than separate vessels: it removes substrate heterogeneity entirely from the H6 comparison.

6.3 Telemetry Mesh & Closed-Loop Instrumentation

The instrumentation layer is what makes restoration an engineering discipline rather than a horticultural one, and it is also what supplies the AI with its training data. The dashboard below renders a replay of the modelled 36-month trajectory for a single instrumented vessel (treatment arm C3, varietal class SiL), demonstrating how controller actuations and soil response co-evolve. Values are simulated from the model of Section 3.3 and exist to demonstrate the telemetry contract — the signals, units, thresholds and alarm semantics the real system will report. They are not observations.

6.4 Laboratory & Field Protocols

Soil physical, chemical and biological analyses follow the methods in Table 2.1. Total metals are determined by EPA 3051A digestion with ICP-MS quantification; bioavailable metals by DTPA extraction, which better predicts phytotoxic exposure than total content because it approximates the plant-available pool. Redox potential is measured in situ with combination electrodes at field capacity, logged at 15-minute intervals.

Disturbance discipline. Sampling uses a fixed georeferenced coring template across all sampling dates. Wireless nodes are installed to fixed depth in permanently marked sleeves so that repeated insertion does not progressively compact the measurement volume. This matters more than it appears: sensor insertion is itself a disturbance that would otherwise become a confounding trend in a 36-month series.

6.5 Amendment Protocol (Arms C1 and C3)

1. Physical decompaction without inversion. Subsoil shank or broadfork at 35–45 cm. Ploughing and inversion are prohibited: tillage crushes existing fungal hyphae and oxidises residual organic matter, compounding the damage it is intended to remedy.
2. Quenched biochar application at 1.5–2.5 t ha⁻¹, charged with compost extract, fulvic acid, fish amino acids and paramagnetic basalt powder, then cured 14–21 days under breathable tarp until thermal stability indicates microbial colonisation of the pore structure.
3. Fungal and microbial inoculation at ≥100,000 viable propagules m⁻², with indigenous microbial inoculum and saprophytic decomposer slurry applied beneath mulch.
4. Succession cover polyculture — a twelve-species cocktail combining taproot drillers, nitrogen-fixing legumes, fungal-biomass builders and pollinator/dynamic accumulators. Bare soil is never left exposed.
5. Organic armour. A 5–10 cm carbonaceous mulch to suppress evaporation, buffer temperature swing and shield soil biology from UV.

7.1 Composite Restoration Index

No single variable constitutes soil health. The programme therefore constructs a composite Restoration Index (RI) from z-scored, directionally normalised endpoints, weights fixed a priori rather than fitted to the observed data, so that the index cannot be reverse-engineered to produce a desired result:

$z_i^{*}$ is the directionally normalised z-score of endpoint $i$ (sign-corrected so that higher is always better); $w_i$ is the pre-specified weight; $k = 8$ for the primary index.

7.2 Controller Constraint Set

The controller's admissible action set is a convex polytope rather than an unconstrained action space. This single design decision is what converts a machine learning controller from a hazard into a safety instrument:

with $E_{\max} = 1.2$ V cm⁻¹ and $\dot{E}_{\max} = 0.1$ V cm⁻¹ min⁻¹. Any action proposed by the learned policy is projected onto $\mathcal{K}$ before actuation; a projection onto a compact convex set is idempotent and cannot produce an out-of-bounds command 28.

7.3 Lyapunov Stability of the Closed Loop

The programme's central theoretical claim is that the constrained closed loop cannot run away. The argument is stated in full below; it is deliberately elementary, and elementary is the point — an esoteric proof that reviewers decline to check provides no safety assurance.

7.4 Controller Action Matrix

The rule-based fallback controller — the one the system reverts to if H4 fails — operates on the following state machine. It is specified in full so that the fallback is auditable without reference to the learned policy.

The TOXICITY and SATURATED rows are hard interlocks: they override the learned policy entirely, including the project's target trajectory. No learned policy may act while an interlock is asserted.

The programme is organised into six work packages sequenced so that each gate is a genuine decision point — at each gate the programme can legitimately stop, redirect, or continue on the basis of accumulated evidence. A programme with no stopping rules is a programme that cannot be held to account.

8.1 Gate Criteria & Stopping Rules

Risks are scored on a 5×5 matrix (likelihood × impact) and each carries a named owner and a trigger threshold that initiates the mitigation — not a vague intention to monitor. Select any cell below to inspect the risks assigned to it.

L = likelihood, I = impact, each 1–5; Score = L × I. R15 exists because a programme that publishes only positive results has already failed as science, irrespective of its technical merits.

