More than half of this library is not market manipulation
Of 2,637 records, 1,256 (48%) carry only tags from the "related but distinct" family: insider trading and Ponzi schemes account for most of them. Just 982 carry a manipulation-family tag, and 399 carry no technique tag at all. The split differs sharply by regulator, which means the record describes who publishes and what they publish at least as much as what happens in markets.
This site’s front page says how many enforcement actions it holds. That number is accurate, and it is easy to read as “the number of market manipulation cases”. It is not, and the difference is large enough to be worth putting on the table.
The library assigns every record one or more technique tags, and every technique belongs to one of seven families. Six are manipulation families defined by what the manipulator controls. The seventh, related but distinct, holds conduct that is often called manipulation and is not: insider trading, Ponzi schemes, churning, front running, naked short selling and hacking to trade. It is in the library on purpose, so that a reader looking for the difference can see both sides. It also means the library’s total counts both.
All figures below are as of 2026-10-05 and are computed from the case files in cases.json.
The split
Of 2,637 records, 1,256 carry only tags from the related-but-distinct family. That is 48%. Another 982 carry at least one tag from a manipulation family, and 399 carry no technique tag at all: after a record is read against its primary document, a tag the document does not support is removed rather than replaced by one that does not fit. A record can carry several tags — 248 do — so a family’s count is the number of records that touch it, not a partition.
Two techniques account for most of the first group. There are 750 insider-trading records and 474 Ponzi-scheme records. Everything else in the library’s top ten is a real manipulation technique, and the largest of them has 157.
None of this means insider trading and Ponzi schemes are unimportant. They are the most frequently announced kinds of securities enforcement, and a reader who wants to see how they differ from manipulation should be able to. It does mean that a statement like “the library holds 2,637 market manipulation cases” would be wrong by more than half.
It depends who you ask
The split is not the same at every regulator.
Among SEC records, 36% carry a manipulation-family tag. Among CFTC records it is 49%. At the UK’s Financial Conduct Authority it is 65%; at the Ontario Securities Commission it is 14%; among the 209 records of France’s AMF it is 29%. AMF France’s records are mostly insider dealing decisions (119 of 209). The SEC’s enforcement docket is dominated by insider trading and Ponzi schemes, and the CFTC’s by order-book conduct in futures. The FCA’s records are the only ones where the benchmark family is a large share, which reflects the LIBOR and foreign-exchange settlements of the 2010s.
These are differences in what each regulator publishes, and they should not be read as differences in how much manipulation occurs in each market. Three further limits apply to any comparison:
Coverage is uneven by design. The SEC and CFTC are the deepest sources. FINRA, where much US order-book enforcement is brought, is not collected because its robots.txt asks crawlers not to read its disciplinary database. DOJ press releases are a rolling window, not an archive. In Canada only Ontario is collected, and CIRO, the national order-book regulator, is not. The sources page lists each gap and what it costs.
Sources start in different years. CFTC actions and SEC administrative orders are collected from 2013, SEC court filings from 2015, and AMF France’s Commission des sanctions decisions from 2004. AMF France’s records are decisions as published in French and read by AI agents against the decision; older ones name their defendants only as anonymised letters, and its penalties are in euros. A 2013 or 2014 record therefore comes from a narrower set of sources than a 2016 record does.
A count of records is not a count of schemes. One scheme can produce a filing release, a settlement release and several follow-on orders, each of which is a record.
Records versus matters
The library groups records that appear to concern one matter, at build time and only on strong evidence: the same lead defendant with an overlapping technique tag and filing dates within three years (one year for a firm unless both records name the same court), a shared court docket number, or a release that cites another record’s release number. As of 2026-10-05 the 2,637 records fall into 2,260 matters. Of those, 301 matters hold two or more records (678 records in all), and only seven span more than one agency. The rules err towards missing a link, so the true number of matters is probably somewhat lower. The grouping is ours, not the regulator’s, and each case page that has siblings says so.
What the timeline can and cannot say
The series peaks in 2018 and is lower after 2020, with a sharp fall in 2025 and 2026. The honest reading is that this is a record of what was announced and collected, and it does not distinguish between fewer cases, slower announcement, and coverage that has not yet caught up with recent filings. The 2015 figure includes a batch of settled orders under Rule 105, a strict liability rule discussed in its own post. Sweeps like that can move a year on their own.
Thirteen techniques have no records
The library defines 59 techniques. Thirteen of them have no record at all: box squeezes, cornering, custodianship shell hijacking, delivery squeezes, dilution death spirals, ETF and NAV abuse, market maker loan arrangements, marking the open, options expiry pinning, painting the tape, perpetual-market spoofing, quote stuffing and sandwich attacks. Three others were on the list of fourteen until 2026-09-20 and now have records: EDGAR filing fraud with ten, social media ramps with nine and short and distort with three. The thirteen are not the same as the fourteen on 2026-09-20: analyst manipulation now has five records (three after source checks retagged them, two from AMF France), and marking the open, which had a record then, has none. Momentum ignition, empty on 2026-10-03, now has one record, a French decision. Custodianship shell hijacking, dilution death spirals and options expiry pinning had records on 2026-10-02; a second round of reading records against their primary documents removed tags the documents did not support, and left these three with none.
Those three were empty because of the tagging, not because the library lacked the documents. The cache held SEC releases about false tender offers filed on EDGAR, promoters selling into their own Twitter, Discord and Stocktwits followers, and a hedge fund adviser found liable by a jury for some false statements about a company he was short. The keyword rules that assign tags did not describe them, and now do. The three short and distort records are one matter, in its complaint, jury verdict and final judgment releases.
For the eleven that were empty before that second round, a search of the cached documents for the technique’s own terms and for descriptions of the conduct found nothing that fits, and the library cannot say why. The regulators it reads may not have brought the charge in the period covered. The charge may have been brought under another label, or by a body the library does not collect, such as FINRA or the exchanges. An empty technique page should be read as “this library holds no record”, never as “no such case exists”.
What to take from it
Three practical rules follow.
Read the family before the total. If you want manipulation cases, filter to the six manipulation families. The technique index is organised that way.
Compare within a regulator, not across. Differences between the SEC’s and CFTC’s mix reflect their mandates and what they publish.
Treat every gap as a caution. An empty facet, a missing year or a regulator that is absent says something about the library first and about the world second.
If a figure here does not match cases.json when you recompute it, tell us through the corrections process.