Publications

The submitted LL.M dissertation. It defines economic warfare against the law of armed conflict, works case studies in the Democratic Republic of Congo, Nigeria and Yemen, treats private military and security companies and cyber-economic aggression, examines SAMLA and the Magnitsky legislation, and proposes a United Kingdom response. Retained here as the thesis of record, unaltered.

A revised publication edition of the 2024 dissertation. It keeps the thesis, the chapter sequence and the voice of the submitted text, and much of its wording; the abstract, the figures, most of Chapter 2, sections 3.2 and 4.1–4.2 and the statutory scheme in 6.2 are new or substantially rewritten, and the positions on the Magnitsky Act, humanitarian law in cyberspace, non-State involvement in conflict and the use of force are revised. Material first available after submission, including the March 2024 attribution concerning APT31, is marked as such rather than written back into the original chronology. Papers 1 to 6 develop every major branch this one opens.

Replaces the embargo-centred conception of economic warfare with a functional account of privately controlled communications, capital, imagery, targeting and infrastructure. Starlink is treated chronologically, including assistance rendered before the formal contract; the absence of a mutual-defence treaty is distinguished from the law of collective self-defence; and cyber operations are classified by function and effect rather than by the presence of a keyboard.

Treats autonomy as an allocation of decision authority rather than a property of a model. Six build architectures and six end-to-end workflows — State–PMSC commissioning, subcontracted and time-zone-separated coding, modular development with no single informed programmer, open-weight deployment, jurisdiction-selected compute, and autonomous continuation after political withdrawal — joined to State responsibility, due diligence, criminal and humanitarian law.

Distinguishes the victim's assembled campaign from each territorial State's partial view. Refuses strict liability for mere territorial transit; develops notice, knowledge, capacity and feasible protection; shows how sharding may be used deliberately to manufacture ignorance; and places the correlation burden on the controller that retains the operational map.

Answers the operational question Paper 3 leaves open, as a specified service rather than a slogan: architecture, data and evidential models, chain of custody, contrary hypotheses, manifests, correction, preservation, hostile-node testing, United Kingdom legal integration, a twelve-month pilot, a three-year core programme, procurement structure and staged acceptance gates.

A taxonomy of infrastructure, intelligence, PMSC, platform, intrusion and autonomous-capability sovereigns. Treats reciprocal risk, consumer and shareholder exposure, the two distinct Lebanese campaigns, beneficial ownership, subcontract chains, maximum reasonably credible harm and the economics of cheap war, and proposes a Strategic Private Capacity Act built on speed, control maps and the smallest effective dependency.

Converts the analysis into a consolidated statutory scheme. Preserves speed as a constitutional requirement; integrates reporting, an incident register and MOSAIC; regulates strategic private capacity and PMSCs; distinguishes payer, operator, contractor, territorial host and State; and constrains the power by parliamentary notification, rapid review, live-target verification, expiry, correction, remedy and compensation without making immediate action impossible.

The law of decision, which the series otherwise lacked. Separates technical, legal and decisional attribution; builds a traceable claim graph and correction lifecycle; prescribes different thresholds for preservation, confidential engagement, public attribution, sanctions, civil restraint, criminal process, countermeasures and force; and closes with operational tests, a threshold matrix and an Attribution Decision Certificate.

Labour displacement is answered fiscally and dangerous modification is answered by market-entry rules, and neither instrument can do the other's work: a tax cannot deliver a quantity and confers no right of access, while a market-entry rule is discharged before the risk it manages has arisen. The paper proposes that the state retain title and pass possession under a lease. The quantity is graduated — possession as of right up to a household entitlement, a presumptively granted application band above it, a binding cap only at industrial scale — and the middle band generates the deployment data from which the binding band is calibrated, which no fiscal instrument can do. Stating the defence limb plainly exposes the paper's most serious finding: “purpose” in the definition of a military objective means intended future use, so a standing statutory reservation over every unit of a class arguably confers a military purpose on all of them — and raises a question about the dwellings holding them — before any emergency. The contractual civil-reserve arrangements on which the paper draws escape this by accident of location; aircraft sit at airports and ships in ports. A reserve held in housing does not. 27,009 words, 65 notes.

