We believe legal AI has reached an inflection point. The first generation asked how AI could make lawyers more productive. The second generation asks how AI can redesign the institutions within which lawyers operate. Our work belongs to the second generation.
These are experiments in that direction — not claims to have solved it. The argument that follows is offered in the spirit in which it was developed: as a rigorous attempt to understand present design failures by tracing them to their origins.
This essay is the theoretical foundation of the Procedural Gap Project — five working papers on structural failures in legal procedure, each with a corresponding diagnostic tool on kyc.co. The monetary design argument and the legal procedure argument are the same argument at different registers.
The projects described on this site — KYC.co and Moral.Money — are not presented as solutions. They are design experiments engaging from first principles with a set of problems that the current wave of monetary innovation — blockchain, cryptocurrency, and CBDC — has not resolved and, in most cases, has made structurally worse.
The argument here is not about individuals. Politicians, central bankers, blockchain founders, and regulators are operating within inherited frameworks. The failure is architectural. You cannot blame a builder for a load-bearing wall that was designed incorrectly two hundred years before they arrived on site. What you can do is identify the error, trace it to its source, and ask whether first-principles design would have made the same choices.
This is that exercise. It is offered as a contribution to academic and design debate — not as financial advice, not as investment proposition, and not as a claim to superior insight over any individual or institution.
Most legal AI asks how to make legal work more efficient. The question is reasonable. The tools are real and the efficiency gains are measurable. A contract review engine that runs in seconds is genuinely useful to the party who would otherwise have waited weeks.
Most legal AI asks how to make legal work more efficient. We ask why the work exists in the first place, and whether a better decision architecture could make much of it unnecessary.
These are not the same question. The first assumes the system is fundamentally correct and asks how to reduce the cost of operating within it. The second asks whether the system is optimising for the right things — whether the rules it enforces are producing the outcomes they claim to serve, and whether a different architecture would generate those outcomes at a fraction of the current cost.
The historical argument that follows is an attempt to identify what a better optimisation target looks like. The evidence predates every centralised legal and monetary system currently in operation. What happened between then and now is the story of how a design that embedded dispute resolution into the instrument itself was replaced by a design that delegated it — and then never quite built the delegation layer correctly.
A legal system is not merely a collection of rules. It is an optimisation system — a structure that, by design or by drift, rewards certain behaviours more than others and thereby produces, at scale, exactly what it has been designed to incentivise.
Every optimisation system produces what it rewards. If the dominant incentive rewards procedural activity, the system produces procedure. If it rewards adversarial persistence, the system produces conflict. If it rewards billable hours, the system produces more legal work. None of these outcomes requires any participant to act in bad faith. The participants are behaving rationally within the structure they inhabit. The output is the structure, expressed at scale.
This is the systems engineering observation underlying the five procedural gap papers. Each identifies a different point at which the legal system's actual optimisation target — what the rules materially reward — has diverged from its stated objective. The competence gap exists because the credential system rewards qualification over proximity to the specific problem. The proximity gap exists because the independence framework rewards self-declaration over measurable network distance. The disclosure gap exists because the silence incentive rewards the party holding private information over the party who needs it. The escalation gap exists because the billing architecture rewards process duration over resolution. The self-reporting fallacy exists because the adversarial framework rewards argument over truth.
Changing the software without changing the optimisation target does not change the system. What this project attempts is to identify the optimisation targets that produce these outcomes, make them measurable, and design instruments that shift them — by surfacing what the existing system has no mechanism for surfacing and making the cost of misaligned incentives visible before it becomes irreversible. This is not a philosophical preference for a different kind of law. It is a systems engineering argument about what happens when the reward structure of a decision architecture diverges from its stated purpose — and what it takes to correct that divergence.
