An Open Letter on Ontario's Machine-Readable Regulatory Framework

Comment on the OSC's proposal to publish a machine-readable dataset of Ontario's securities regulation

This submission is made by Material, an AI-native services firm for public companies whose work involves rendering securities regulation into a form that software can reason over reliably.

We welcome the Ontario Securities Commission (the "OSC")'s initiative. Our purpose here is to address questions on which the OSC has invited comments at the level where, in our experience, initiatives of this kind succeed or fail, and to set the OSC's choices against the experience of comparable jurisdictions that have already travelled this road.

A single principle organizes our comments. Rendering regulation machine-readable is not a neutral act of transcription. Decisions about structure, hierarchy, exceptions, and cross-references embed interpretive choices, and once a dataset is relied upon at scale, those choices begin to shape how market participants understand their obligations before any human legal analysis intervenes. The attendant risk is that interpretive influence migrates, however unintentionally, from the regulator and the courts toward the designers of datasets, compliance software, and AI systems, and that machine-generated readings come to be treated as authoritative notwithstanding their errors or limitations^[1].

We raise this not as an objection to the initiative, which we support, but because it clarifies what good design must accomplish. The OSC's foundational choice, to annotate the authoritative text rather than to rewrite the law as code, is itself the first and most important safeguard against this risk, for it keeps interpretation anchored to the regulator's own words. The three matters we address below, the licensing and distribution of the dataset, the expression of relationships within it, and its eventual pairing with machine-readable filings, are each, in their way, a means of ensuring that interpretive authority remains where it belongs.

1. Terms of Access Will Determine Whether an Ecosystem Forms

The OSC has asked for feedback on licensing preferences, and we would encourage the Commission to treat this as among the most consequential decisions in the entire initiative rather than as an administrative matter to be settled late.

The overview's own stated aims, to mitigate barriers to entry for service providers and to support both new and existing RegTech solutions, are achievable only under a particular class of licensing and distribution choices. We would urge three commitments. First, that the dataset be free and available in bulk, rather than accessible only through a hosted interface. Second, that it be governed by an open, published schema, so that third parties can both build upon it and independently validate it against the authoritative source. Third, that the schema and its controlled vocabularies remain neutral, and not ossify around the patterns of any single vendor's taxonomy.

The second and third of these commitments bear directly on the question of interpretive authority. An open schema that can be checked against the authoritative text is the structural mechanism by which the regulator's interpretation, rather than a vendor's silent encoding of it, remains the operative one. Where the dataset is open and validatable, any divergence between the encoded structure and the law itself can be detected and corrected in the open. Where it is closed, such divergences accumulate unexamined, and the dataset's designer acquires, by default rather than by mandate, a measure of interpretive influence that properly belongs to the OSC. The OSC's preliminary work has been valuable, but the durability of an annotation-based approach depends on the labels expressing the structure of the law itself rather than the conventions of any one supplier. We note the OSC's intention to hold future consultations on taxonomy governance, and we regard a governance model that preserves this neutrality, and that provides for transparent version control and error correction, as essential to the project's long-term credibility.

The governing precedent is XBRL. When the United States Securities and Exchange Commission (the "SEC") adopted its final rule mandating structured financial data in January 2009^[2], it published an open specification and made the resulting corpus freely available in bulk. The instructive evidence is the market's response. An entire layer of providers, both commercial and open-source, now builds directly upon that foundation, ranging from hosted services that deliver filings at scale to freely available libraries that render any filing into structured, analyzable data without keys or fees^[3]. The regulator did not need to anticipate these use cases; the openness of the standard and the accessibility of the data were sufficient to call them forth. Access confined to a portal, or encumbered by restrictive licensing, would have capped that activity at whatever the regulator's own interface supported.

The XBRL experience is not isolated. The same dynamic recurs wherever a regulator or government has published authoritative legal text as open, documented, bulk-accessible data. In the United Kingdom, The National Archives publishes the entire statute book at legislation.gov.uk under the Open Government Licence, with a RESTful API that was released in July 2010, ahead of the public website itself, and that returns content as HTML, XML, and RDF^[4]. In the United States, the Office of the Law Revision Counsel publishes the U.S. Code in bulk under the United States Legislative Markup (USLM) schema, an open, semantically-oriented XML format maintained publicly and interoperable with the international standard^[5], and the Federal Register and Code of Federal Regulations are likewise distributed in structured XML through documented APIs that require no key. The European Union's EUR-Lex publishes legislation as linked open data through its Cellar repository, addressable by the European Legislation Identifier^[6]. In each instance the openness of the standard and the availability of bulk data, rather than any effort by the regulator to build applications itself, is what summoned the surrounding ecosystem into being. The lesson transfers to Ontario with little modification: an open schema with bulk access allows the market to supply the tooling and allows that tooling to be checked against the authoritative text. This is precisely the dynamic by which barriers to entry fall and a RegTech ecosystem of the kind the overview envisions comes into being.