10.1 Regulatory & Permitting Matrix

The programme is scoped to avoid regulated-activity thresholds wherever scientifically possible. This is a deliberate design choice, not an oversight: a research programme that requires fewer exemptions is faster, cheaper, and more likely to complete.

Permitting determinations are provisional and must be confirmed with the relevant authority before ground disturbance. Nothing in this table constitutes legal advice.

10.2 Governance Structure

10.3 Ethics, Publication & Data Policy

• Pre-registration. Hypotheses, primary endpoints, exclusion criteria and the analysis plan are registered before WP2 completes, so that the analysis cannot be retrofitted to the data.
• Null-result parity. Negative findings are published with the same prominence as positive findings. This is contractual, not aspirational (risk R15).
• Selective reporting audit. All measured endpoints appear in the publication regardless of the direction of the result.
• Data release. De-identified telemetry, laboratory data and analysis code are released under an open licence within 12 months of each gate (see Section 15).
• Stewardship commitment. The island is treated as a long-term research asset, not a development site. No construction beyond what the research requires.

11.1 Carbon Co-Benefit Projection

Restoration produces a measurable carbon co-benefit. The model below is a first-order projection based on IPCC Tier 1 stock-change factors for mineral soil, modified for the biochar carbon-fixing pathway 3031. Adjust the parameters to test sensitivity; the output is deliberately conservative and should be treated as an order-of-magnitude estimate, not a verified offset.

12.1 Intended Impact

A proposal that does not articulate its own weaknesses has not been reviewed. The seven objections below are, in the authors' assessment, the strongest arguments against this programme. Each is stated at full strength, followed by the response — and, where the response is weak, that is acknowledged.

13.1 Residual Limitations

• Statistical power. Twenty plots per treatment across four blocks gives limited power for small effect sizes. The programme is powered for the large effects it hypothesises, not for subtle ones; a null result must be read as a confidence interval, not as proof of absence.
• Unmeasured legacy contaminants. The fourteen-endpoint battery may miss contaminant classes not included in the baseline, such as persistent organics. A broader screening scan is recommended at WP1.
• Equipment maturity. Commercially available reactive media and ISFET assemblies are not purpose-built for this application; some development effort is assumed.
• Model transferability. The kinetic parameters in Section 3.3 are literature-derived and site-specific only after WP2 validation. Pre-validation model output is illustrative only.
• Climate confounding. Thirty-six months spans two complete seasonal cycles, which is the minimum at which seasonal covariates are estimable. More would be better; the design does not claim otherwise.
• Analyst degrees of freedom. Pre-registration mitigates but does not eliminate discretion in exploratory analysis. All exploratory analyses are labelled as such and reported separately from confirmatory results.

The Terra Vivens Sanctae Mariae programme proposes a falsifiable test of whether electrokinetic remediation and biological restoration can be made mutually compatible through physical decoupling, and whether a constrained closed loop can operate that decoupled system materially efficiently.

The scientific contribution is not a new remediation technology. Electrokinetic remediation has existed for four decades, and mycorrhizal pedogenesis has been understood for longer still. The contribution is the coupling — an architecture that treats the antagonism between the two as a design problem to be solved rather than a constraint to be accepted, together with the measurement discipline to determine whether the solution actually works.

14.1 Expected Contributions

1. An operational definition of pedogenic function expressed as a pre-registered, weight-specified composite index over independently validated endpoints — and, more valuably, a full sensitivity analysis showing how much such indices depend on weighting choices.
2. A quantitative assessment of whether physical decoupling of electrolysis fronts and field from a root zone is achievable, and at what cassette geometry.
3. A validated reactive-media formulation with published working capacity and passivation kinetics for a St. Lawrence lowland soil.
4. An open, well-characterised BACI dataset with 36 months of high-frequency telemetry from a hydrologically bounded site — a resource of standalone value to the modelling community.
5. A reusable, safety-certified control architecture in which the learning component is provably bounded and provably removable.
6. A published negative or null result on H6, should that be the outcome — contributing to a cleaner picture of the sub-Faradaic bioelectric literature than currently exists.

14.2 Closing Statement

The organisation approaches this programme as an ecological trust and a research instrument, and it approaches the St. Mary's Island property as an opportunity rather than a prerequisite. The science is testable whether or not the purchase completes, and the three scenarios in Section 2.1.2 exist so that no single funding decision can determine whether the programme proceeds. The organisation's first obligation is to the integrity of the result, and its second to the site — and both are served by the same design choice: to specify in advance what would prove the approach wrong, and to stop if it does.

