Sources

This site compiles enforcement data from the SEC (litigation releases, administrative proceedings and administrative law judge decisions), the CFTC, DOJ, Ontario's OSC, the UK's FCA and Australia's ASIC. US federal releases are works of the United States government and are not subject to copyright; our compilation and writing are published under CC BY 4.0 and are free to reuse with attribution.

Machine-readable exports

Start here if you want the data rather than the pages.

All of it is free to reuse with attribution to this site and a link back.

Sources we collect

SEC litigation releases

The Commission’s published notices of actions filed or resolved in federal court, from 2015 forward. This is the single largest contributor to the library. Each release names the respondents, carries a release number, and usually links the complaint.

Licence: work of the United States government, not subject to copyright.

SEC administrative proceedings

Orders instituting administrative proceedings, published as PDFs, from 2013 forward — 10,100 orders read. Text is extracted from the PDF, the order type is identified from its own title, and only substantive findings orders go on to technique classification.

Note that this source now reaches back two years further than SEC litigation releases, which begin in 2015. Any comparison of 2013 or 2014 with a later year is therefore comparing a smaller set of sources, and the year columns on the charts say so.

Most of this docket is not enforcement for misconduct, and we do not record it as though it were. Of a 1,500-order sample, 39% were substantive. What is excluded, and why:

A Rule 102(e) order that finds the professional personally engaged in the underlying violation — an accountant who traded on inside information, an attorney who authored the false filings themselves rather than merely failing to catch someone else’s — is recorded like any other enforcement action. The distinction is read from the order itself, not assumed from the respondent’s job title.

The ALJ initial decisions source below applies the identical test. The two used to disagree — one recorded every substantive 102(e) order, the other excluded every one regardless of what it found — and now defer to the same function so a 102(e) order is judged once, the same way, wherever it turns up. See DECISIONS.md §20.

Every exclusion is written to the ingest log with the order type and the reason, so the decisions are auditable rather than invisible.

Licence: work of the United States government, not subject to copyright.

SEC administrative law judge initial decisions

Initial decisions of the Commission’s administrative law judges: the outcome of contested administrative proceedings, and the richest documents the SEC publishes. A judge who has heard evidence sets out findings of fact about what a respondent did, which is exactly what a technique classifier needs.

Three things are checked before any of it is recorded.

Whether the decision is final. An initial decision is not the end of a matter — it can be appealed to the Commission, and it becomes final only when a finality order issues. Recording one that was later reversed, or one of the many vacated and reheard after Lucia v. SEC (2018) because the judge who wrote it had been unconstitutionally appointed, would publish findings that were formally set aside. The index links a finality order where one exists, and only decisions carrying one are recorded. That drops 278 of 1,160, and the shape of the loss is the tell: 2017 has 80 decisions and 23 finality orders.

What kind of order it is. Just under half this docket is section 12(j) revocations of delinquent filers, and another tenth follow-on bars. Neither is recorded, for the reasons set out under administrative proceedings above.

Whether the matter is already here. A proceeding can produce several initial decisions, and its order instituting proceedings may already be in the library. Both are keyed on the file number, so a matter already recorded is not recorded twice.

One further check exists because of an error this source made: the index names the presiding judge after the respondents, and the library briefly recorded Chief Administrative Law Judge Brenda P. Murray as a respondent in an accounting case. Judges are not parties. Their names are now removed before the respondents are parsed.

Licence: work of the United States government, not subject to copyright.

SEC trading suspensions

Suspensions ordered under Exchange Act section 12(k), taken in full from the Commission’s own index: 1,295 of them, from 1995 to the present.

These are stored separately from the case library, at data/trading-suspensions.json, and are not recorded as enforcement actions. A trading suspension is a protective measure that halts trading for up to ten business days when public information about an issuer is questionable. It alleges nothing against anyone, and recording it as an enforcement action would misrepresent it. It feeds the suspensions chart and nothing else.

