Native on-chain issuance stalls at asset servicing, not settlement — and the industry keeps demoing the easy half of the trade
Settlement demonstrations are now routine: tokenised securities against tokenised cash, atomic, on ledgers of every governance flavour. What nobody demonstrates is a rights issue with fractional entitlements, an election deadline propagated through four custody layers, or a withholding-tax reclaim. This essay argues the omission is the industry's most informative fact. The economic product of the CSD-custody complex was never the transfer of securities — it is asset servicing: the machinery of announcements, entitlements, elections, payments, claims, and tax that keeps a security's legal life synchronised with its ledger life. That machinery is the part no protocol has displaced, and the reasons are structural, not temporary.
The boundary is precise and worth drawing: coupon-bearing digital bonds service themselves adequately because they contain no choices. Everything with optionality — elections, entitlements, reclaims — remains anchored in the intermediated stack wherever a stack exists (the closely held, direct-register share services itself for the opposite reason: there is no chain to traverse), because corporate actions are legal events born off-chain, because the claims machinery is calibrated to a settlement interval that continuous settlement deletes, and because the binding constraint is not data distribution but liability for getting an election wrong.
The tokenisation announcement has a stable grammar. A platform, a network, a bank or three; an instrument tokenised; a trade executed; both legs settled — atomically, natively, programmably. September 2026 supplied a clean specimen: the first intra-day repo represented in the Common Domain Model and settled on a permissioned network with a stablecoin cash leg (Tokenovate, 14 September 2026). The same vendor's product architecture, read alongside the announcement, makes the essay's distinction for it: corporate-actions processing is sold as a separate workflow — evidence not that servicing is impossible on-chain, but that settlement and servicing are distinct implementation problems, and only one of them was demonstrated. The trade is the demo. The servicing is the roadmap.
The pattern is general and, once seen, unmissable. Across a decade of pilots, the demonstrated lifecycle runs from execution to settlement and stops. Nobody demonstrates the mandatory-with-options event where an election deadline must reach a beneficial owner through an omnibus chain and return in time; nobody demonstrates the market claim on a trade that straddled the record date; nobody demonstrates relief at source on a cross-border dividend. Quo Vadis catalogued where tokenised markets are being absorbed by incumbent infrastructure; this essay isolates the mechanism doing the absorbing. The industry keeps demoing the easy half of the trade because the hard half is not a settlement problem at all.
Outside post-trade operations, almost nobody can describe the machinery, which is itself part of the moat. A corporate action begins as an issuer decision — a board declares a dividend, approves a rights issue, accepts a merger — and must then traverse a pipeline: the announcement, captured from the issuer's agent and normalised into a golden record; the entitlement calculation, fixing who is owed what as of a record date; for voluntary and mandatory-with-options events, the election window, in which instructions flow down the custody chain to beneficial owners and back up against a deadline; the payment or delivery; and then the residue — market claims where trades straddled the key dates, transformations where pending settlements must be restated over the event, buyer protection where an election right travelled with an unsettled trade (CAJWG market standards; AMI-SeCo corporate events standards).
The domestic cash dividend, the event everyone reaches for as the example, is the trivial case: mandatory, no elections, one currency, one date-pair. Walk instead through a rights issue at one remove of custody. The issuer announces terms; the offer must reach holders several intermediary layers away from the register; entitlements include fractions that must be rounded, sold, or compensated under rules that differ by market; lapsed rights have a value that someone must realise; and every step runs against a deadline that shortens at each layer of the chain, because each intermediary needs processing time before passing instructions upward. The event is not a payment. It is a small, time-boxed logistics operation with legal liability attached to every hop.
The domestic dividend everyone cites is the trivial case. The rights issue nobody demos is the product.
The first structural reason the protocol stalls here: corporate actions are legal events, and they are born off-chain. A dividend exists because a board resolved it under company law; a merger exists because courts and regulators approved it; a tax rate applies because a treaty and a domestic statute say so. A ledger — any ledger, distributed or central — can distribute these facts; what it cannot supply, by itself, is the legal authority that originates them or the responsibility that adjudicates them when they are ambiguous, which they routinely are: conditional events, revised terms, competing offers, jurisdictional divergence over the same instrument. This is the oracle problem restated as company law. The chain's canonical state is authoritative about token movements and derivative about everything that gives the token its economic content.
The intermediated stack is, on this reading, not a legacy accident but an adjudication hierarchy. The CSD's golden record of an event is not valuable because it is a database — databases are free — but because an institution with legal responsibility has interpreted the issuer's language, resolved the ambiguities, and accepted liability for the interpretation. Smart contracts execute rules; corporate actions routinely require someone to decide what the rules are in this instance, and to be legally answerable for the deciding. That function does not decentralise, because its economic content is the concentration of responsibility.
