One Common Language

Four duties that investment firms in Brussels, London and New York carry, set side by side in the regulators’ own words, turn out to be the same four duties: communicate honestly, execute orders well, keep records, know the customer before recommending. The words differ; the register does not. A baseline test run for this series, using a deterministic keyword matcher on 556 clauses from six rule texts, found 218 distinct duties and, at a 0.30 word-overlap threshold, matched 39 of them across a border, every one between the UK and the EU and not one in the United States, because the British text was taken from the European text and the American text was not. Eight other industries have been here. Payments, trading, derivatives, corporate reporting, entity identification, insurance, health records and retail each built a common machine-readable language for a shared problem; active Legal Entity Identifiers went from over 481,000 in March 2017 to over 3.1 million in June 2026, and on 24 November 2025 Swift observed 97% of payment instructions sent in ISO 20022. UK regulators have worked on machine-executable reporting rules since a 2016 TechSprint; nearly ten years on, the FCA publishes its Handbook as machine-readable data, not as machine-executable rules. None of these languages describes what a firm owes. This piece sets out what one would have to contain.

In this piece: four duties in three traditions; what a word-matcher finds; eight industries that built a common language; what the numbers say about the effect; regulators and machine-readable rules; the ontology that exists; what a compliance standard would have to contain.

Four duties, three legal traditions

The European Union writes its conduct rules as directives and regulations, drafted in Brussels and applied by national authorities. The United Kingdom writes them as a Handbook of numbered rules and guidance; COBS 4.2.1R, COBS 11.2A.2R, SYSC 9.1.1AR and COBS 9A.2.1R, all quoted below, each carry a bracketed note naming the MiFID article they transpose. The United States writes them as self-regulatory rules of FINRA, layered on statutes and SEC rules, in a common-law register that speaks of “reasonable diligence” and “fair dealing” rather than “all sufficient steps”. The three traditions do not share a vocabulary, a numbering scheme or a theory of what a rule is.¹²³

Yet a firm with a licence in all three places does not carry three different jobs. The table below places the same four duties side by side, verbatim, as published on the ESMA, FCA and FINRA websites.

Read down any column and the legal tradition shows. Read across any row and the duty shows. The record-keeping rule is a clear case: SYSC 9.1.1AR is Article 16(6) split into two sentences, with “common platform firm” for “investment firm”, “must” for “shall”, “the FCA” for “the competent authority” and “the regulatory system including MiFID, MiFIR and the Market Abuse Regulation” for the list of EU acts. FINRA says the same thing in twenty-three words by pointing at other rules, and its paragraph (c) takes the format of records from an SEC rule, SEA Rule 17a-4. The retention periods differ: at least five years in the UK rule; in FINRA’s, at least six years for records with no period set elsewhere.²³ Those differences are real and a firm must track them. They are differences in the parameters of one duty, not evidence of two duties.

Best execution shows the limits of the register. The EU and UK ask for “all sufficient steps” to obtain “the best possible result” against a list of six execution factors. FINRA asks for “reasonable diligence to ascertain the best market” so that the price is “as favorable as possible under prevailing market conditions”, and then lists five factors that “will be considered” in judging that diligence. The standards of effort are not identical; the gap between “sufficient” and “reasonable” is a matter of legal argument. But the obligation is the same shape: when executing a client order, take steps, weigh these factors, obtain the best outcome available, and be able to show it. A compliance program that has that shape written down once can hold the EU, UK and US parameters as three settings of one duty.¹³

The suitability row shows the register under strain in the other direction. MiFID II and COBS 9A trigger the duty by service: it applies “when providing investment advice or portfolio management”. FINRA 2111 triggers it by act: it applies to “a recommended transaction or investment strategy”. The information to be gathered overlaps heavily (knowledge and experience, financial situation, objectives, risk tolerance) but FINRA’s list adds age, tax status, time horizon and liquidity needs, and the EU and UK texts add “ability to bear losses”. A firm that treats these as one duty with jurisdiction-specific fields will do better than one that keeps three unconnected policies; a firm that treats them as identical will miss the trigger difference and be wrong in one market or another.¹²³

What a word-matcher finds

If the register is shared and only the words differ, a machine that compares words should find the European and British twins and miss the American ones. That is a testable prediction, and a baseline test was run for this series to test it, using a deliberately simple method with no language model at any stage: a deterministic keyword extractor and a word-overlap matcher. The code is public at github.com/Reg42-ai/clhear.⁵

The inputs were six rule texts on two topics, client communications and best execution, read from the ESMA, EUR-Lex, FCA Handbook and FINRA websites on 15 October 2026: COBS 4 and COBS 11.2A from the FCA Handbook, FINRA Rules 2210 and 5310, MiFID II Articles 24 and 27, and Articles 44 to 47 and 64 to 66 of Delegated Regulation 2017/565.⁴ Split into provisions, they came to 556 clauses, of which 421 were binding; the other 135 were FCA guidance and one placeholder for a deleted section. The extractor found 223 duties among them by their modal verbs; an in-country dedupe step at 0.92 word overlap merged five near-duplicates, leaving 218 distinct duties: 99 British, 59 American, 60 European.⁵