The series has a shape worth knowing before you use it. Suspensions ran in the low tens per year through the mid-2000s, rose as the Commission began suspending batches of dormant shell companies from 2012, peaked at 113 in 2020, and then collapsed: 2 in 2022, 4 in 2023, 2 in 2024. That collapse is real and not a gap in our collection — we checked the index directly. Counts before about 2008 are much lower than later years, which reflects a genuinely smaller programme, though we cannot confirm the Commission’s online index is complete for its earliest years; treat those figures as a floor.

Licence: work of the United States government, not subject to copyright.

CFTC enforcement actions

The Commission’s enforcement action index and the press releases it links, from 2013 forward. CFTC headlines are unusually informative — they routinely name the respondent, the conduct and the amount — which improves extraction accuracy for this source relative to the others.

Licence: work of the United States government, not subject to copyright.

Department of Justice

Partial coverage, and the gap is structural. The Department’s filtered news listing returns HTTP 403 to automated clients. This project does not attempt to defeat bot protection, so historical DOJ backfill is not possible through that route.

What we do collect is the site-wide press-release RSS feed, which is a rolling window of recent items. That gives ongoing capture of new criminal market-manipulation matters and no historical depth.

In practice most criminal matters still reach this library, because they are announced alongside a parallel civil action and are recorded through the criminalParallel flag and the document links on that action’s page. But the DOJ action itself will usually not have its own record, and the criminal parallel chart understates the true criminal share for every technique as a result.

Licence: work of the United States government, not subject to copyright.

FCA final notices

Final notices from the UK Financial Conduct Authority: the document that concludes an enforcement matter and sets out what the firm or individual did and what was imposed for it.

We reach these through the FCA’s sitemap rather than its search interface, and that is a permissions decision rather than a technical one. The FCA’s robots.txt sets Disallow: /*?, which puts the whole paged search UI out of bounds, while explicitly allowing /sitemap.xml and /sitemap-main.xml. So we walk the sitemap and fetch nothing else carrying a query string.

Most of this docket is not market conduct. The FCA publishes far more final notices for consumer-credit permission cancellations and unpaid fees than for market abuse, so the large majority are read, found to describe no technique in our taxonomy, and skipped. That is the intended outcome: the classifier decides, not the file name.

Two sections of every notice are removed before classification, because both describe conduct that is not the conduct being sanctioned. The annex reproduces the relevant statutes verbatim, so a notice that annexes the Market Abuse Regulation quotes the language of manipulation whatever the firm actually did. And the penalty calculation lists the respondent’s previous fines — the 2025 Barclays notice is an anti-money-laundering case whose history block cites LIBOR, London Gold Fixing and foreign exchange in a single paragraph, none of them the matter at hand. Reading either as findings produces confident, wrong records.

Licence: published by the FCA under its own terms. We link rather than reproduce.

ASIC media releases

Media releases from the Australian Securities and Investments Commission.

Licence: published by ASIC under its own terms. We link rather than reproduce.

Ontario Securities Commission (Capital Markets Tribunal)

Ontario enforcement, taken from the Capital Markets Tribunal, which has decided the OSC’s enforcement proceedings since 2022 and publishes one page per proceeding with its documents attached. That page is the unit of a case here. Records name the OSC as the agency, because the OSC prosecutes, and the Tribunal as the forum, the same way a US record names a district court.

We use the Tribunal rather than the OSC’s own site deliberately. osc.ca publishes over 15,000 pages under orders, rulings and decisions, overwhelmingly exemptive relief and routine rulings that no URL distinguishes from enforcement, and its robots.txt disallows the paged enforcement listings. The Tribunal’s proceedings are the enforcement docket, and nothing there is restricted.

The OSC’s open-data inventory does publish enforcement datasets, and we looked at those first. They record respondents, dates and sanction types but give no account of the conduct, so nothing in them can be classified by technique.

A proceeding that ended in a notice of withdrawal is skipped. Its statement of allegations describes conduct nobody was found to have engaged in, and the regulator itself abandoned the claim.

Licence: published by the Capital Markets Tribunal under its own terms. We link rather than reproduce.

A note on currencies

The FCA states penalties in sterling, ASIC in Australian dollars and the Capital Markets Tribunal in Canadian dollars, and this site does not silently convert currencies. Converting needs an exchange rate on a date the release does not give, and a wrong rate is worse than no figure at all.