The vocabulary of "servicing" bundles three separable problems, and the rest of the essay takes them in order. Event truth: what legally happened. Entitlement truth: who is owed what, and at which cut-off. Liability: who pays when either answer is wrong. This section has argued the ledger cannot supply the first alone; the next two sections take the second and third.
The second structural reason is stranger, and largely unexamined: much of the entitlement machinery is calibrated to exactly the settlement interval that tokenisation is dedicated to deleting. A record date does several jobs — it fixes a determinable entitlement population and gives a fragmented market one agreed reference point — but the compensation apparatus around it exists for the interval: market claims exist because a trade can be executed cum-entitlement and settle ex; transformations exist because pending instructions can straddle an event; buyer protection exists because an election right can travel with an unsettled trade. That apparatus — the claims layer, not the record date itself — is a patch for the gap between trade date and settlement date. Shorten the gap and the patch shrinks: the move to T+1 mechanically compressed the claims window. Delete the gap — true atomic, continuous settlement — and the claims layer has, in principle, nothing left to compensate: entitlement could collapse to whoever is the authoritative holder at a specified timestamp.
In principle. In practice the deletion buys a harder problem than the one it solves. Entitlement-at-an-instant requires the market to agree on the instant — one authoritative timestamp across every venue, ledger, and jurisdiction where the instrument trades — and fragmentary tokenisation delivers the opposite: the same economic instrument held across a CSD register, one or two permissioned ledgers, and a wrapper or two, each with its own finality clock. A record date is a coordination device: crude, slow, and universally agreed. Continuous settlement replaces one agreed date with a synchronisation problem across systems that do not share a clock — and the market claim, invented for the paper age, quietly returns as the reconciliation item between ledgers. The interval was never the bug. It was the tolerance that let a fragmented market agree on who owns what.
The third structural reason is the one the protocol literature avoids almost completely: withholding tax. A cross-border dividend is paid net of a treaty-dependent withholding, and the difference between the statutory and treaty rate must be recovered — at source, if the paying chain can certify the beneficial owner's status in time, or by reclaim, a paper process measured in months and abandoned by a meaningful share of entitled investors. The EU's own remedy legislates the timescale: the FASTER directive, offering member states standardised relief-at-source and quick-refund systems, transposes by end-2028 and applies from 2030 (Council Directive, OJ 10 January 2025). The contrast is the point: settlement can be harmonised by regulation on a defined timetable, while cross-border withholding relief needs half a decade of institutional build — a legislated admission of where the difficulty lives.
Tax is also where the servicing gap has already been monetised at scale, criminally. The cum-ex complex — dividend-arbitrage structures engineering multiple refund claims on a single withholding — became one of Europe's largest tax scandals by exploiting precisely the ambiguity this essay describes: who, across a chain of trades straddling a record date, held the entitlement to the tax credit; loss estimates vary substantially by scope and method, and the mechanism, not the total, is the point (parliamentary inquiries, DE/EU, 2017–). Cum-ex was, mechanically, an asset-servicing exploit. And the requirement it proves is the one the token standards do not carry: a token can identify an address; the servicing stack must establish the legal person behind it, the beneficial-owner status, and the tax treatment valid for the event. A chain that can pay a dividend cannot yet prove who may receive it gross — and until it can, the gross-to-net gap chains every "native" instrument back into the intermediated stack each time it pays.
Honesty requires the boundary to be drawn from the other side, because there is a genuine success to account for. Digital bonds largely service themselves. The ECB trials, the Swiss central bank's pilot settling digital bonds in wholesale central bank money, the supranational issuers' programmes — coupons calculate, accrue, and pay on-ledger without incident, and have for years, inside an institutional frame of issuance terms, paying agents, and cash infrastructure that did not disappear (SNB Project Helvetia; EIB digital bond programmes). The temptation is to read this as asset servicing solved and scaling. The accurate reading is narrower: a fixed-rate coupon is a corporate action with no choices in it — mandatory, deterministic, pre-programmed at issuance, requiring no election, no golden-source interpretation, and, for the wholesale investor base in question, no retail tax chain. It is the one event class that is genuinely code.