The converse is equally well evidenced. The United Kingdom's own history is the cautionary case: the predecessor Statute Law Database was originally conceived on a model that contemplated charging for access to historical law and that did not engage the wider community of legal publishers, and it was only after that approach was abandoned in favour of free, openly licensed bulk data and an API that the third-party application and publishing ecosystem now built on legislation.gov.uk emerged. Closed or fee-gated distribution does not merely slow adoption; it forecloses the very validation against the authoritative source on which the integrity of an annotation-based dataset depends.

Two technical choices materially affect whether the dataset is usable in practice, and both bear on the OSC's question about priority content. The first is format. Maintained, annotated text in a structured, plain-text representation is considerably more reliable for software, and for language models in particular, to parse than formats in which visual layout obscures logical structure. This is why the leading jurisdictions have all moved from presentation-oriented formats toward semantic XML: the USLM schema, for example, was deliberately designed as a second-generation markup that expresses structure and meaning rather than the visual characteristics that its predecessor captured.^[7] Where the OSC is already applying expert annotations, preserving them in structured text compounds that effort rather than subordinating it to presentation. The second is versioning, which we address below in connection with linkage, because in this corpus the two cannot be separated.

2. Relationships Are How the Regulator Retains Interpretive Authority

When the relationships among provisions are left for software providers to infer, it is they who supply the interpretation, and their inferences that market participants come to rely upon. When the OSC instead expresses those relationships itself, with its own authority, the interpretation carries the regulator's imprimatur rather than a vendor's conjecture. Typed linkage is thus not merely a matter of dataset quality; it is the means by which interpretive authority over the structure of the rules is retained by the OSC rather than ceded by omission.

In a corpus as densely interconnected as Canadian securities regulation, the quality of the dataset is in any event largely a function of how faithfully it captures relationships that carry legal meaning, and a relationship recorded merely as a generic link discards the very information that makes it useful. The schema should permit the OSC to express what kind of relationship obtains: that one provision supplies the controlling definition for another; that an instrument imposes a condition upon an exemption; that an order excepts certain issuers; that a later instrument supersedes an earlier one; that a companion policy interprets a rule.

The statutory and regulatory treatment of the "insider" for example, illustrates why untyped relationships are insufficient, and it is no marginal example. In Ontario, the principal insider reporting obligations reside in Part XXI of the Securities Act (Ontario) (the "Act") rather than in the national instrument, while the concepts that scope those obligations are distributed across several instruments. The Act defines "insider." National Instrument 55–104 introduces the narrower notion of "reporting insider," which in fact determines who must file, and supplies the definition of "significant shareholder" keyed to a ten per cent voting threshold. The applicable exemptions do not reside in any single place; they are spread across NI 55–104, NI 51–102, NI 62–103, NI 71–101, and NI 71–102, a dispersion the OSC's own guidance acknowledges by directing the reader to all of them. A companion policy governs how "director," "officer," and "beneficial ownership" are to be construed. For software to reach a sound conclusion as to whether a given person must file, it must apprehend not merely that these documents are connected, but that this provision narrows that one, that this instrument carves an exemption from that requirement, and that this definition controls that term. A plain link from an obligation to the word "insider" conveys none of this. A typed relationship does, and the signpost construct is the natural place to express it. Left to inference, each of these relationships becomes an interpretive judgment made by a software provider; declared by the OSC, each becomes a statement of the regulator's own understanding.