Switzerland already holds a civil reserve in private hands, and holds it the harder way: compulsory stockpiling imposes a contractual obligation on some 280 undertakings and the Confederation guarantees their bank financing. The retained-interest model obtains the same reserve by inverting the property relation instead — no obligation to impose, no financing to guarantee, and capability in continuous use rather than commodities held idle. The submission works the comparison against art. 102 and art. 26 of the Federal Constitution, the Landesversorgungsgesetz, cantonal labour-market differentiation, and the Mutual Recognition Agreement.

Amended Protocol II imposes five obligations on emplaced devices — recording, marking, self-destruction, detectability and clearance after hostilities. An armed robotic platform emplaced for thirty days escapes all five, because the definition turns on a fuze: a mine is exploded by presence, while a robotic system observes presence and decides. Four of the five obligations are wholly indifferent to whether the device can classify what triggers it. The regime asks less of these systems than they can deliver, because it asks nothing of them at all. 11,956 words, 73 notes. Version 2 reproduces the body of the version of record unchanged and adds a postscript, with five further notes, on the report the CCW Group of Governmental Experts adopted on 4 September 2026 (CCW/GGE.1/2026/3): the autonomous-weapons process has let go of both mine-law measures its December 2025 text carried, and what survives is a measure of intervention owed only “to the extent possible or feasible”.

Ten years of multilateral effort to define autonomy in weapons has produced no instrument, and the paper argues the failure is a property of the mandate rather than of the disagreement beneath it. It proposes regulating the interval between the last human decision and the effect, together with what was knowable across it, and shows that the rate at which the facts underlying an authorisation decay is what should set the lawful interval. Article 36's temporal reach is broad across procurement and zero across operational life — the breadth runs along the wrong axis. 17,676 words, 94 notes. Version 2 reproduces the body of the version of record unchanged and adds a postscript, with twelve further notes, on the report the CCW Group of Governmental Experts adopted on 4 September 2026 (CCW/GGE.1/2026/3): the review provision it carries is Article 36 restated at the four pre-employment stages, the re-review of modified systems that the drafts carried was dropped, and the timing of the attack now appears in the sentence that defines what the characterisation excludes.

Endurance for a fielded ground robot is published as a drive-hour figure. The mission examined here — holding ground — is defined by not moving, so drive hours do not measure it. The paper proposes watt-hours per emplaced-day at a stated readiness level as the quantity in which an emplaced armed system should be specified, and computes it for a thirty-day emplacement from an eleven-parameter vector. “Offline” conflates two independent properties — independence from third-party services, and not emitting — and only the first is architectural. An emplaced armed platform should be architected as an unattended ground sensor that has effectors rather than as a vehicle that has stopped; the resulting budget is 1.22 watt-hours per emplaced-day, against roughly 360 for a conventionally architected platform. Ninety-one per cent of that budget is bought by identification and release judgement: holding ground without discriminating costs 110 milliwatt-hours a day, and looking first costs about eleven times as much. A denial test shows that a stop-command channel is fail-dangerous under communications denial, whereas a periodically renewed permission-to-continue channel safes the platform after a bounded interval. Version 2.12.4 updated the treaty analysis to the report the CCW Group of Governmental Experts adopted on 4 September 2026 (CCW/GGE.1/2026/3). Version 2.12.5 restates as targets and plans reported figures that had been given as accomplished, withdraws figures not found in the sources cited for them, narrows three statements about the reported employment of ground robots, and corrects the earlier-title note and seven cross-references; no model input, equation or numerical result changes, and the reproducibility archive of v2.12.2 remains controlling. Versions through v2.10.0 circulated as Holding Ground with Machines That Do Not Move; the two published legal companions refer to it, without a title, as the companion paper on closed-system ground robotics.

Argues that Article IV of the 1967 Outer Space Treaty is an artefact of a Cold-War bargain that prohibits weapons of mass destruction in orbit but says nothing about conventional, kinetic or dual-use systems, leaving “orbital denial” legally contested rather than clearly prohibited. Through an interpretive analysis of Article IV, a mapping of where kinetic, conventional and anti-satellite systems fall in law, an account of the liability, registration and debris regime, and a survey of the turn since 2022 towards “responsible behaviours”, it argues that the gap is best filled not by a new convention but by targeted, nationally anchored measures built on the authorisation-and-supervision duty that Article VI already imposes, each resting on its own domestic legal basis.