Before money as we currently understand it, reputation was the currency. In medieval England, the dominant mechanism for recording debt and credit was the tally stick: a length of wood on which notches were cut to record a transaction, then split lengthwise so both parties held a matching half. The two halves were the proof. If they matched, the debt was real. If they did not, someone was lying. No institution required. No intermediary. The record was peer-to-peer, bilateral, and self-verifying.
The English Exchequer used this system for over seven centuries — from the Norman period through to 1826, when the practice was formally abolished. The surplus tally sticks accumulated by the Exchequer were incinerated in a furnace beneath the House of Lords in October 1834. The fire spread through the walls. Both Houses of Parliament burned to the ground. The physical record of England's pre-centralisation monetary system was destroyed by the institution that had replaced it.
What the tally stick system was doing — beyond recording debt — was embedding dispute resolution into the monetary instrument itself. The physical object was the proof. The match between the two halves was the verification. There was no need for a third party to adjudicate because the instrument carried its own evidentiary standard.
When reputation was the currency, dispute resolution was not a separate layer. It was built into the instrument. The transition to centralised monetary systems did not just change who issued money — it removed the self-verifying quality of the original instrument and handed verification to an institution.
Parallel to the tally stick tradition in England, the great trading networks of medieval Europe developed their own reputation-backed monetary instruments. The Champagne fairs of the 12th and 13th centuries were the clearing house of European commerce — twice-yearly events in the Champagne region of France where merchants from Venice, Genoa, Bruges, and London settled months of accumulated credit. The instrument that made this possible was the bill of exchange, developed in detail by Venetian and Florentine merchant banks: a written order by one party to another to pay a specified sum, backed not by gold but by the creditworthiness — the reputation — of the issuing merchant.
What is critical about the bill of exchange is not its mechanics but its epistemology. The question it answered was: can this person be trusted to make good on this commitment? The answer was not provided by a sovereign, a regulator, or an institution. It was provided by the accumulated record of the merchant's past behaviour in the network. Reputation was verifiable. Reputation was tradeable. Reputation was, functionally, money.
The bill of exchange also had built-in dispute resolution — not as an add-on layer, but as a structural feature. When a bill was dishonoured, the mechanisms for resolution were embedded in the network of merchants who issued and guaranteed the instruments. The same parties who created the credit were the parties who adjudicated its failure.
The arc from 1694 to 1931 is the story of reputation being replaced by authority. The question can this person be trusted? was replaced by the question does this institution say this is valid? The bilateral self-verifying instrument was replaced by a third-party-issued token. And critically: the dispute resolution mechanism that had been embedded in the original instruments — the match between the two tally sticks, the network of merchants who could adjudicate a dishonoured bill — was not carried forward. It was silently removed. The new system had issuance. It had regulation. It had legal tender status. It did not have dispute resolution at the transaction layer.
The cryptocurrency movement that emerged from the 2008 financial crisis was, in part, a response to exactly the problem described above: centralised monetary issuance controlled by institutions that are not accountable to the participants whose transactions they intermediate. The diagnosis was correct. The design response — while technically innovative — reproduced the same structural omission it set out to fix.
Proof-of-work — Bitcoin's consensus mechanism — rewards those who can afford the most mining hardware. The entity that controls the most computational power earns the most newly minted currency. This is not a neutral mechanism. It concentrates new money issuance in the hands of those who arrived at the table with the most capital to spend on hardware. The distributional logic is identical to the Bank of England's founding: those closest to the source of new money accumulate advantages that compound over time.
Proof-of-stake — the architecture adopted by Ethereum and most subsequent networks — rewards those who already hold the most tokens. Staking returns are proportional to stake size. This means the largest holders earn the most from new issuance, and the gap between large and small holders widens structurally over time. The mechanism differs from Proof-of-Work; the outcome is the same.
Neither architecture provides any mechanism for dispute resolution at the transaction layer. A smart contract that executes incorrectly, a delivery that does not occur, a service that is not rendered as specified — none of these have a resolution path built into the protocol. What blockchain built was a ledger that cannot be falsified. What it did not build was a system for adjudicating whether the real-world event underlying a transaction actually occurred.