These penalties are therefore recorded as published, in their own currency, and shown on the case page as penalty as published. They are deliberately excluded from every US dollar total, median and chart on this site. A statistic described as covering penalties means US dollar penalties, and the non-US records are not silently folded into it at an invented rate.

Sources we cannot reach

Recorded here so the gap is visible rather than silent.

What the FINRA gap costs this site

This is the omission that most distorts the library, and it is worth being precise about how.

FINRA is the primary enforcer of order-book manipulation in US equities. Layering, spoofing, marking the close and quote stuffing are brought there far more often than at the SEC. Because those actions are missing, this site’s pages for those techniques rest on CFTC futures matters and a thinner set of SEC cases, and are not representative of where such cases are usually brought or of the size of a typical penalty. Any count, chart or median here describing US equity order-book manipulation should be read as covering the SEC and CFTC only.

We previously recorded this gap as bot protection. That was wrong: the pages answer automated requests normally, and the barrier is the instruction in robots.txt, which we follow whatever the server would let us do. Every request the pipeline makes is now checked against the publisher’s robots.txt before it is sent, so this rule is enforced by the code rather than by our memory of it.

Arbitration awards are a different thing, and would not fill that gap

FINRA’s arbitration awards are the obvious substitute for the disciplinary docket, and they are closed by the same robots.txt — one directive for the awards database, another for the directory of award documents. But the more important reason we would not use them is that they are not enforcement, and an award is not a finding of wrongdoing.

FINRA arbitration is private dispute resolution between a customer and a firm or broker. Under FINRA Rule 12904, an award must state the parties, a summary of the issues, the damages requested and awarded, the fees, the arbitrators and the dates. It need not state what anyone did. Rule 12904(f) says the award may contain a rationale; an “explained decision” under 12904(g) is produced only when both sides jointly ask for one before the prehearing exchange, and even then gives only “the general reason(s) for the arbitrators’ decision”, without legal authorities or damage calculations.

So a typical award records that a panel ordered someone to pay, and nothing about why. Filing that here as a manipulation case would assert a violation that no adjudicator found and no document states — against a named broker. The gap in this library is enforcement actions for order-book manipulation, and a private damages award is not one, however closely the underlying complaint might have resembled one.

Why the CSA Disciplined List is not a shortcut

It is the obvious thing to reach for — one national register of everyone disciplined by a Canadian securities regulator or self-regulatory organisation, which looks like it would fill the Canadian gap in a single pass. It would not, for reasons worth setting out because they apply to several sources we have turned down.

Three things stand in the way. The site answers automated clients with a JavaScript challenge, and we do not solve those. There are no documents to read even if it answered: across the 6,331 URLs in its sitemaps there are exactly three disciplined-list pages, none of them an entry, because the register is a search application rather than a set of published decisions. And most fundamentally, a register records who was sanctioned, in which jurisdiction, when, and for how much — not what they did. Every case on this site is tagged by technique, and a technique cannot be read off a sanction amount.

That last point is why we also declined the OSC’s open-data enforcement datasets, which are reachable and unrestricted, and why we did not rebuild the FINRA docket from BrokerCheck. A list of sanctioned people is a different kind of thing from a library of what market manipulation looks like, and merging one into the other would produce records that assert nothing and cite nothing.

The register remains the right tool for the question it answers. If you want to know whether a particular person has been disciplined in Canada, use it directly: CSA Disciplined List.

What the CIRO gap costs this site

CIRO is Canada’s national self-regulatory organisation for investment dealers and marketplaces, and it enforces the Universal Market Integrity Rules, which govern manipulative and deceptive trading on Canadian marketplaces. Ontario’s provincial enforcement is now collected from the Capital Markets Tribunal, but that does not close this gap: the provincial commissions prosecute securities-law offences such as fraud, illegal distributions and insider trading, while order-book conduct on Canadian marketplaces — layering, spoofing, wash trading between accounts — is CIRO’s jurisdiction under UMIR. So this library still has no Canadian order-book enforcement, and its coverage of Canada is Ontario-weighted besides, since British Columbia, Alberta and Québec are not collected either. Nothing here supports a comparison between Canadian provinces, or between Canada and the United States, on how often such cases are brought.

