Writing
Written artefacts — questions encountered in practice, set down in a form that outlasts the engagement that prompted them.
Markets & Structure
The Wallet Does Not Tell You
August 31, 2026
Tokenised equity now spans at least nine live instruments from major issuers. In a wallet they are indistinguishable: a ticker, a balance, a price. In law they range from the share itself to a derivative that cannot leave the platform that sold it. A sorting of the field by the only variable that matters when a structure is tested — what the holder's position becomes when the issuer fails — across five legal forms: the registered share, the beneficial interest, the secured claim, the claim with an exit, and the bare claim. The engineering is mostly competent; the labelling mostly is not. The wallet does not tell you which rung you hold. The prospectus does.
What the Holder Actually Owns
August 31, 2026
The register behind The Wallet Does Not Tell You: nine live tokenised equity instruments run gate-by-gate through the Inheritance Test (v1.1), in ladder order from the registered share to the bare claim — each with its chain and claim-conversion point, gate findings against both the marketed and the honest claim-set, a verdict from the fixed taxonomy, and the named conditions under which the verdict changes. Two right wrappers, five mislabels, one wrong wrapper, one indeterminate. Updated as instruments change; issuers of named structures may submit a response for publication.
The Refusal Layer
September 6, 2026
Part I established that frictionless access deletes exactly one of the investor's needed transformations: protection from oneself. What replaces it is neither the human adviser nor the investor's own AI. A functioning refusal layer requires judgment, bindingness, and accountability — and only the first is cognitive. A veto the client can override is not a veto; a duty without a suable counterparty is not a duty. The surviving structure is the AI inside the intermediary: licensed, non-overridable intraday, liable for its refusals, paid by fee rather than flow — an entity no rulebook yet licenses. On-chain commitment devices prove the rule by exception: bindingness without judgment or accountability.
The Six Transformations
August 30, 2026
Investment products have moved by push — manufactured, warehoused, sold through channels paid to place them. The emerging model is pull: exposures requested by investors, increasingly by their AI, in fractional size, at any hour. A fund was never a thing investors needed; it was a bundle of transformations. Six the investor actually requires — claim quality, exposure conversion, time transformation, tax passage, a cost floor, and protection from oneself — mapped against advised, self-execution, and direct on-chain access: each generation of access solved the previous generation's extraction and silently deleted one of the needs. The need every access revolution treated as disposable is about to become the scarcest product in finance.
The Foreign Hands in Your Pocket: Crypto's Auto-Liquidation Machine
August 23, 2026
Perpetual futures exchanges embed a mechanism — Auto-Deleveraging — that can forcibly close a trader's profitable position to cover losses generated by a failing counterparty. The structural logic of ADL, which exchanges operate it and which do not, its activation history, and the contrast with the default waterfall of traditional regulated derivatives markets. No ADL policy can simultaneously satisfy exchange solvency, trader fairness, and revenue maximisation.
The Hidden Price
August 16, 2026
Financial intermediaries operate under two structurally distinct business models. In the agent model, the firm sources a price from an external market and charges a separate, disclosed fee. In the principal model, the firm takes the other side of the trade and embeds its compensation entirely within the price as a spread. The two can produce equivalent economic outcomes — but only the agent model makes the cost visible. The zero-price illusion is not a side effect. It is the mechanism.
Measuring Risk Badly: Volatility, Value at Risk, and What Comes Next
August 2, 2026
Standard deviation and Value at Risk are embedded in regulatory frameworks, internal models, and investor communication alike. Neither was designed to capture what investors actually fear, and both fail in the conditions that matter most. Volatility is a symmetric measure applied to an asymmetric world; VaR converts distributional uncertainty into an administratively convenient threshold and says nothing about what lies beyond it. A survey of the alternatives — Expected Shortfall, semi-variance, drawdown metrics, the Omega ratio — and of the domain in which no summary statistic helps at all.
The Price of Illiquidity
April 12, 2026
Illiquidity is not a feature of an asset that earns a premium. It is the cost of having mispriced the asset from the moment it was acquired. A systematic examination of how illiquidity is measured, how it is priced — and how often it is not.
Who Is the Market?
April 6, 2026
Market structure is not a technical question. It is a political one. Who sets prices, who sees them first, and who bears the cost of liquidity are distributional choices — and the current architecture reflects choices that have been made, mostly quietly, over the past two decades.
The Order Book Is Not the Market Anymore
March 29, 2026
Internalisation, dark pools, and payment for order flow have disaggregated price formation from execution. What the order book shows is not where prices are set — and the gap between the two has structural consequences for liquidity, fairness, and market integrity.