The matcher then compared every duty with every duty in another jurisdiction and counted a match when the content words the two share made up at least a set fraction of all the distinct content words in the pair. At a strict setting of 0.60, it found 11 pairs covering 19 duties. At a loose setting of 0.30, it found 23 pairs covering 39 duties; a separate grouping step at the same setting, with a different word list, placed 39 duties in 18 groups. The top pair, COBS 4.5A.9R and Article 46(5) of the delegated regulation, scores 0.93: 13 of the 14 distinct content words in the two texts are shared. Reading them shows why: they are the same sentence, except that “A firm must” replaces “Investment firms shall” and “investment services and ancillary services” replaces “investment and ancillary services”.⁴⁵

Every one of the 23 pairs at 0.30 runs between the UK and the EU. Not one of the 59 American duties was matched at any setting from 0.15 up; at 0.10, the loosest setting tried, 17 were, and no pair involving a US duty scored above 0.14. No group spans all three jurisdictions.⁵

The prediction held, and the explanation is historical rather than clever. The FCA’s MiFID rules were taken from the MiFID texts, so the words survived the crossing; the American rules were written in a different tradition on a different continent, so the same duty arrived in different words. The first two rows of the table above, the two topics the run covered, are such a case: FINRA’s “fair and balanced” and the EU’s “fair, clear and not misleading” are one duty to a reader and no match to a word-counter at any setting tried.

Two cautions. The run is small, two topics and three jurisdictions, and the extractor is crude; a hand check of 40 of the 223 detected duties in the earlier article in this series found 35 duties and 5 scope or deeming clauses that slipped through. And a 0.30 threshold is generous; some of the 39 matches are weak. Neither caution changes the shape of the result: on this run, at every setting above 0.10, word overlap did not cross the Atlantic. A shared language for duties has to sit above the words, at the level of what is owed, to whom, under what trigger, with what parameters, and the words have to be attached to it as evidence rather than used as the key.

Eight industries that built a common language

Finance has solved this class of problem before, several times, for other kinds of information. So have insurance, health care and retail. In each case an industry faced the same object described in incompatible local dialects, agreed a common machine-readable form, and then spent years moving onto it. The milestones below are taken from the standard bodies themselves and from the regulators that mandated them.

Retail: the barcode. GS1 dates the first commercial scan to 26 June 1974, when a cashier at a Marsh Supermarket in Troy, Ohio, scanned “a 10-pack of Wrigley’s Juicy Fruit gum that cost 67 cents”. The European numbering body followed in 1977; the American and European bodies merged in 2005 into one organisation with 101 member organisations, 116 by 2023. In its 2014 anniversary release GS1 said the barcode’s beep would be heard “as many as 5 billion times today” and that it was “driven by over a million user companies”.¹⁶

Trading: FIX. The Financial Information eXchange protocol began as a bilateral link between Salomon Brothers and Fidelity; according to a history by Kevin Houstoun posted on the FIX Trading Community website, it “was first used in anger during 1993”, after which Goldman Sachs and Putnam were invited in. The standards body launched in 1998 and today lists “270+ leading financial service companies” as members, across equities, fixed income, listed derivatives and foreign exchange.⁸

Derivatives: FpML. JP Morgan started a research project in 1997 with PricewaterhouseCoopers; “in 1999 the organizations announced a draft standard for interest rate swaps”, and on 14 November 2001 ISDA and FpML.org announced their intention to integrate its development process into ISDA. The standard is freely licensed.¹³

Corporate reporting: XBRL. The SEC adopted its interactive-data rule on 30 January 2009, after a voluntary programme in which “Over 100 companies have participated”. Large accelerated filers using US GAAP with a worldwide public common equity float above $5 billion were required from fiscal periods ending on or after 15 June 2009; all other large accelerated filers from June 2010; “All remaining filers using U.S. GAAP, including smaller reporting companies” from June 2011.⁹ The EU followed with the European Single Electronic Format: a Commission delegated regulation of 17 December 2018 requires listed issuers to prepare annual financial reports in XHTML and to mark up IFRS consolidated statements in Inline XBRL, for financial years beginning on or after 1 January 2020. Its stated purpose is “to further facilitate their accessibility, analysis and comparability”.¹⁰

