Canton Intelligence Framework Proposal
Development Fund Proposal Submission
Proposal file: Canton Intelligence Framework
Abstract
Canton is becoming the coordination network for institution-grade financial workflows, bringing together issuers, trading venues, custodians, validators, infrastructure providers, service providers, and application developers under shared privacy guarantees. That architecture is one of Canton’s strongest advantages: it allows regulated participants to coordinate without exposing sensitive data beyond what each workflow requires.
As Canton grows, that same strength creates a new network-level challenge. Market integrity, compliance intelligence, valuation, underwriting, and risk monitoring increasingly depend on signals that no single participant can observe in full. A trading venue may see order-book behavior. A custodian may see wallet movement. A validator or infrastructure provider may see operational anomalies. A settlement participant may see timing patterns. A compliance provider may see links to entities already under review. Individually, each signal may remain below the threshold for action. Together, the pattern may reveal wash trading, circular flows, false liquidity, suspicious settlement behavior, or emerging systemic risk.
This is not a problem that can be solved by centralizing data. The relevant information is fragmented by design. Customer activity, trading records, wallet data, custody activity, settlement information, internal review outcomes, and proprietary risk labels are constrained by regulation, contract, privacy obligations, and competitive sensitivity. Institutions cannot simply pool that information into a shared database. Canton should not ask them to compromise the privacy and control that brought them to the network in the first place.
The Canton Intelligence Framework introduces a privacy-preserving, open-source coordination layer for collaborative intelligence on Canton. Participants train shareable models from their own local data, while raw data remains inside their own environments. Canton and Daml provide the governance and coordination layer: participant permissions, training rounds, model commitments, evaluation records, contribution attribution, audit trails, and incentive logic. Training remains off-ledger; Canton records the governed process, provenance, and commitments that make collaboration auditable without centralizing sensitive information.
The first use case is market integrity because it is immediate, measurable, and directly tied to institutional confidence in Canton-based markets. The Phase 1 pilot will focus on developing models for market integrity across key Canton stakeholders. The objective is practical and testable: demonstrate that a participant’s local risk model can improve through governed collaborative training, without receiving another participant’s private records.
The Canton Intelligence Framework is best understood as a new ecosystem primitive for coordinated intelligence. It gives any governed group of Canton participants the ability to form a training consortium, define participation rules, select or operate aggregators, record model commitments, measure contributions, and improve shared models without centralizing sensitive data. The first implementation is market integrity, but the same framework can support underwriting, collateral risk, compliance, valuation, litigation finance, and continuous asset monitoring. T-RIZE’s role is to deliver the initial Canton-native architecture and reference implementation; the value of the framework is that it can be reused, extended, and governed by the ecosystem itself.
Demand is being anchored through a concrete first-adopter path. MPCH, a cybersecurity and infrastructure company operating close to governed signing, validator, recovery, key-management, and regulated blockchain environments, has expressed significant interest in the framework and is being advanced as a Phase 1 design and implementation partner. Its relevance is strategic: MPCH sits near the infrastructure layer where market integrity, key management, validator operations, governed access, and compliance workflows intersect. T-RIZE is also a direct user of the capability through its Canton-based institutional asset programs, including litigation-finance and private-credit workflows where collaborative valuation, underwriting, risk scoring, and continuous monitoring can improve capital allocation.
The research base is already in place. T-RIZE has spent roughly three years and approximately $2 million of its own capital developing the underlying privacy-preserving machine learning work, supported by more than $3 million of Canadian federal research funding through Mitacs and NSERC Alliance. The work is led through the T-RIZE Industrial Research Chair at École de Technologie Supérieure under Prof. Kaiwen Zhang, with a specialized research team and prior work across decentralized federated learning, secure aggregation, differential privacy, federated evaluation, contribution attribution, and incentive mechanisms. Phase 1 is therefore focused on translating existing research and prototype work into a Canton-native open-source framework, rather than beginning from basic research. Canton is not being asked to fund the full cost of inventing this capability: the requested grant funds the final-mile Canton integration, open-source release, and market-integrity pilot validation of a research base that already exists.
The ask is intentionally staged. Phase 1, covering milestones 1 and 2, requests 6,500,000 CC total to deliver a bounded validation package: Canton-native architecture, Daml governance primitives, open-source reference implementation, developer documentation, participant onboarding materials, and a working market-integrity pilot. Success will be evaluated against pre-registered metrics before the pilot begins, including improvement over isolated-model baselines, precision and recall performance, reduction in undetected suspicious activity, auditability of training rounds, and confirmation that raw participant data remains local. Milestone 3 proceeds only after the Committee can evaluate real results and agrees to continue the development.
The long-term sustainability model is built into the framework. Aggregators, trainers, evaluators, model users, and governance participants can be recognized through configurable contribution and usage mechanisms. A consortium can operate its own aggregator. Infrastructure providers can support enterprise-grade deployment. Participants can contribute data, compute, evaluation, or governance work without exposing their datasets. Models can support usage-based economics or customer inference revenue sharing where governance permits. This creates a path for maintenance, adoption, and ecosystem growth beyond the grant period.
Canton’s privacy is what makes it credible for regulated institutional workflows. The Canton Intelligence Framework makes that privacy scalable. It gives participants a practical way to improve fraud detection, market surveillance, compliance, underwriting, valuation, and risk scoring while preserving data control. It turns fragmented private signals into governed shared intelligence and expands Canton from a network that coordinates assets into a network that can also coordinate trusted intelligence across institutions.
Checklist
- [ x ] Proposal file added under
/proposals/ - [ x ] Milestones and funding amounts defined
- [ x ] Acceptance criteria included
- [ x ] Alignment with Canton priorities described