Bitcoin fixed the counterfeiting problem. It did not fix the dispute problem. The tally stick had both. The bill of exchange had both. Every monetary architecture since 1694 has had neither.
Central Bank Digital Currency architecture, as currently specified by the Bank for International Settlements and implemented downstream by national monetary authorities, closes the compliance distance between the issuing institution and every transaction endpoint. Programmable money allows granular enforcement, conditional spending, and direct tax collection at the transaction layer. It is, from a design perspective, the logical completion of the trajectory begun in 1694: the full digitisation of institutional monetary authority with no mediating layer between the issuer and the participant.
What the current BIS architecture does not include — and does not appear to be designed to include — is any dispute resolution mechanism at that same layer. The compliance infrastructure is being built at scale. The adjudication infrastructure is not. As the velocity and volume of digital transactions grows under CBDC rails, the gap between the enforcement capacity of the system and its dispute resolution capacity will widen exponentially. The traditional court system, operating within a procedural framework largely unchanged since the nineteenth century, has no structural ability to absorb that load.
This is not a conspiracy. It is a design failure with a 330-year lineage. Each generation of monetary innovators has inherited a system that lacked dispute resolution and has built the next layer without asking whether the omission should be corrected. The tally stick had the answer. The bill of exchange had the answer. The question was abandoned when reputation was displaced by authority.
When the BIS and national monetary authorities began designing CBDC, they confronted a technical problem that decentralised networks had already partly solved: how to move value at scale, with finality, without the friction of legacy correspondent banking. Their design response was to adopt the essential architecture that blockchain had developed — digital wallets, programmable transaction logic, distributed ledger concepts — and rebuild it as a centralised system under institutional authority. The CBDC and the blockchain are, in this architectural sense, close relatives. One distributes control. The other concentrates it. But both are built on the same underlying settlement pattern.
What neither system inherited from the other — because neither had it — was dispute resolution at the transaction layer. Blockchain built a ledger that cannot be falsified. CBDC builds a ledger that the issuing authority controls completely. Both chose their settlement architecture from a design space that did not include the question of what happens when the real-world event underlying a transaction is disputed. A smart contract executes whether the delivery occurred or not. CBDC programmability enforces compliance rules whether the underlying obligation is legitimate or not. Neither system has a mechanism for the question that the tally stick and the bill of exchange had already answered: what do two parties do when the instrument and the reality it was meant to record no longer match?
The macroeconomic cost of this shared omission is not measured by either system. Unresolved disputes freeze capital — parties cannot redeploy resources tied up in litigation. Disputed obligations reduce the supply of voluntary exchange: counterparties who have been through a dispute become risk-averse, contracts shrink in scope, credit becomes more expensive. The aggregate friction of unresolved disputes across an economy is a real quantity. It appears in productivity statistics as deadweight loss, in credit markets as elevated risk premiums, in investment patterns as systemic underinvestment in relationships that would otherwise have been productive. Neither the blockchain protocol nor the CBDC specification has any mechanism for measuring this quantity, let alone reducing it. The settlement layer is specified with precision. The dispute layer does not exist. And the economy absorbs the difference.
The design philosophy of the tally stick and the bill of exchange was, in part, a macroeconomic architecture. Embedded dispute resolution meant that disputes could be resolved at transaction cost rather than at litigation cost — a difference of several orders of magnitude. When that layer was removed by the transition to centralised authority, the cost did not disappear. It was transferred: to the parties, to the courts, to the insurers, to the broader economy that must absorb the friction of a system designed without it. The blockchain and the CBDC each inherit that transfer. That neither recognised the omission when they had the opportunity to correct it is not a failure of individual judgment. It is what happens when a settlement architecture is designed without asking what the instrument needs to do when settlement fails.