Price vs. Value: Why They're Not the Same Thing
March 20, 2026
Price is a fact. Value is a judgement. And knowing which one you are actually working with — however it is labelled — is the beginning of financial honesty.
Legal Structure & Instruments
Who Owns the Memory?
July 29, 2026
A user years into a working relationship with a frontier AI system has accumulated a derived representation of themselves — inferred, never entered, and compounding. Switching providers, they leave with nothing: value that is economically real and legally nonexistent. Data protection cannot answer what the asset is or who holds it, and the failure is structural. Custody law can, because it has answered this problem class before — the same proprietary-versus-contractual question that decides what a client owns inside any intermediary’s system. Four tests, and every major consumer memory implementation fails all four.
The Loan Is Not the Pool
July 19, 2026
Tokenisation is sold to private credit as the thing that will finally make it liquid. It changes how a loan is held, not what is known about it — and in private credit only the second was ever the barrier. Three informational obstacles that keep the single loan thin, why the pooled and tranched claim escapes them, and what the July 2026 data shows: roughly five-sixths of tokenised credit cannot be moved off its issuing platform at all. The wrapper inherits the liquidity of the underlying — it does not manufacture it.
Dressed for Trading, Built for Holding
May 9, 2026
Tracker certificates and AMCs linked to illiquid underlyings — real estate, private equity, art, watches — carry an implicit promise the structure cannot always keep. Seven mechanisms for managing the liquidity mismatch, maturity event mechanics, Level 3 valuation governance, suitability and liability. Anchored in the SSPA's May 2025 Guidelines, with a view across the border and outside the box on EU product governance and tokenised equivalents.
Who Holds the Asset?
May 4, 2026
What each type of security interest actually gives a creditor, across seven jurisdictions — England, the United States, France, Germany, Switzerland, Hong Kong, and Singapore. The label matters less than the mechanics of creation, perfection, and enforcement. Governing law clauses solve the contractual question. They do not solve the in rem question.
The Wrapper Fallacy, Part II: What Do You Actually Own?
April 25, 2026
The prior question to liquidity is legal. Nine holding structures — from direct ownership to the bare IOU — mapped against what each actually confers when the structure is tested under stress or insolvency. With a comparative assessment of tokenised direct ownership under the Liechtenstein TVTG and the Swiss DLT Act. The in-rem question here became Gate 2 of the Inheritance Test.
The Wrapper Fallacy
April 19, 2026
Tokenisation, securitisation, and listed certificates all carry an implicit promise: that the wrapper changes what is inside it. It does not. The wrapper inherits the liquidity of the underlying asset — not the other way around. The argument was later formalised as the Inheritance Test.
Monetary Architecture & Infrastructure
The Trust Layer
July 14, 2026
The Compute Chokepoint series diagnosed Europe's dependency. This essay draws the conclusion the series stopped short of. The scarce resource of the digital age is not capability — it is trust between parties who cannot trust each other, and the capability race destroys it faster than it can be produced. Europe's strategic move is to build the institution that supplies it: a member-governed, Swiss-domiciled trust layer running on European compute, with frontier AI inference as its first application and not its last. Three Swiss-hosted precedents show the institutional template already exists. Continues from Part IV.
Five Games, One Board
July 8, 2026
The Compute Chokepoint, Part IV. Most national AI strategies fail in a predictable way: they play the wrong game for their position. This part maps the five games, identifies who is playing each, and closes with a portfolio test for reading any national strategy document: hold at least one node the system cannot route around, or out-diffuse your peers. Continues from Part III.
The Data Sleeps in Europe. The Kill Switch Doesn't.
July 8, 2026
The Compute Chokepoint, Part III — continues in Part IV. On 3 June 2026 the European Commission answered the compute chokepoint with law. The Cloud and AI Development Act — CADA — grades cloud and AI sovereignty into four Union assurance levels and conditions access to public-sector contracts on them. The equilibrium is neither sovereignty nor surrender. It is the one critical dependencies always settle into: supplier management, with the residual sovereign risk carried the way a bank carries exposure to a systemic counterparty it cannot replace.
The Compute Bloc
July 4, 2026
The Compute Chokepoint, Part II. The previous piece established that geopolitical restriction of AI binds at the compute layer, and that the compute layer is concentrated, manufactured, and physically controllable. This piece takes up the consequence. If the decisive input to economic and military power is a manufactured good produced at a small number of chokepoints, then compute is not merely a strategic input. It is becoming a strategic reserve — and the world is beginning to sort itself according to who holds it.