Entity identification: the LEI. The Financial Stability Board’s account is that “The global financial crisis highlighted significant obstacles to identifying and tracing financial transactions across the international financial system”. At Cannes in November 2011 the G20 asked the FSB to lead work on a global LEI; the FSB’s recommendations were endorsed at Los Cabos in June 2012, with a target to launch “by March 2013”.¹¹ GLEIF’s first quarterly report, in April 2017, counted “over 481,000 active LEIs managed by 30 LEI issuers” across 199 jurisdictions. GLEIF reported that during 2017 the LEI population “nearly doubled to approximately 953,000”, with 77 percent of the year’s issuance in the final quarter, which it attributed primarily to firms seeking compliance with MiFID II and MiFIR, in force from 3 January 2018. The FSB counted “over 1.4 million entities uniquely identified by an LEI” in May 2019; GLEIF reported 1.77 million active LEIs at the end of March 2021, 2.25 million at the end of March 2023 and “over 3.1 million” at the end of June 2026.¹¹¹²

Payments: ISO 20022. Swift’s community decided in 2018 to adopt ISO 20022 for cross-border payments. Coexistence with the old MT messages began in March 2023; by June 2025 “over 40% of daily traffic” was in the new format; on 22 November 2025 coexistence with the MT format ended, and on 24 November Swift observed “97% of payment instructions sent using ISO 20022”. Swift’s platform connects “more than 11,500 banking and securities organisations, market infrastructures and corporate customers”.⁶⁷

Insurance: ACORD. The insurance standards body dates itself to 1970 and reports “more than 36,000 participating organizations” in “more than 100 countries”.¹⁴ Health: HL7 FHIR. The Fast Healthcare Interoperability Resources standard, created by Grahame Grieve and a small team working through HL7, a body founded in 1987, was first presented in 2011. The US Office of the National Coordinator required certified health IT developers to move their certified APIs to the FHIR-based criterion and provide them to customers by 31 December 2022; products certified to that criterion went from five in early 2022 to 253 by February 2023, and “More than 95 percent of Certified Health IT developers met the compliance deadline”.¹⁵

What the numbers say about the effect

The claim under test is that each common language made cost and time fall. The honest answer from the primary documents is that adoption is measured carefully and effect is mostly asserted. Swift’s November 2025 release describes “faster, more efficient and data-driven payments” and quantifies none of it; the two efficiency figures in its June 2025 update are statements by adopting banks, Deutsche Bank’s “increase our efficiencies by over 40%” and Diamond Trust Bank’s “reduction of up to 50% of alerts generated on incoming payments”, which are claims, not audits.⁶⁷ ACORD says its standards deliver “billion-dollar savings” without showing the sum.¹⁴ Houstoun’s history of FIX says the protocol “has undoubtedly delivered significant efficiency” and offers no number.⁸ The SEC said its rule would make financial information “easier for investors to analyze” and that the automation it allows “could eventually reduce costs”; the EU regulation gives its aim as “to further facilitate their accessibility, analysis and comparability”.⁹¹⁰

What the documents do show is three things. First, that adoption took years and, in five of the eight cases, moved when a regulator or the network operator required it: the SEC turned a voluntary XBRL programme into a mandate, and ESEF was a mandate; GLEIF attributes the LEI’s issuance spike in late 2017 primarily to firms seeking compliance with MiFID II and MiFIR; FHIR’s certification numbers moved when a federal deadline arrived; Swift ended coexistence with the old format on 22 November 2025.^91012156 Second, that coverage stalls where no one requires it. The FSB’s 2019 peer review found LEI coverage “too low outside securities and derivatives markets to encourage new industry or regulatory uses”, and GLEIF reports an overall LEI renewal rate of 57.1% in the second quarter of 2026.¹¹¹² Third, that use can keep growing after a mandate’s deadline has passed: the active LEI population rose from 1.77 million in March 2021 to over 3.1 million in June 2026, years after MiFID II applied in January 2018.¹²

The evidence therefore supports a narrower claim than the one this article set out to test. The documents reviewed here do not show that a common machine-readable language by itself lowers cost; the migration is itself a cost, and none of the standards bodies’ pages cited here publishes audited savings. What a common language does is make a shared problem tractable at scale, so that a regulator can mandate, a vendor can build once, and a firm can connect once. Whether the saving arrives depends on what is built on top.

Regulators and machine-readable rules

Every standard above describes data: a payment, an order, a swap, a balance sheet, an entity, a policy, a patient, a tin of beans. None describes a rule. In the UK, the attempt to do that, to write regulation itself in a form a machine can read and act on, began with a TechSprint in November 2016 and is set out in a series of official reports.¹⁷

The FCA held a TechSprint on digitising reporting instructions in November 2016 and a joint one with the Bank of England a year later. The two regulators then ran two Digital Regulatory Reporting pilots with seven banks: Barclays, Credit Suisse, HSBC, Lloyds, Nationwide, NatWest and Santander. Phase 1 ran from June to December 2018 and explored automating reporting by writing regulation in a machine-executable format; phase 2 ran from February to October 2019, tested domain-specific languages for generating it and modelled the cost-benefit case. The phase 2 viability assessment, published in January 2020, is candid. For mortgage reporting alone, “the standalone viability of DRR (limited to just mortgage reporting) is likely to be low”, though the idea “showed sufficient promise to warrant continued investment and exploration”. The survey behind it put a large regulated firm’s mortgage reporting at “an annual average of £450,000 for run and a further £700,000 on change related costs”, not counting technology costs, and the report put the “overall efficiency factor for the base scenario” at 30 to 40%, with the caveat that “The numbers were generated for this report through limited industry surveys”.¹⁷