The path from blockchain to CBDC was not direct. The intermediate architecture was the stablecoin: a token pegged to an existing currency and backed by fiat reserves, which added price stability to the digital wallet and programmable logic that decentralised networks had developed. Stablecoins also added one capability that institutional actors found particularly significant — the ability to freeze a wallet or transaction at the issuer's discretion. This made a stablecoin, in functional terms, a private digital currency with compliance controls that the issuer could enforce unilaterally. What CBDC did was take this architecture and complete its centralisation: moving issuance from a private company to a sovereign central bank, replacing opt-in compliance with mandatory regulatory programmability, and backing the instrument with sovereign authority rather than commercial reserves. The central bank digital currency is, architecturally, a stablecoin with institutional control. The innovation was significant on its own terms. It was not an answer to the question the tally stick had already asked.
The shared omission across the full chain is precise: every instrument records the transfer of a token without verifying whether the real-world obligation that token was meant to represent was actually completed. A blockchain transaction executes when the cryptographic condition is met. A stablecoin freezes or releases on the issuer's instruction. A CBDC enforces compliance rules at the transaction layer. None of them distinguishes between a token that represents a completed obligation and one that represents a promise of performance that was never delivered. The instrument records the transfer. It has no mechanism for the prior question: was the performance delivered? was the obligation honoured? This is the question the tally stick answered by making the match between the two halves the verification standard. The bill of exchange answered it by making the merchant's reputation the counterparty to every commitment. Every architecture since has answered it by not answering it — by delegating the question to a dispute resolution system that was never built into the instrument and was never adequately built anywhere else.
The CBLT token is an attempt to design the next element in this chain. It is structured as a security token — a legally compliant instrument in which holding the token means assuming an obligation, not acquiring an asset. The token cannot be realised until the underlying obligation is verified as complete: it moves through a defined state architecture — from allocation to escrow to eligibility to redemption — and economic benefit is recognised only at the final stage, only after both contracting parties have approved completion and an independent expert has not blocked release. Once escrowed in a contract, the token cannot be transferred. This is not primarily a security feature, though it functions as one. It is an architectural consequence of the design premise: an escrowed CBLT is an obligation unit, not a tradeable instrument, and the holder's interest is in completing the work, not in moving the token. The dispute resolution mechanism is not downstream of the contract. It is embedded in its structure. The full specification is in the KYC.co Blackpaper ↗.
There is a structural reason why every architecture since the tally stick has left this layer out. Civil law — the legal framework governing commercial disputes — does not, in practice, produce the enforcement consequences necessary to change the behaviour of parties who are strategically motivated to delay or deny performance. Criminal procedure produces different behaviour because its consequences are real: there is genuine deterrence, genuine risk, genuine change in the calculus of the parties involved. Civil procedure operates with nominal sanctions — penalties that are rarely imposed, rarely proportionate, and easily absorbed by a well-resourced party with a strong incentive to extend proceedings. A dispute that would resolve rapidly under genuine criminal threat can persist for years in civil procedure, because the cost of bad-faith conduct is small, intermittently applied, and structurally subsidised by the billing architecture that rewards continuation over resolution. The CBLT does not attempt to reform civil enforcement. What it attempts is to make enforcement structurally unnecessary for the common case: the party who does not honour the milestone does not receive the allocated tokens. The expert who cannot verify completion blocks the release. The consequence is built into the instrument's architecture and does not require a court to impose it.
The relationship between civil procedure's enforcement gap, the plausible deniability that gap creates for actors who choose to exploit it, and the design of instruments that alter the underlying incentive structure is the subject of ongoing work. Each of the five papers of the Procedural Gap Project identifies a specific structural failure that can be traced to this more fundamental problem: a legal architecture whose incentive structure rewards the appearance of performance more reliably than it rewards actual performance. The paper that follows this sequence addresses the problem directly. What is described here — the CBLT's design, the milestone escrow structure, the expert adjudication layer — is its architectural response.