This is not novel territory. The international standards developed precisely for machine-readable law already model typed relationships as a first-class feature rather than an afterthought. Akoma Ntoso, adopted as an OASIS standard in 2018 and used in derivative form by the European Parliament, the United States, Italy, and the United Nations, models legal documents on the FRBR hierarchy and supports typed references and typed amendment effects rather than generic links^[8]. The United Kingdom's legislation.gov.uk goes further and publishes a structured feed of amendment effects in which each change is typed, for example as words substituted, a provision repealed, or an application restricted. And LegalRuleML, adopted as an OASIS standard in 2021, was designed specifically to capture the features that distinguish legal norms, including the ability to attach several alternative interpretations to a single provision, each carrying its own provenance metadata^[9], a capability directly responsive to the concern that animates this letter. The OSC need not adopt any of these standards wholesale, but it should treat them as evidence that typed relationships are both achievable and expected of a serious legal dataset.

Versioning is the temporal dimension of the same problem, and it warrants first-class representation in the schema rather than relegation to prose. Here too the comparative practice is settled. Point-in-time retrieval, the ability to obtain the law as it stood on a given date, is a solved problem implemented as a matter of course by the United Kingdom, Australia, and New Zealand, and, tellingly, by Canada's own federal government, whose Justice Laws website provides previous-version access to consolidated Acts and regulations^[10]. That a sister regulator's own national government already maintains point-in-time legislation makes the standard difficult to characterize as aspirational. The insider regime again makes the point concrete: NI 55–104 contains provisions that turn upon whether a transaction occurred before or after a stipulated date, and the Ontario insider trading provisions were extended from reporting issuers to all issuers at a fixed point in time. A dataset that records only the current consolidated text, or that relies upon embedded qualifiers such as "for transactions after this date," compels every downstream user to reconstruct the state of the law as it stood at a given moment, and every such reconstruction is an occasion for interpretive divergence. The regulator is uniquely positioned to express that history authoritatively, and to do so is essential both for ongoing compliance and for any matter in which the governing law is the law as it stood on a prior date.

3. A Machine-Readable Rulebook Attains its Full Value Alongside Machine-Readable Filings

The consultation is scoped to the regulatory instruments themselves. We would encourage the OSC to look one layer further, because a machine-readable rulebook becomes fully useful only when obligations can be tested against actual disclosure, and that in turn requires programmatic access to filings. Here Canada exhibits a structural gap that the present initiative renders difficult to overlook.

SEDAR+ affords no free, bulk, programmatic access. The contrast with the United States is instructive. EDGAR is free to anyone without an API key, full-text searchable across filings submitted since 2001, bulk-downloadable, and served by documented REST APIs that the SEC launched in 2021^[11], and two decades of regulatory technology have been constructed upon that foundation. That openness is a substantial part of the explanation for the depth of the American RegTech and analytics ecosystem. Nor is the United States alone in treating open, machine-readable access to disclosure as the destination. The European Union has legislated a European Single Access Point that will, from 2027, provide free, machine-readable access to corporate and financial disclosures through a mandatory API, on the express principle that the information be available to users free of charge^[12]. The international current, in other words, runs decisively toward free and programmatic disclosure access.

The contrast is sharpened by the prevailing direction of travel on this side of the border. SEDAR+ operates on a cost-recovery basis, and under final amendments to Multilateral Instrument 13–102 – System Fees, published by the CSA on 10 July 2025, system fees rose by approximately sixty per cent on 28 November 2025, with further annual increases of approximately three per cent in each of the following four years through 2029^[13]. The merits of cost recovery for the filing function may be debated on their own terms. But a disclosure system that is both fee-gated and non-programmatic sits uneasily beside a stated ambition to render the rulebook machine-readable, and it diverges from the path that the United States and the European Union have chosen. The rules layer and the filings layer are where value compounds, for it is only together that obligations may be mapped to the disclosures intended to satisfy them. To address one and not the other is to strand most of the benefit of both. A machine-readable rulebook unaccompanied by any machine-readable means of testing compliance against it is, in the end, a half-built bridge.

We recognize that SEDAR+ resides at the CSA level and is not the OSC's to resolve alone. Our request is accordingly specific: that the OSC exercise its influence within the CSA to advocate for free, programmatic, bulk access to SEDAR+ data, of the kind that EDGAR has long provided and that the European Single Access Point will shortly provide. This is no departure from the OSC's mandate. It is central to it, to capital formation, to fair and competitive markets, to the reduction of burden, and to the regulatory technology innovators the overview identifies as intended beneficiaries of this very initiative.

We appreciate the opportunity to comment and would welcome the opportunity to engage further as the work proceeds, including the questions of taxonomy governance and implementation that the OSC has signaled it will revisit.

Respectfully,

Material