Drawn precisely, then: asset servicing is solved on-chain exactly as far as the instrument contains no optionality. Everything deterministic — coupons, redemptions at maturity, pre-set amortisation — automates cleanly and always did; events requiring interpretation, election, or identity — the equity complex, convertibles, restructurings, anything with a tax personality — remain, generally, on intermediary rails. Which is why native issuance clusters so heavily in plain bonds, and why the instrument class tokenisation most wants, listed equity, is the class whose servicing load it cannot yet carry. The boundary of on-chain asset servicing is the boundary of choice.
There is a second counterexample, and it draws a second boundary. Switzerland has for several years had native shares of unlisted companies issued as ledger-based securities under Art. 973d ff. CO, held directly by their holders with nothing between the ledger and the share register — on one platform the ledger is the register, on another the registrar updates the register from the ledger in a single step — with dividends and votes going to whoever is on that register; in the United States a registered transfer agent keeps a few issuers' native share registers on-chain as the official record (Art. 973d ff. CO; Aktionariat and daura, Switzerland; Securitize as transfer agent, US). Elections work there not because the events contain no choices but because there is no chain for the election to traverse: the holder is on the register, the deadline reaches him in one hop, the issuer's own agent interprets the terms, and the withholding is domestic, recovered through the holder's own tax return rather than through a chain of custodians. Optionality is one boundary; intermediation depth is the other. On-chain servicing works where the event has no choices or the holding has no depth, and fails where both are present — which is a description of listed equity. The moment such a company lists, acquires foreign holders through nominees, or needs a treaty reclaim, the stack reappears, because the moat was never the asset class. It is the intermediated structure around it.
The absorption verdict in Quo Vadis — register and notary functions to the CSDs, economics to the custodians — can now be given its precise mechanism. The CSD anchors the market's book-entry infrastructure — in many legal architectures the authoritative register itself — along with the record date and the golden source of the event; the seat is regulatory: CSDR requires securities traded on venues to exist in book-entry form, which keeps a notary function at the market's core, though it assigns no servicing monopoly (CSDR Art. 3). But the servicing economics sit with the custodians, and the moat is not the database. It is liability. A custodian that misses an election deadline, misapplies an entitlement, or certifies the wrong tax status makes the client whole — that is the product, priced into every basis point of the servicing fee. The custody chain is a chain of parties against whom liability is allocated in a known sequence — who can, bluntly, be sued in the right order — which is what an earlier essay called bindingness as the surviving product: institutions exist where someone must be accountable, and asset servicing is a machine built from accountability.
This is also why the challenger economics disappoint. A protocol can replicate the data flows of servicing at near-zero cost, and several have; none can replicate the balance sheet standing behind an erroneous election, and none is trying to — much of the value proposition of disintermediation is precisely the removal of the party who would absorb the mistake. The incumbents' least glamorous desk turns out to be their deepest moat precisely because its cost structure is dominated by a liability nobody wants to disrupt into their own hands.
The gap between demonstrated settlement and undemonstrated servicing persists because the economics run with it, at every position on the field. The mechanism favours incumbent custodians first: every "native" issuance that must route its dividends, elections, and reclaims back through the stack converts a disruption narrative into a service contract. The platform side gains from the demo economy's definition of live — settlement milestones raise rounds and sign partnerships years before end-to-end servicing could be shown. And issuers gain from not asking: a tokenised instrument marketed on settlement efficiency need never disclose that its lifecycle events run on the infrastructure it claims to bypass.
The holder's diagnostic, for any tokenised or natively issued instrument, is one question with three parts. Ask the platform: who processed your last contested corporate action — a revised offer, a straddling trade's market claim, a reclaim — who bore the liability when it went wrong, and what does the instrument's documentation say happens to an entitlement the code cannot classify. A platform that answers with its settlement statistics has answered.
Settlement is a moment: two movements, one instant, done. Servicing is a relationship — with an issuer whose language must be interpreted, a tax authority whose certainty must be certified, and a mistake whose cost must land on a balance sheet that agreed in advance to receive it. The decade of settlement demonstrations has proven, thoroughly, the half of post-trade that was always the easier half, and the silence around the other half is not a scheduling artefact. It is the shape of the actual moat.
The ledger was never the product. The register decides who holds the right; it was never the product either. The product is the institution that decides what the issuer meant, moves the entitlement through the chain on time, faces the tax authority, and pays for its own errors. Until a protocol wants that liability, the dividend — the ordinary, quarterly, unglamorous dividend — remains the hard part, and the infrastructure that handles it remains, unadvertised, in charge.
Written in a personal capacity. Analytical views only — not legal or investment advice.
Julian Gretzinger — Investor and writer on monetary history, real wealth mechanics, and financial markets. substack.com/@juliangretzinger