Legal disputes are commonly described as conflicts between parties. At the surface level this is accurate. At the level of system design, most disputes are better understood as information failures: moments when the quality of information available to decision-makers has been degraded by incentives, ambiguity, competence mismatch, adversarial framing, or the simple absence of any mechanism for surfacing what is actually true.
Each of the five procedural gaps documented in the academic papers is, at its root, an information quality failure. The expert witness who is not independently competent for the specific matter at hand is an information quality problem. The proximity network connecting that expert to a party is an information quality problem. The disclosure asymmetry that makes silence rational — and transparency costly — is an information quality problem. The compounding cost trajectory invisible at filing but visible in the structural pattern of similar files is an information quality problem. The professional's filed evidence that contradicts their own stated position is a specific kind: a self-referential distortion in the record that the record-generating institution has no structural incentive to surface.
The academic work on the self-reporting fallacy makes this precise in one domain. Legal procedure routinely produces statements made without regard for their truth — not lies in the strict sense, but assertions whose epistemic quality is secondary to their strategic function. False certainty, mischaracterised evidence, institutional hedging, and adversarial rhetoric each degrade information quality in ways that compound across a proceeding. The problem is not primarily ethical. In most civil proceedings, a party whose evidence is contaminated by strategic misrepresentation is not in legal jeopardy for the misrepresentation — the civil sanction for producing low-quality information is negligible. The problem is architectural: as information quality deteriorates, every decision made on the basis of that information deteriorates with it. The system progressively loses its ability to distinguish between what happened and what was successfully argued. Bullshit, in this framing, is not primarily a moral failure. It is an information architecture failure — and it compounds.
From this perspective, the role of AI in legal systems is not to automate document work. It is to restore the information quality that the adversarial system has structurally degraded. This distinction matters: a faster process running on distorted information produces distorted outcomes faster. The instrument needs to improve the information before it can improve the outcome.
The failure of every monetary instrument since the tally stick to include a dispute resolution layer is not, on its own, enough to explain the legal system's current state. Something more specific happened: the resolution layer that was eventually built was priced on the wrong variable.
The driller is paid per metre drilled. The lawyer is paid per hour billed. The expert is paid per report filed. The court clerk is paid to process the volume in front of them. None of these actors is paid for resolution. Resolution is the one outcome in the system that ends the billing. Every other outcome — the motion, the adjournment, the further expert instruction, the amended pleading, the additional hearing — generates more of it. A structure metered on process and not on outcome will, without any individual intending it, over enough time, grow better at process and worse at resolution. The distance between the moment a dispute is filed and the moment it closes is not a failure of the system. Measured by what the system rewards, it is its primary product.
This is what the doctor on her eighteenth hour of a shift and the patient with a cut on their leg do not share: a perspective. From the outside, a visit to hospital is an exceptional event, and the expectation is full presence and clear judgment. From the inside, it is the forty-third case of the shift, processed by someone whose margin of attention was rationed out six hours ago. The gap between those two perspectives is not a moral failure on either side. It is what a system optimised for throughput looks like from inside versus from outside — and legal procedure, optimised for billable throughput across its entire professional layer, has produced the same gap at industrial scale. The parties before the court believe the matter is being read whole, with full attention, by someone with nothing at stake in the answer. The practitioners inside the system know what the docket looks like. Neither is wrong. They are looking at the same structure from opposite ends of the meter.
The conclusion is uncomfortable but arithmetically precise: a system that charges for process has no financial interest in resolution, and a system with no financial interest in resolution will, at scale, produce more process than resolution, more dispute than settlement, and a cost of litigation that approaches or exceeds the value of the matter in dispute — not as a design failure, but as the design working correctly. The five papers of the Procedural Gap Project each named one presumption the system grants for free; underneath all five presumptions is this single pricing mechanism, which is why the presumptions have survived scrutiny for three centuries. They were not simply overlooked. They were load-bearing. The revenue passed through them.