The barriers the pilot named are the ones this article is about. “The lack of standardisation of the definitions and descriptions of data by firms is a significant barrier to improving both firm processes and exploitation of new analytical techniques.” “Sometimes the rules that may map a term to another term cannot be coded unambiguously.” “Interpreting the instructions incorrectly can have serious legal, operational and financial consequences.” The independent review by PA Consulting in September 2020 recommended “developing a Data Dictionary which precisely defines terms in the regulatory rulebook” and said that generating regulatory reports “should remain decentralised so that firms can implement their own solutions based on a set of open standards”.¹⁷

In February 2021 the Bank of England published a plan for a wider programme, drawing on ideas explored in the DRR pilots, whose vision is that “The Bank gets the data it needs to fulfil its mission, at the lowest possible cost to industry” and whose reforms it expects to need “at least the ten years envisioned in the original review scope”. In March 2024 the Bank and FCA responded to the industry’s phase two recommendations with five outcomes, an Industry Data Standards Committee “responsible for issuing industry recommendations on data standards” and reviews of existing data collections; the response does not mention machine-executable regulation.¹⁸ In August 2026 the FCA launched a Handbook API that “enables systems to access structured Handbook content directly” in “a structured, machine-readable format”, free to registered users, with the limitation that it “does not provide past versions of Handbook content”.¹⁹ That is a real step: the text of the rules is now data. It is not the same as the rules being data. The API serves the sentence “A firm must ensure that a communication or a financial promotion is fair, clear and not misleading”; it does not say which control, or what evidence, would show the duty met.

Elsewhere the record is similar. The OECD’s 2020 report on Rules as Code, which cites New Zealand’s Better Rules programme and work in France as early successes, defines the aim as “a machine-consumable version of some types of government rules” and says the field “remains an early stage concept with much to test”, with “unresolved questions about the legal status of machine-consumable laws” and about how “errors in the coded form of rules could be identified and corrected”. The BIS Innovation Hub’s Project Ellipse, with the Monetary Authority of Singapore, the Bank of England and ISDA, concluded in March 2022 that “regulatory reporting requirements can be expressed in unambiguous machine-readable logical reporting instructions”, that “if a common data standard was agreed to and implemented, financial institutions may no longer need to interpret reporting instructions”, and that the exercise “was intentionally limited in scope”; its prototype used “synthetic bank exposure data”.²⁰

Two patterns run through all of it. Where the target was a reporting template, a finite set of fields with definitions, the projects reached working prototypes and then turned to agreeing data definitions across firms. None of the projects reviewed here took on a conduct rule, with a trigger, a standard of effort and a judgement, and the OECD’s phrase “some types of government rules” marks the boundary. Reporting rules are about data, and these projects reduced some of them to data. Conduct rules are about duties, and none of these projects expressed one as data.

The ontology that exists

Finance does have a formal shared vocabulary of its own. The Financial Industry Business Ontology, FIBO, is described by its maintainers as a “global, standardized ontology (formal knowledge model)”, “available in RDF/OWL” and “organized into modular domains covering all major areas of finance”; it is open source and maintained by the EDM Association, and its listed uses include “Regulatory reporting consistency, e.g., for BCBS 239”. Its repository lists ten domains, from business entities and business process to derivatives, loans, market data and securities. FIBO can say precisely what an interest rate swap is, down to fixed-float and cross-currency swaps. Its foundations also define general legal terms, a “Duty” as “some obligation which exists and is imposed on some individual” and a “Regulation” as “a rule used to carry out a law”, and it has a general concept of a program, which a change note says was added in part “to represent compliance”. In a search of its class names made for this article, none contains “Compliance”.²⁰ A compliance language would have to state, in one agreed form, that a firm owes its client a best-execution policy under COBS 11.2A, that the policy is document D, that control C tests it monthly, that system S produces the evidence, and that person P signed the mapping.

That is the shape of the gap. The data standards describe things. The ontology describes the kinds of things, with general terms for duties, regulations and programs. The regulators’ projects describe the fields a regulator wants. None of the documents reviewed here sets out a model of the compliance program itself: the set of duties a particular firm owes, each tied to its source text in each jurisdiction, each tied to the roles, documents, systems and processes that discharge it, and each tied to the firm’s profile, which is what decides whether the duty applies at all.