Article IV of the 1967 Outer Space Treaty does not comprehensively prohibit conventional weapons, anti-satellite capabilities or all military activity in outer space — but that proposition does not create a legal vacuum. This draft reassesses the proposal made in version 1.0 (July 2026) of Orbital Denial to anchor national restrictions on irresponsible counterspace conduct in Article VI, finds Article VI narrower than that proposal suggested, and replaces it with nationally anchored market conditionality: licensing, export controls, financial sanctions, procurement, insurance and government-backed finance, each on its own legal basis, triggered by conduct rather than capability and bounded by published criteria, independent review and proportionality.

The companion legal paper that ARGUS-PD/POD-7K Paper II defers to. Characterises the hypothetical system under Article IV of the Outer Space Treaty (conventional-kinetic research outside the Article IV(1) WMD prohibition; Article IV(2) celestial-body limits regardless of civil labelling), treats Article VI authorisation and continuing supervision as the operative regime, analyses registration and liability for objects launched from an orbiting platform, engages the argument that existing law already prohibits debris-creating destructive testing, and proposes an independent-oversight model in which authorisation advances only through the technical series’ own published evidence gates — offered as policy consistent with existing law, not as existing law.

Argues that AI has become the “perfected proxy” — a deniable, jurisdictionally slippery, self-executing intermediary through which actors circumvent international-legal norms while remaining formally unaccountable — and that the answer is not a new global treaty but its opposite: localised, targeted national legislation on the Magnitsky model. It theorises AI as an accountability-diffusing “double veil” (the corporate veil plus the algorithmic responsibility gap), explains why targeted designations cascade through market access, dollar-clearing, correspondent banking, compliance de-risking and legislative emulation, and proposes an original doctrinal mechanism of AI-conduct-triggered designation keyed to responsible control rather than to a completed, attributable wrong. It is tested against three worked scenarios and against the strongest objections — sovereignty, over-reach, fragmentation and due process.

Generative systems act on the cost of generating a claim and leave the cost of checking it substantially unchanged. Where a claim is cheap to make and dear to verify, its cost has been relocated to the checker rather than reduced — and where verification cost exceeds what a checker can bear, the checker stops verifying and adopts a proxy uncorrelated with truth. The paper defines a verification ratio, derives “verification drag” as a mechanism by which a real productivity improvement registers as a measured slowdown, and predicts a movement of the firm boundary: functions whose output is an unverifiable assertion should be internalised, not outsourced. 23,350 words, 159 notes.

Every instrument the international legal order has for responding to economic coercion — the prohibition on intervention, state responsibility, countermeasures, the security exceptions of the trading system and the sanctions built on them — presupposes a provable author. The paper argues that the presupposition no longer holds. Five of the six legal routes fail at the point of proof; the sixth, creeping expropriation, reasons from effect but fails on standing and forum. Two WTO disputes the European Union brought against China mark the boundary: what decides whether conduct can be established is whether it leaves a documentary trace in the hands of someone other than the acting state. Where attribution falls below the point at which a proportionate countermeasure can deter, the countermeasure channel closes but untargeted retorsion, which needs no attribution, does not, so retaliation migrates to instruments that cannot distinguish the responsible from the rest. A study of 2024 trade at tariff-line resolution, reduced to 323,929 importer-product cells, finds severe chokepoints in production inputs rare — 1,533 cells, 0.77 per cent of input trade by value — and reciprocal severe dependency no more common than the degree distribution predicts. The paper closes with an evidential mechanism modelled on Corfu Channel, which shifts an evidential burden but not the legal burden or the standard of proof. 27,075 words, 139 notes.