This is not an accusation directed at any individual. The lawyer billing by the hour is not a bad actor — they are operating rationally within a structure they inherited and did not design. The judge managing a docket of eight hundred active matters is not negligent — they are processing what the structure placed in front of them. The observation is structural: a billing architecture that meters effort rather than outcomes will produce institutions that are excellent at effort and indifferent to outcomes, and what has been built, measured by what it rewards, now functions at least as well as an apparatus for sustaining disputes as one for resolving them.
Respect is commonly treated as an ethical aspiration — a quality that well-designed institutions ought to maintain. This framing, while not wrong, renders respect invisible to the instruments that measure economic performance and absent from the architectures that legal systems are built on. This project treats respect differently: as a measurable economic variable with quantifiable effects on system performance.
High respect between parties reduces friction at every transaction that passes through it. High respect increases voluntary compliance with decisions — reducing the enforcement overhead required to make outcomes stick. High respect improves information quality: parties who trust the process disclose more accurately, cooperate more fully, and accept findings they did not prefer. High respect reduces litigation: many disputes that escalate to adversarial proceedings would, in a higher-trust environment, resolve at negotiation cost. The working hypothesis is that legal systems become exponentially more expensive as respect is depleted — that the cost of operating a low-trust legal architecture grows non-linearly as trust falls, because each unit of lost trust requires a disproportionate increase in procedural enforcement to substitute for it.
When respect is destroyed — by expert testimony that is structurally compromised, by disclosure asymmetries that leave one party holding false information, by procedural manipulation that the other side cannot name but can measure in the cost of continuing — the system must replace it with enforcement: more process, more authority, more procedure, more cost. The substitution is not measured anywhere. The legal system produces no reading of the trust content of a proceeding and has no mechanism for detecting the moment at which trust became irretrievable. A proceeding in which trust is intact resolves. A proceeding in which trust has been destroyed proceeds for as long as the least cooperative party has an interest in its continuation — which, given the billing architecture already described, is often indefinitely.
The instruments on this platform are an attempt to produce a partial trust accounting. A structural score that identifies competence mismatch, proximity conflict, disclosure asymmetry, or self-referential contradiction in a proceeding is, among other things, a proxy for the amount of respect that proceeding is consuming — and a signal that the window in which resolution remains achievable at reasonable cost is still open.
Every unnecessary dispute increases legal entropy. Every inaccurate pleading increases it. Every competence mismatch assigned to a technical matter increases it. Every avoidable appeal, every procedural motion filed not because it is necessary but because the billing architecture rewards filing it, every institutional delay that compounds into a further institutional delay — all of these increase legal entropy.
The concept is borrowed from thermodynamics deliberately. Entropy in a physical system measures the degree to which the system's energy has become unavailable for useful work. Legal entropy, in the same sense, measures the degree to which the legal system's capacity for resolution has been consumed by process — by activity that does not advance the question of what actually happened and what should follow from it, but that must be paid for nonetheless, and that forecloses other uses of the same resources.
A legal system optimised for activity will produce activity. Without an instrument for measuring the ratio of useful work to total work, the system has no mechanism for detecting when that ratio has fallen below the threshold at which it can be said to be functioning as a resolution architecture at all. The adversarial system, measured by its own metrics, is functioning well. It is processing the volume in front of it. What it cannot ask — and cannot answer, given the instruments it currently possesses — is whether that processing is producing resolution, or merely more process.
The purpose of the decision architecture described on this platform is not only to predict entropy. It is to identify the structural signals that indicate entropy is accumulating in a specific proceeding — before it accumulates to the point where no resolution other than exhaustion remains. The settlement window visible in the structural pattern of similar proceedings. The competence mismatch detectable at assignment. The liability migration appearing in the record long before it appears in any claim. These are entropy signals. Making them legible — before the decisions that would have changed the outcome have already been made — is the operational purpose of the instruments.