Terra Vivens Sanctae Mariae — The Living Soil of Saint Mary

“Earth Inspired Self Contained Self Reliant Systems”

Glossary of Terms

Notation

Source Protocol Documents

The amendment sequence in Section 6.5 derives from the organisation's internal protocol library: Protocol 01 — Soil Resurrection & Rhizosphere Genesis, Protocol 02 — Alchemical Formulations, Protocol 03 — Electroculture Schematics, and Protocol 04 — Restoration Phasing, together with the four specialist agent dossiers (soil, alchemy, electroculture, and AI). Where those documents describe practices whose efficacy is not supported at peer-review standard — specifically water-structuring and monatomic mineral preparations — this white paper deliberately excludes them from the confirmatory design and, where acknowledged at all, confines them to a non-confirmatory exploratory sub-study.

1. Acar, Y. B., & Alshawabkeh, A. N. (1993). Principles of electrokinetic remediation. Environmental Science & Technology, 27(13), 2638–2647. doi:10.1021/es00049a002
2. Acar, Y. B., Gale, R. J., Alshawabkeh, A. N., Marks, R. E., Puppala, S., Bricka, M., & Parker, R. (1995). Electrokinetic remediation: Basics and technology status. Journal of Hazardous Materials, 40(2), 117–137. doi:10.1016/0304-3894(94)00066-P
3. Alshawabkeh, A. N., Yeung, A. T., & Bricka, M. R. (1999). Practical aspects of in-situ electrokinetic extraction. Journal of Geotechnical and Geoenvironmental Engineering, 125(1), 27–35. doi:10.1061/(ASCE)1090-0241(1999)125:1(27)
4. Reddy, K. R., & Cameselle, C. (Eds.). (2009). Electrochemical Remediation Technologies for Polluted Soils, Sediments and Groundwater. John Wiley & Sons. doi:10.1002/9780470523650
5. Schulman, J., Wolski, F., Dhariwal, P., Radford, A., & Klimov, O. (2017). Proximal policy optimization algorithms. arXiv:1707.06347. arXiv:1707.06347
6. Probstein, R. F., & Hicks, R. E. (1993). Removal of contaminants from soils by electric fields. Science, 260(5107), 498–503. doi:10.1126/science.260.5107.498
7. Cang, L., Zhou, D. M., Wang, Q. Y., & Fan, G. P. (2007). Impact of electrokinetic remediation on microbial community in heavy metal-contaminated soil. Journal of Hazardous Materials, 146(1–2), 294–302. doi:10.1016/j.jhazmat.2006.12.019
8. Zhang, X., Wang, Z., Cheng, S., & Wu, B. (2024). Sustainable remediation of soil and water utilizing arbuscular mycorrhizal fungi: A review. Microorganisms, 12(7), 1255. doi:10.3390/microorganisms12071255
9. Stewart-Oaten, J., Murdoch, W. W., & Parker, P. G. (1986). Environmental impact assessment: Pseudoreplication in time. Ecology, 67(4), 964–969. doi:10.2307/1940823
10. Underwood, B. J. (1994). Beyond the restoration paradigm. Ecological Applications, 4(3), 541–549. doi:10.2307/1942110