A domain-general methodology for estimating cost, schedule and risk on very large, capital-intensive autonomous and infrastructure programmes before commitment. It integrates a work-breakdown-structure CapEx estimate with explicit uncertainty and learning-curve effects, critical-path scheduling under deep uncertainty via an event-sourced engine (Helm) with Monte Carlo finish-date distributions, explicit go/no-go feasibility gates, a procurement layer (make-versus-buy, long-lead management), and an integrated feasibility index triangulated against reference-class benchmarks. A fully de-identified worked illustration on a generic autonomous-freight network demonstrates the method end to end; all numbers are notional and clearly labelled.

Why a regulated AI programme has to answer two legal questions at once — is it lawful to build and use (EU AI Act risk-tiering) and is it lawful to move (EU Dual-Use Regulation 2021/821 and Wassenaar export screening) — and how Sentinel answers both in one deterministic pass that makes no network calls. The note sets out the rules-engine method, in which every reason line carries one of four tags — legal fact, assessment, verify or action — and every position is dated and pinned to the Digital Omnibus on AI (Regulation (EU) 2026/1744); the “regulated-by-design” thesis (human authority, auditability and export screening built into the control layer); and explicit scope limits — decision-support, not legal advice, and not an official control classification.

A working reference architecture for financial-crime and regulatory analysis built on three commitments: an offline, dependency-free client — autonomous in the sense that it needs no server, external service or network connection — where no subject data leaves the analyst's machine; a deterministic “peel-chain” method that traces layered value transfers as exact integers under a checked value-conservation invariant, byte-for-byte reproducible from its inputs; and a provenance-first data model that labels every value by source and surfaces gaps rather than imputing them. It specifies the peel-chain algorithm and a 23-module engine framework that applies a common deterministic discipline across beneficial-ownership tracing, sanctions and adverse-media screening, an insolvency-recovery waterfall, and a structured compliance knowledge base. No performance benchmarks are claimed; every numeric example is labelled illustrative.

The methodology of Meridian, a provenance-first research system that maps real-time regulatory instruments and changes onto a universe of roughly 1,300 issuers and infers potential — not measured — impact. Every value carries its own source key, timestamps, hashed raw artifact and explicit verification class; missing values are shown as missing, and an instrument is linked to a company only through a quotable, sourced business fact. It also sets out an operating discipline for strict API budgets — a daily-cap governor, need-driven rotation, and an “outage is not absence” rule — and reports no empirical findings, only illustrative mechanism.

A dependency-free forecasting engine implementing the classical exponential-smoothing family (SES, Holt, additive Holt-Winters), OLS regression with seasonal dummies, and Croston's method, with seasonal-naive and moving-average baselines. Rather than committing to one model, it performs per-series rolling-origin (walk-forward) backtesting and selects a champion by lowest held-out error across five commercial domains with different cadences. The finding is that model choice should be left to each series, with its margin and sensitivity reported: Croston wins one of three intermittent lines, seasonal-naive the one whose zeros are almost all scheduled, and between 13 and 20 of 31 champions change with the backtest's horizon or number of origins — using semi-synthetic, clearly-labelled benchmark data.

A practitioner’s handbook for the person handed a problem nobody owns in an organisation too small to have a department for it — a charity, a community interest company, a training provider, an owner-managed business or a single-asset project. Part I sets out a method that does not depend on the domain: taking the call and setting the mandate, including the operator’s own liability and pay; the first seventy-two hours; diagnosis of the decisive constraint; governance and controls that survive an audit; people; regulators, funders and money; the operator’s own judgement; and leaving. Part II applies it to three domains with the law stated as at October 2026: crisis and cash in English charities and companies (the creditor duty after Sequana, wrongful trading, the moratorium, restricted funds, the thirteen-week cash flow, creditor strategy and exits); compliance architecture for a publicly funded training provider (culture, the obligations pipeline, AI, funding assurance, the Data (Use and Access) Act 2025, safeguarding and speak-up); and capital and structure for a small solar project in the Philippines (venture or asset, technical normalisation, the project vehicle and lease, risk allocation, debt sizing, and integrity as a financing condition). Each chapter of Part II ends with a worked case built on invented organisations and figures, with a wrong turn, a dilemma the method did not settle and its cost. Appendices provide checklists, templates, dated regulatory maps and a reading list mapped against the standard playbooks. Not legal, financial or investment advice.