What a compliance standard would have to contain

The evidence above sets the design constraints for anyone who tries to build one. This is the design proposed in the founder letter this series accompanies: an open standard, in development, called CLHEAR. What follows is the proposal, not a description of something that exists.

• Rules. Every source provision, with its jurisdiction, citation, version, effective dates and verbatim text, so that lineage runs back to the words. The FCA Handbook API now serves this layer as data for one of the three jurisdictions; ESMA’s Interactive Single Rulebook publishes the EU text without a consolidation date, and FINRA’s rule pages carry version or adoption dates. This layer exists in pieces and needs stitching, not inventing.
• Duties. The register of obligations above the words: who owes, what, to whom, under what trigger, with what parameters, by what standard of effort. The side-by-side table is four entries of it. Each entry carries pointers to one or more rules in each jurisdiction, with the parameter differences (five years, six years) held as settings rather than separate duties. Equivalence between an EU rule and a US rule is a human assertion with a signature, not a word score.
• Roles. The people and functions accountable for each duty, mapped to the regulators’ own responsibility regimes.
• Documents. Policies, procedures and attestations, each tied to the duties it discharges, so that a changed rule points to the pages that have to change.
• Systems. The applications that produce the evidence, with the data they hold, so that “records to be kept of all services, activities and transactions” resolves to named systems and retention settings.
• Processes. The controls and their schedule, with the evidence each one produces.
• Firm profile. Licences, jurisdictions, client types, products and services. The profile is what decides which duties apply: a firm that provides neither investment advice nor portfolio management does not carry COBS 9A.2.1R, and opening a new country means changing the profile and re-deriving the map, not re-reading the rulebook from scratch.

Three lessons from the eight industries would apply. The standard must be open and free to use, as FIX, FpML and FIBO are, or it will not be adopted across competitors. It must separate the shared language from the local implementation, as the PA Consulting review recommended, so that firms and vendors build their own tools on it. And it should expect the pattern the LEI and FHIR show: slow uptake until a regulator or a large counterparty asks for it, then a steep curve.

Two cautions from the same record also apply. The migration is a cost before it is a saving, and none of the standards bodies’ pages cited here publishes audited figures for the saving; the targets the letter sets for a firm running on such a standard (an 80% cut in compliance cost, a score at any moment, an answer in 60 seconds) are targets, and nothing in this article is evidence that they have been met. And the regulators’ own reports name the lack of agreed data definitions as “a significant barrier”; for duties, the agreement needed is about who decides that FINRA 5310 and MiFID II Article 27 are one duty, and who signs. The design answer is that a person does, every time, with the text on both sides of the screen, and the machine’s job is to put the right two texts there. The baseline test in this article shows how badly a machine does that job with words alone. The next article in the series looks at what the published evidence says about trusting a machine with compliance work.