11. Lindsay, W. A. (1978). Stability relations of soil minerals. Soil Science Society of America Journal, 42(4), 660–672. doi:10.2136/sssaj1978.03615995004200040002x
12. Sposito, F. (1981). The Chemistry of Soils. Oxford University Press.
13. Bohn, R. K., & Fendorf, J. (2009). Soil chemistry. In Soil Chemistry (pp. 343–387). Elsevier. doi:10.1016/B978-0-444-63584-4.00010-9
14. Smith, S. E., & Read, D. J. (2008). Mycorrhizal Symbiosis (3rd ed.). Cambridge University Press. doi:10.1017/CBO9780511545562
15. Lin, H. (2010). Earth's critical zone and hydropedology: Concepts, characteristics, and advances. Hydrology and Earth System Sciences, 14(1), 25–45. doi:10.5194/hess-14-25-2010
16. Lal, R. (2004). Soil carbon sequestration impacts on global climate change and food security. Science, 304(5677), 1623–1627. doi:10.1126/science.1097396
17. Six, J., Conant, R. T., Paul, E. A., & van Littere, M. (2002). Stabilization mechanisms of soil organic matter: Implications for C-saturation of soils. Plant and Soil, 238(1–2), 59–76. doi:10.1023/A:1018045019093
18. Bareau, I., Callewaert, G., & Cornelis, G. (2018). pH buffering capacity of biochar and its effect on the bioavailability of metals in soil. Environmental Sciences Europe, 30(1), Article 10. doi:10.1186/s12302-018-0137-6
19. Uchimiya, M., Lima, I. M., Klasson, K. T., & Wartelle, L. H. (2010). Contaminant immobilization and nutrient release by char for environmental applications. Chemosphere, 80(8), 935–940. doi:10.1016/j.chemosphere.2010.05.020
20. Zhu, J., & Holthausen, M. M. (2020). The effect of biochar on soil microbial communities and their function: A meta-analysis. European Journal of Soil Biology, 100, Article 103256. doi:10.1016/j.ejsobi.2020.103256
21. Qu, W., & Wang, W. (2016). Biochar and soil ferrous/manganese oxide transformations and the effect on metal immobilization. Journal of Environmental Management, 171, 87–95. doi:10.1016/j.jenvman.2015.12.013
22. O'Carroll, D., Sleep, B., Krol, M., Boparai, H., & Kocur, C. (2013). Nanoscale zero valent iron and bimetallic particles for contaminated site remediation. Advances in Water Resources, 51, 104–122. doi:10.1016/j.advwatres.2012.02.005
23. Lehmann, J., & Joseph, S. (Eds.). (2015). Biochar for Environmental Management: Science, Technology and Implementation (2nd ed.). Routledge. doi:10.4324/9781316438530
24. Tiwari, J., Ma, Y., & Bauddh, K. (2022). Arbuscular mycorrhizal fungi: An ecological accelerator of phytoremediation of metal contaminated soils. Archives of Agronomy and Soil Science, 68(3), 283–296. doi:10.1080/03650340.2020.1829599
25. Salvalaio, M., Oliver, N., Tiknaz, D., Schwarze, M., Kral, N., Kim, S.-J., & Sena, G. (2022). Root electrotropism in Arabidopsis does not depend on auxin distribution but requires cytokinin biosynthesis. Plant Physiology, 188(3), 1604–1616. doi:10.1093/plphys/kiab587
26. Zhao, Y., Gao, Y., Mao, J., Wang, Z., & Xu, Z. (2021). Wearable and implantable sensors for plant health monitoring. ACS Sensors, 6(5), 1735–1748. doi:10.1021/acssensors.1c00280
27. Bolt, G. H., & Bruggenwert, B. M. G. M. (2002). Soil Chemistry: A. Basic Elements (2nd ed.). Elsevier.
28. Saxton, K. E., & Rawls, W. J. (2006). Soil water characteristic estimation by machine learning. Soil Science Society of America Journal, 70(4), 1032–1040. doi:10.2136/sssaj2005.04.0239
29. Millington, R. J., & Quirk, J. P. (1965). Permeability of porous media. Nature, 206(4980), 479–480. doi:10.1038/205479a0
30. Karpatne, A., et al. (2019). Towards a universal model for the prediction of soil properties. Proceedings of the 25th ACM SIGKDD International Conference on Knowledge Discovery & Data Mining, 1347–1357. doi:10.1145/3292500.3330669
31. Goh, K. H., Lim, T. T., & Dong, Z. (2008). Application of layered double hydroxides for removal of oxyanions: A review. Water Research, 42(6–7), 1343–1368. doi:10.1016/j.watres.2007.10.043
32. LoRa Alliance. (2023). LoRaWAN Specification Document (LoRaWAN 1.0.4 / 1.1). LoRa Alliance. Open specification
33. IPCC. (2019). 2019 Refinement to the 2006 IPCC Guidelines for National Greenhouse Gas Inventories. Intergovernmental Panel on Climate Change, Geneva.
34. Sanderman, J., Joerg, S. E., Havlin, J. S., Socolar, J., Elbert, S. L., & Glaser, B. (2017). A climate debt of 12,000 years. Science Advances, 3(10), eaa1348. doi:10.1126/sciadv.aa1348
35. Hernandez-Orallo, J. H., & Fernbach, C. (2019). The Measure of Minds: Evaluating the Cognitive Capabilities of Machine Learning Models. Cambridge University Press. doi:10.1017/9781108608809

Note on reference scope. This bibliography is a curated, representative evidence base sized for a proposal document rather than an exhaustive systematic review. Entry 29 is an open technical specification rather than a peer-reviewed source; it is included because LoRaWAN topology and node power budgets determine the feasibility of the telemetry mesh. Full retrieval will be completed under a systematic search protocol (databases, date range, inclusion and exclusion criteria) at programme commencement, and that protocol will be pre-registered. Note that the two entries supporting H6 (25, 26) document genuine electrotropic and plant-sensing phenomena at field strengths orders of magnitude above the sub-Faradaic regime; they are cited in support of the programme's position that H6 is not established, not in support of it.