Our description of this gap was also wrong until recently, in the opposite direction to FINRA’s: we said CIRO’s robots.txt permitted the pages. It does not. Alongside the challenge, it disallows the path the enforcement bulletins are published under — a line further down the file than we had read. The pipeline now checks the recorded gaps as well as the collected sources, so a claim like that one fails a check rather than sitting on this page indefinitely.

Whistleblower award orders are open, and still the wrong shape

Unlike everything else on this page, nothing blocks us here. The SEC’s index of final whistleblower award orders carries no robots restriction and no bot challenge, and goes back to 2012 — by the standard the rest of this page uses, it is the easiest source available. We read three sample orders directly before deciding not to collect it, because the barrier turned out to be the document itself.

Each order redacts the claimant’s identity, which the statute requires (15 U.S.C. § 78u-6(h)(2)) and which we would expect and respect regardless. But each order also redacts the “Covered Action” it pays out on — the caption, the release number, the respondent are all replaced with the word “Redacted”. What is left is a claim number, a payout, and the Claims Review Staff’s reasoning about why a tip was valuable. There is no respondent to record, no technique to tag, and no case in this library for the award to attach to, since the one thing that would let us connect an award to the action it rode on is the one thing withheld.

This is the same test applied above to FINRA arbitration awards and the CSA Disciplined List: an open source can still be the wrong shape for this library. An arbitration award states an outcome without saying what anyone did. A disciplined list states who was sanctioned without saying what for. A whistleblower award order states that a claimant’s tip was valuable in some action, without saying who or what. Recording it here would produce a dollar figure with no case attached to it, which is not an enforcement record in the sense the rest of this site uses.

Matters from these sources are added by hand where they are significant enough to warrant it. If you maintain any of these services and would prefer we collect properly rather than not at all, get in touch — we would rather be rate-limited, throttled or required to identify ourselves than excluded.

What one record is, and what its date means

A record is one regulator document: one litigation release, one administrative order, one press release. A single matter that produces a filing release, a settlement release and a later judgment therefore appears as several records, and the same conduct can be tagged in more than one of them. The library does not merge them, because each carries its own date, money figures and link to its own primary document. Counts of records are not counts of schemes.

dateFiled is the date of the document the record links to. For CFTC and most SEC records that is the date the regulator published the release, which can be a day after the order or complaint it announces. Where we have read the underlying order and it carries an earlier date, the record uses that date. The two can therefore differ by a day between records from the same regulator.

How collection works

• Permission. Every URL is checked against the host’s robots.txt before it is requested, using the standard’s own precedence rules, and a disallowed URL is never fetched or read from cache.
• Politeness. Requests are serialised per host with a delay between them, well below the rate the SEC’s own guidance permits. The user agent identifies the project and carries a contact URL.
• Caching. Every document fetched is written to a local cache keyed by URL and is never re-requested. Re-running the pipeline costs the source nothing.
• Resumability. State is tracked per source, so an interrupted run resumes rather than restarts.
• Logging. Every decision — written, skipped, why — is appended to data/ingest-log.ndjson.

What we do not do with source text

We do not copy release text onto this site.

US federal releases are uncopyrighted, so this is not a licensing constraint. It is an editorial one. Structured fields are extracted mechanically; the summary and narrative on each case page are then composed from those fields, in our own sentences. Quotation is limited to short phrases where exact wording is legally load-bearing.

The constraint is enforced in the pipeline rather than left to discipline: the validation step fails the build if any case narrative shares a long run of consecutive words with a cached source document.

Licence for what is ours

Our writing, our technique taxonomy, our classification, and our compiled datasets are published under CC BY 4.0. Reuse them freely, including commercially, with attribution and a link.

The underlying government releases are not ours to license and need no licence from us.

Citing this data

Attribute to this site and link the specific page or export. When quoting a figure, note that it measures enforcement, not conduct — this data records manipulation that was detected, charged and announced, which is a different population from manipulation that occurred, in ways that cannot be estimated from the data itself.

Last updated September 20, 2026.