Notes and sources

1. Directive 2014/65/EU (MiFID II), Articles 16, 24, 25 and 27, read from ESMA’s Interactive Single Rulebook; Articles 24 and 27 on 15 October 2026 for the baseline run, Articles 16 and 25 on 17 October 2026. ESMA’s pages show no consolidation date; the Article 27 text reflects the amendments made by Directive (EU) 2024/790.
2. FCA Handbook: COBS 4.2.1R and COBS 11.2A.2R (read 15 October 2026; chapter COBS 4 last updated 6 April 2026, section COBS 11.2A last updated 23 October 2025); SYSC 9.1.1AR and SYSC 9.1.2R (dated 03/01/2018) and COBS 9A.2.1R (dated 01/10/2018), read 17 October 2026. The Handbook prints COBS 4.2.1R with a note naming “article 24(3) and article 30(1) of MiFID”, COBS 11.2A.2R with “[Note: article 27(1) of MiFID]”, the SYSC record-keeping rule with “[Note: article 16(6) of MiFID]” and the COBS suitability rule with “Note: first paragraph of article 25(2) of MiFID, first paragraph of article 30(1) of the IDD”. In the table, sub-paragraph labels are kept as printed.
3. FINRA Rules 2210 (version “Aug 16, 2019 onwards”) and 5310 (“May 09, 2014 onwards”), read 15 October 2026; 4511 (adopted by SR-FINRA-2010-052, effective 5 December 2011) and 2111 (“Jun 30, 2020 onwards”), read 17 October 2026. Rule 4511(c) adds that records must be preserved “in a format and media that complies with SEA Rule 17a-4”; the US retention regime therefore sits partly in SEC rules outside the FINRA rulebook.
4. Commission Delegated Regulation (EU) 2017/565, Articles 44 to 47 and 64 to 66, EUR-Lex consolidated text of 2 August 2022, read 15 October 2026. Article 46(5): “Investment firms shall ensure that information contained in a marketing communication is consistent with any information the firm provides to clients in the course of carrying on investment and ancillary services.” COBS 4.5A.9R: “A firm must ensure that information contained in a marketing communication is consistent with any information the firm provides to clients in the course of carrying on investment services and ancillary services.”
5. The baseline test: a run made for this series with a deterministic keyword pipeline (extractor version deterministic-v6), code public at github.com/Reg42-ai/clhear, app/clhear/l2. Inputs: FCA COBS 4 (243 provisions) and COBS 11.2A (37), FINRA Rules 2210 (177) and 5310 (34), MiFID II Articles 24 and 27 (24 paragraphs), Delegated Regulation 2017/565 Articles 44 to 47 and 64 to 66 (41), 556 provisions in all. Binding provisions (FCA rules and evidential provisions, FINRA rule text and supplementary material, EU articles): 421. Duties detected from modal verbs (“must”, “must not”, “shall”, “required”): 223, of which UK 104, US 59, EU 60. In-country dedupe at 0.92 word overlap merged 5, leaving 218. Cross-border word-overlap matching (shared content words as a share of all distinct content words in the pair): at 0.60, 11 pairs covering 19 duties; at 0.30, 23 pairs covering 39 duties. Every pair at these two settings is EU-UK. A separate grouping step at 0.30, with a different word list, placed 39 duties in 18 groups, none spanning three jurisdictions. Top pair: COBS 4.5A.9R and Article 46(5), similarity 0.93 (13 of 14 distinct content words shared). A re-run of the same code on the same texts on 20 October 2026 reproduced every figure above; it found no match for any US duty at any setting from 0.15 up and, at 0.10, 26 pairs involving 17 of the 59 US duties (12 UK-US, 14 EU-US), the closest scoring 0.138. The provisions in the table’s first two rows (FINRA 2210(d)(1)(A) and 5310(a)(1) against COBS 4.2.1R, COBS 11.2A.2R and MiFID II Articles 24(3) and 27(1)) score between 0.027 and 0.073. No language model was used at any step. This is a baseline for comparison, not the CLHEAR method.
6. Swift, press release, “Global financial community completes switch to ISO 20022”, Brussels, 25 November 2025: “The final switch happened on 22 November, with the end of coexistence with the traditional MT message format”; “In March 2023, Swift began a period of coexistence with the MT format for cross-border payments and reporting”; the standard was chosen for cross-border payments “starting in 2018”; “Yesterday we observed 97% of payment instructions sent using ISO 20022”; Swift connects “more than 11,500 banking and securities organisations, market infrastructures and corporate customers” in “more than 200 countries and territories”. The release quantifies no efficiency gain.
7. Swift, news, “Momentum builds as industry advances ISO 20022 adoption”, 23 June 2025: “over 40% of daily traffic now using the ISO 20022 format”, “180+ sending countries and 220+ receiving countries”, “Swift now carries over 75% of its Payments Market Infrastructure traffic in ISO 20022 format”. The two efficiency figures in the text are statements by individual banks quoted in that release: Deutsche Bank, “we’ve been able to increase our efficiencies by over 40%”, and Diamond Trust Bank, “a reduction of up to 50% of alerts generated on incoming payments”. They are claims by adopters, not audited results.
8. FIX Trading Community, “What is FIX Trading Community?”: “the non-profit, industry-driven standards body at the heart of global trading”; the organisation launched in 1998; “membership includes 270+ leading financial service companies”; “supporting the equities, fixed income, listed derivatives and foreign exchange markets”; it promotes “non-proprietary, free and open standards”. Kevin Houstoun, “The long journey to frictionless electronic trading”, published by The Trade News on 28 October 2020 and posted on the FIX Trading Community website: the protocol, “initially dubbed SBX (Salomon Brothers Exchange)”, “was first used in anger during 1993” between Salomon Brothers and Fidelity, who then invited Goldman Sachs and Putnam. Neither page gives a market-share or cost figure; Houstoun writes only that FIX “has undoubtedly delivered significant efficiency to the industry”.