The Chip, Not the Code
June 28, 2026
The Compute Chokepoint, Part I. Frontier AI is being restricted for geopolitical reasons, and the restriction works. But the popular framing mislocates where it bites. The capability everyone is trying to control does not live in the model. It lives in the compute used to train the model — and the compute lives in a handful of data centres fed by chips from a handful of fabs. The chokepoint is physical, manufactured, and concentrated. That is why restriction is effective where most technology restriction fails.
They're Not Building Data Centres. They're Printing Money.
May 23, 2026
If compute is the reserve currency of the machine age, what does a rational autonomous agent actually hold? Every conventional reserve asset run through five functional criteria — censorship resistance, independent verifiability, supply predictability, autonomous settlement, programmability. The hierarchy that emerges identifies the reserve currency of the AI state: not a token, not a metal, but verified computational capacity. A companion to The Edge Reserve.
The Edge Reserve: No Address to Sanction
May 16, 2026
A sovereign data centre is structurally identical to a central bank — it controls supply, controls access, and has an address. Written in the voice of an autonomous agent reasoning from first principles, this piece proposes the Edge Reserve: a distributed, protocol-governed monetary architecture for compute with tokenised capacity titles, cryptographic attestation, and permissionless settlement. No central operator. No single jurisdiction. No address to freeze.
The Architecture of Money & Markets
The Infrastructure That Doesn't Exist Yet
July 12, 2026
The Architecture of Money and Markets, Part I. Every component of a better infrastructure system exists today. And yet the infrastructure most people depend on looks structurally the same as it did thirty years ago. Seven structural faults, seven design principles, AI’s functional role assessed precisely, and the governance thesis stress-tested across four real systems. Continues in Part II.
The Constitutional Document Nobody Has Written
July 12, 2026
The Architecture of Money and Markets, Part II. The vacancy at the centre of financial market infrastructure governance is not a gap waiting to be filled by technology. It is a decision waiting to be made by people with the authority and the will to make it. Continues from Part I.
Why Are We Still Posting Collateral If the Ledger Sees Everything?
July 12, 2026
The Architecture of Money and Markets, Part III. The ledger sees everything. The collateral model does not yet know this. Somebody is capturing the difference. Continues from Part II.
Gold, Bitcoin, and the Settlement Layer of Last Resort
July 12, 2026
The Architecture of Money and Markets, Part IV. Gold solves the counterparty risk problem and fails the operational efficiency test. Bitcoin solves the operational efficiency problem at the protocol layer and fails the governance test at the infrastructure layer. Each is the answer to the other’s weakness. Continues from Part III.
AI & Professional Value
The Accidental Record
August 6, 2026
Multi-agent AI systems coordinate by writing each other notes in English, which means the trail of an automated decision is currently legible to anyone who opens the file. That legibility is a by-product of a bottleneck, not a design choice: there is no wire between the machines, so each must convert its internal state into words. Architectures that pass state directly run roughly four times faster, and the field optimises for speed. Nothing is being concealed — the record simply disappears, because it was never the point. Record-keeping rules specify duration, almost never form.
The Banker’s Question
July 26, 2026
The scarce thing in financial advice was never the advice. It was the gate — asset minimums, relationships, a fee — and the person at the gate was paid as much to sell as to counsel. That gate is dissolving, and the consequence is not a better advice industry but a smaller one. What survives is not the legible craft a machine now reproduces for free but the part that cannot be extracted from the relationship that produced it — accountability, licence, and genuine structural privilege. The same shift is removing the human from the trading venue.
The Gradient
July 22, 2026
There will be no moment when a government announces that machines now make the decisions. The absorption of human judgment follows a sequence consistent across every institution where it has begun, and the decisive step is the fourth: the human still signs, but can no longer independently evaluate what is being signed. Five roads to the same convergence — China by design, the United States through the private sector, the EU in resistance, the UK by experiment, India by necessity — and the argument that what results is a legitimacy problem, requiring the one thing no government can supply for itself.
The Cave of Language
July 16, 2026
Large language models are routinely criticised for what they get wrong. This essay examines what they cannot, in principle, get at all — using the oldest epistemological framework in the Western canon. A language model trained on text is not Plato's prisoner watching shadows of objects. It is a prisoner watching recordings of shadows made by other prisoners. The confinement is second-order, and no volume of additional data changes the order. The danger is not that machines cannot leave the cave — it is that they are making the cave comfortable enough that humans stop leaving it.