If you were designing a monetary system from first principles — starting not from "how do we improve on the existing system" but from "what does a monetary instrument actually need to do" — you would arrive at something that looks more like the tally stick than the banknote.
A monetary instrument needs to record a commitment between two parties. It needs to provide a verification mechanism for that commitment. And it needs to provide a resolution mechanism for the case where the commitment is not honoured. The tally stick did all three. The bill of exchange did all three. Neither the pound sterling, nor Bitcoin, nor the proposed CBDC architecture does all three.
KYC.co is an attempt to build the missing layer into the contract itself: dispute resolution embedded at the contract layer, expert adduction built into the instrument, backed by CBLT — a token whose issuance is tied to verified productive contribution rather than capital ownership or hardware possession. It is an experiment. It may be wrong in important ways. It is offered as a contribution to a design debate that the dominant monetary architectures of 2025 — both blockchain and CBDC — have not seriously engaged with.
Moral.Money is an attempt to make this argument legible to a non-specialist audience — to translate 700 years of monetary design failure into a form that can be engaged with by people who will eventually live inside the consequences of the choices currently being made, whether or not they can parse a BIS working paper.
The five academic papers of the Procedural Gap Project are the same argument applied to specific and documented failures of the current system. It is not a coincidence that the expert witness case study involves the same structural problem — a principal–agent relationship with no verifiable impartiality standard and no quantitative mechanism for adjudicating proximity — that the broader monetary argument identifies as the missing layer in every instrument since the tally stick.
These are experiments in design philosophy. The historical argument is presented as a framework for understanding present failures, not as a complete theory of money. The proposed architectures are early-stage implementations of design principles that deserve scrutiny, critique, and academic engagement. Correspondence from researchers, comparative procedure scholars, monetary historians, and economists is welcomed at hello@kyc.co.
The four diagnostic instruments on this platform are components of a single framework, each addressing a different structural failure in the same information system. Taken together, the readings they produce describe the information quality of a proceeding at a given moment. But the framework is not only diagnostic. It is predictive.
A file with high competence distance, filed in a jurisdiction with measurable disclosure asymmetry, managed by professionals with network proximity to the counter-party, will not resolve efficiently. This is not a probabilistic estimate. It is a structural read: the incentives present in this configuration point toward escalation. The settlement window exists now. The liability migration — the movement of responsibility from original defendant to the filing professionals themselves — is detectable in the record before it appears in any claim. The question is not only what is wrong with this proceeding. It is what this proceeding is becoming, and who bears responsibility for what it becomes.
Current legal AI largely automates what happens inside the adversarial process: drafting, reviewing, searching, summarising. This is useful work. But it assumes the process is correctly designed and asks how to run it more efficiently. This architecture attempts something different: to optimise the decision — the assignment of responsibility, the identification of the settlement window, the structural reading that makes the outcome visible before it is fixed — rather than the document the decision eventually generates.
Legal disputes are the first application. The underlying framework — measuring information quality in institutional decisions, identifying trust deficits before they become entrenched, predicting decision trajectories from structural signals rather than case-by-case intuition — is not specific to legal proceedings. The same logic applies wherever institutional decisions accumulate over time, wherever informational asymmetry between parties goes undetected, and wherever the cost of a bad decision is borne long after the moment at which a different choice was possible.
There is an emerging generation of legal AI that is asking a different question — not how to make legal work faster, but how to make legal obligations machine-readable and institutional decisions computable. Our work belongs to that generation. The instruments are legal in their first application. The decision architecture is not: the same framework — measuring information quality, identifying structural misalignment, predicting institutional trajectories — applies wherever actors accumulate decisions over time and wherever the cost of a bad decision is borne long after the moment at which a different choice was available.
moral.money · Procedural Gap Project · Five Papers · Steven Carroll · kyc.co ↗
Design experiments · Not financial advice · Correspondence: hello@kyc.co