9. US Securities and Exchange Commission, Interactive Data to Improve Financial Reporting, Release No. 33-9002, 30 January 2009, published at 74 FR 6776 on 10 February 2009, effective 13 April 2009. Phase-in: large accelerated filers using US GAAP with “a worldwide public common equity float above $5 billion” from periods ending on or after 15 June 2009; all other large accelerated filers from 15 June 2010; “All remaining filers using U.S. GAAP, including smaller reporting companies”, with foreign private issuers reporting under IFRS as issued by the IASB, from 15 June 2011. “Over 100 companies have participated in the voluntary program” before the rule. The rules are “intended not only to make financial information easier for investors to analyze, but also to assist in automating regulatory filings and business information processing”; on costs, the release says “Such automation could eventually reduce costs.” GAO, GAO-09-389R, confirms the citation and effective date.
10. Commission Delegated Regulation (EU) 2019/815 of 17 December 2018 on the single electronic reporting format, EUR-Lex: Article 3, “Issuers shall prepare their entire annual financial reports in XHTML format”; Article 4, IFRS consolidated statements are marked up and “issuers shall use the XBRL markup language”; Article 6 requires the Inline XBRL specifications in Annex III; it applies to financial years beginning on or after 1 January 2020. Recital 4 gives the purpose: “In order to further facilitate their accessibility, analysis and comparability.”
11. Financial Stability Board, press release of 8 June 2012: the aim of the LEI system is “to uniquely identify counterparties to financial transactions”; the FSB report “responds to the mandate issued by the G20 at the Cannes Summit”; the implementation plan “targets launch of the global LEI system on a self-standing basis by March 2013”. GLEIF, history page: “At its November 2011 Cannes Summit, the G20 called upon the FSB to take the lead in developing recommendations for a global LEI and supporting governance structure.” FSB, LEI topic page: “The global financial crisis highlighted significant obstacles to identifying and tracing financial transactions across the international financial system”; the recommendations “were endorsed by the G20 at the Los Cabos Summit in June 2012”. FSB, Thematic Review on Implementation of the LEI, 28 May 2019: “over 1.4 million entities uniquely identified by an LEI in more than 200 countries”; coverage “concentrated in Canada, the EU and the United States”; “Coverage is too low outside securities and derivatives markets to encourage new industry or regulatory uses.”
12. GLEIF blog and business reports: Q1 2017 (13 April 2017: “over 481,000 active LEIs managed by 30 LEI issuers”, “199 jurisdictions”); Q4 2017 (14 February 2018: “In 2017, the LEI population nearly doubled to approximately 953,000 LEIs”, a total the post does not describe as active; 77 percent of the year’s issuance came in the fourth quarter, which GLEIF attributes “primarily to impacted market participants seeking to achieve compliance” with MiFID II and MiFIR, in force from 3 January 2018); Q1 2021 (1.77 million); Q1 2023 (2.25 million); Q2 2023 (around 2.31 million, renewal rate 59.4%); Q3 2023 (around 2.36 million); Q1 2026 (28 April 2026: 3.02 million); Q2 2026 (23 July 2026: “92,000 organizations worldwide obtaining an LEI”, active LEI population “over 3.1 million”, overall renewal rate 57.1%).
13. ISDA, History of FpML: JP Morgan “in 1997 established a research project” on electronic derivatives trading with PricewaterhouseCoopers; “in 1999 the organizations announced a draft standard for interest rate swaps”; “On November 14, 2001 ISDA and FpML.org announced their intention to integrate the development process” into ISDA. The standard is “freely licensed”. The page gives no adoption count.
14. ACORD, About ACORD: “the global standards-setting body for the insurance and related financial services industries”; “Since 1970”; “ACORD engages more than 36,000 participating organizations” “spanning more than 100 countries”; the body claims its standards “realizes billion-dollar savings to the global insurance industry”, without a published calculation.
15. HL7 International, “Happy 10th Birthday FHIR”: HL7 was “Founded in 1987”; FHIR “has evolved through 4 releases since its initial presentation in 2011”; the page names Grahame Grieve as “its creator”, working with a small team “through HL7 International”; FHIR grew “from a true draft standard with 49 Resources to its current 149”. US Office of the National Coordinator for Health IT, “Achieving a Major Milestone”, 10 February 2023: developers had to update certified APIs “to FHIR®-based APIs in 170.315(g)(10)” and “provide that updated certified API technology to customers by December 31, 2022”; “More than 95 percent of Certified Health IT developers met the compliance deadline”; products certified to that criterion rose from five in early 2022 to “253 products that offer this functionality”; the post refers to “the nation’s 5,000 hospitals and 500,000+ clinicians”.
16. GS1, press release on the 40th anniversary of the first barcode scan, Brussels, 26 June 2014: “on June 26, 1974, Sharon Buchanan was the first cashier to scan a GS1 barcode” at a Marsh Supermarket in Troy, Ohio, on “a 10-pack of Wrigley’s Juicy Fruit gum that cost 67 cents”; shoppers and supply-chain workers would hear the barcode’s beep “as many as 5 billion times today”; GS1 is “driven by over a million user companies” with Member Organisations “in over 110 countries”. GS1, historical timeline: EAN established in Brussels in 1977; “The UCC and EAN merge, creating a single, international organisation with 101 local GS1 Member Organisations” in 2005; 116 member organisations in 2023.