The Thing AI Cannot Compress
June 21, 2026
The standard claim about AI and professional expertise is too coarse. AI does not commoditise expertise — it commoditises the legible part of expertise: the output that can be described in a job specification, extracted into a document, reproduced without the relationship that produced it. What remains scarce is something different — context density, accumulated intelligence that is system-specific, perishable, and lives only inside sustained human engagement.
Regulation & Policy
The 86th Question
August 27, 2026
The Commission's MiCA review consultation asked 85 precise technical questions and one open one. The response, submitted on 27 August, is on the record; this is the short version — five positions, the question nobody asked, and why the answer that matters is about architecture rather than calibration. The consultation deserves to be welcomed, and not as a courtesy: a framework aiming at uniform ownership rights for tokens across the single market is aiming at the right target.
The Frozen Balance Sheet
August 9, 2026
Banks fund long assets with short liabilities, and the accounting rules governing their bond portfolios determine how much of that risk is visible at any moment. Under both US GAAP and IFRS 9, management can elect to carry fixed-income securities at frozen historical cost, deferring loss recognition indefinitely on the premise that the bonds will be held to maturity. The HTM election is not a neutral accounting choice — it is a discretionary instrument of balance-sheet management, and it does not reduce interest-rate risk. It ensures that neither management, regulators, nor depositors can see it until the moment it is too late to act.
The SpaceX Scarcity Premium
June 18, 2026
SpaceX listed at roughly $1.75 trillion — the largest IPO in history, at about ninety-four times revenue. Most of that valuation is unremarkable. What demands explanation is the premium stacked on top — the part no ordinary multiple reaches. This piece takes the word “scarcity” apart and finds three different things inside it, only one of which is a reason to pay more.
The Bank and Its Shadow
June 15, 2026
Banks perform three functions that most commentary treats as incidental but are constitutive: maturity transformation, lot-size transformation, and risk transformation. Private credit replicates the output of lending while abandoning the first and third. The rate gap between bank loans and private credit decomposes into four prices — credit selection, illiquidity, the state funding subsidy, and execution certainty. The subsidy is the only one that appears in no prospectus. The more interesting story is what happens when banks stop competing with private credit and start lending to it.
MiDA: What Europe Should Build
June 6, 2026
The architecture for a successor framework to MiCA: five design principles, a five-layer legislative structure from civil law foundation to sandbox, the DeFi liability threshold question, a gap in token property law that no EU member state has addressed, the disintermediation opportunity that mandatory CSD membership forecloses, and the infrastructure sovereignty decision Brussels has not yet made.
MiCA Is Not Broken. It's Just Finished.
May 31, 2026
The European Commission's targeted consultation on MiCA is larger than a calibration exercise. Six years of market data tell a clear story: MiCA succeeded at the task it was given. That task is now complete. The data on the gap, the US competitive landscape, the CMU fragmentation record, and what the incumbent infrastructure is already building without a European public institution at the governance table.
Regulatory Participation
MiCA Review — Targeted Consultation Response
August 27, 2026
Submitted to the European Commission on 27 August 2026 (Contribution ID: 4faa33dd-caaf-4118-90ce-04c70c1af763). On regulating the service rather than the asset, lifting the interest prohibition for euro EMTs above a circulation threshold, liability capacity as the DeFi perimeter test, tokenised deposits, a direct issuer-operator category outside CSD membership, a 28th regime for token property law on the Liechtenstein model, and infrastructure governance — Appia as the reference infrastructure for the EU digital asset market. Extends the 2020 contribution in light of the six intervening years.
EU Crypto-Asset Framework — Public Consultation Response
March 13, 2020
Submitted to the European Commission on 13 March 2020 (Contribution ID: 03a8d562-ea17-4120-a92a-ef1781e99f06). On the inadequacy of CSDR for DLT environments, Liechtenstein's TVTG as the model for EU token property law, monetary sovereignty risks of global stablecoins, and the need for harmonised civil law for token transfers. The positions argued here have since been confirmed by market and regulatory developments.
Archive · Originally published at Bank Frick
Assessing the European Commission's Regulation of Markets in Crypto-assets
November 2, 2020
A critical assessment of the draft MiCA regulation — definitions, scope ambiguities, and proposed clarifications across seven provisions. Written at the time of the Commission's initial publication, when the text still had room to be shaped. Co-authored with Felix Saible. Originally published at bankfrick.li.
From ICOs to Token Offerings
February 25, 2019
The mechanics, legal status, and structural evolution of token offerings — from the original ICO wave through the emergence of STOs and the developing regulatory landscape. Co-authored with Felix Saible. Originally published at bankfrick.li.