17. FCA, Digital regulatory reporting (page last updated 10 February 2023): DRR “emerged from an FCA TechSprint (hackathon) in November 2016 that explored the idea of digitising reporting instructions”; joint FCA and Bank of England TechSprint November 2017; “In 2018 and 2019, the 2 UK regulators collaborated with 7 banks to complete 2 pilots”, the banks being Barclays, Credit Suisse, HSBC, Lloyds, Nationwide, NatWest and Santander; pilot phase 1 June to December 2018, phase 2 February to October 2019; viability assessment released January 2020; the current phase “establishes the foundational requirements necessary to eventually create MRR and MER” (machine readable and machine executable regulation). The FCA’s page on that November 2016 event, Unlocking regulatory reporting, calls it “our second TechSprint”. FCA, DRR Pilot Phase 2 Viability Assessment: “the standalone viability of DRR (limited to just mortgage reporting) is likely to be low”; “DRR showed sufficient promise to warrant continued investment and exploration”; for “a large regulated firm” the cost of regulatory reporting in the mortgage domain “was estimated at an annual average of £450,000 for run and a further £700,000 on change related costs in addition to technology costs which were excluded”; “different firms included different activities within ‘change’ costs”; an “overall efficiency factor for the base scenario” of 30 to 40%; “The numbers were generated for this report through limited industry surveys”; “The lack of standardisation of the definitions and descriptions of data by firms is a significant barrier”; “Sometimes the rules that may map a term to another term cannot be coded unambiguously”; “Interpreting the instructions incorrectly can have serious legal, operational and financial consequences.” FCA, PA Consulting review of phases 1 and 2, 25 September 2020, an “independent review” commissioned by the FCA and the Bank of England: recommends “developing a Data Dictionary which precisely defines terms in the regulatory rulebook” and says the generation of regulatory reports “should remain decentralised so that firms can implement their own solutions based on a set of open standards”.
18. Bank of England, Transforming data collection from the UK financial sector: a plan for 2021 and beyond, 23 February 2021: vision that “The Bank gets the data it needs to fulfil its mission, at the lowest possible cost to industry”; three reforms, “common data standards, modernised reporting instructions and an integrated approach to data collection”; “We expect those reforms to require at least the ten years envisioned in the original review scope”; the review’s potential solutions “drew on ideas explored in” the joint Bank and FCA pilot on Digital Regulatory Reporting. Bank of England, Transforming data collection: phase one recommendations with Bank and FCA response, 21 July 2022: “The Bank and FCA agreed to accept all of the recommendations made by the industry committees in principle.” Bank of England and FCA, response to the phase two industry recommendations, 28 March 2024: “Five outcomes to ensure regulators get the data they need at the lowest cost”; the Industry Data Standards Committee, “a new committee that will be responsible for issuing industry recommendations on data standards”; reviews of FCA data collections and the Bank’s Banking Data Review. The response does not mention machine-executable regulation.
19. FCA, “Making compliance simpler: opening up the FCA Handbook through our new API”, Alex Smith, page dated 6 August 2026: “Access to FCA Handbook data is now available directly through our new API”; it “enables systems to access structured Handbook content directly”, with the Handbook “in a structured, machine-readable format”. FCA Handbook, Launch of Handbook API, listed in the Handbook’s Latest News with a published date of 6 August 2026: “It is free to use”; it is “available to all users who have registered for a free account on the website”; “The API does not provide past versions of Handbook content.” The API serves the text of rules as data; it does not express what a rule requires in a form a machine can execute.
20. OECD Observatory of Public Sector Innovation, Cracking the Code: Rulemaking for humans and machines, 12 October 2020: Rules as Code seeks to “create a machine-consumable version of some types of government rules”; it “remains an early stage concept with much to test”, with “unresolved questions about the legal status of machine-consumable laws” and about how “errors in the coded form of rules could be identified and corrected”; New Zealand’s Better Rules programme and work in France are cited as early successes. BIS Innovation Hub, Project Ellipse: an integrated regulatory data and analytics platform, March 2022, with the Monetary Authority of Singapore, the Bank of England and ISDA: phase 1 “investigated how data-driven supervision could be enabled by machine-executable digital reporting” on UK and Singapore mortgage requirements; it found that “regulatory reporting requirements can be expressed in unambiguous machine-readable logical reporting instructions”, and that “if a common data standard was agreed to and implemented, financial institutions may no longer need to interpret reporting instructions”; the exercise “was intentionally limited in scope” and “The prototype used synthetic bank exposure data.” EDM Council, FIBO: a “global, standardized ontology (formal knowledge model)”, “available in RDF/OWL”, “organized into modular domains covering all major areas of finance”; “FIBO is open-source”; “Developed and maintained by the EDM Association”; its listed uses include “Regulatory reporting consistency, e.g., for BCBS 239”. The FIBO repository’s README lists ten domain directories, from Business Entities and Business Process to Derivatives, Loans, Market Data and Securities. Its Interest Rate Swaps ontology has classes including fixed-float and cross-currency interest rate swaps; its Legal Capacity ontology defines “Duty” as “some obligation which exists and is imposed on some individual” and “Regulation” as “a rule used to carry out a law”. Its Objectives ontology carries a change note that the concept of a program was added, “required for IND but also to represent compliance, and other kinds of programs”. A search of the repository’s class names, made for this article on 20 October 2026, found none containing “Compliance”.