Civil cases with a criminal parallel

816 of 2,428 records (34%) carry the corrected criminal-parallel flag: 32% for the SEC, 32% for the CFTC and 81% for ASIC. The library holds no DOJ records, so no civil-to-DOJ gap can be timed. Eleven matters, each read by hand, span two regulators; eight were released the same day.

Published 2026-10-04 · 10 min read

A flag called “criminal parallel” sits on 816 of the library’s 2,428 records, 34%. It is the only link the library keeps between a civil or regulatory case and a criminal one. This post reports the flag as corrected on 4 October 2026, asks how often it appears by regulator, technique and outcome, what sentences sit beside it, what the flag still cannot show, and how many matters show up under more than one regulator. The last of those cannot be answered the way it was posed, and the reason is itself a finding.

The finding first

After correction, the flag marks 816 of 2,428 records (34%): 32% of SEC records, 32% of CFTC records and 81% of ASIC records. It stood at 836 (34%) before the corrections. A reading of the SEC and CFTC release texts found 113 flags that rested only on a prosecutor’s name or on criminal vocabulary with no case behind it, and 15 records that described a criminal case but were unflagged; all 128 were corrected. Eleven records that carry a recorded sentence but no flag were then flagged, which made 805. Later resolution rounds, which researched more of the “filed” records, added a net 11 more, giving 816.

Four other results follow.

• The rate varies much more by route than by technique. SEC litigation releases carry the flag 43% of the time and SEC administrative orders 7%. ASIC’s 81% is a different thing, because many ASIC records announce the criminal case itself.
• In 86% of flagged SEC and CFTC records the release itself states a criminal case. Of 696 flagged records whose cached release was read, 601 do. The other 95 rest on later documents the record cites, such as a judgment offset by criminal restitution; the flag cannot show that without those documents.
• Sentences are recorded on 285 records, over a third of those flagged, and every one is flagged. The median is 36 months. Ponzi-scheme records have a median of 84 months and insider-trading records 18.
• The library holds no DOJ records. There is nothing to time a civil-to-DOJ gap against. What it does hold is 11 matters, each read by hand, with records from two regulators. The library’s own grouping rules find seven of them. Eight were released the same day.

What the flag means in this library

The field is criminalParallel, a boolean that defaults to false (src/content.config.ts). Nothing in the schema or in DECISIONS.md defines it beyond its name. The case-page template writes “The release references a parallel criminal proceeding” when it is true, so the working definition is that a record refers to a criminal proceeding connected to the matter.

It was first set by keyword rules at ingest. For SEC and CFTC records the rule fired when the release text contained any of eight phrases. Five are statements about a case: indictment, pleaded guilty, criminal charges, criminal complaint, parallel criminal. Three are only the names of a prosecuting body: “Department of Justice”, “U.S. Attorney” and “United States Attorney”. ASIC and FCA records use different, shorter rules (convicted, sentenced, pleaded guilty, criminal charges and similar); OSC records use the same rule as the SEC’s. The DOJ ingest script would set it true on every record it wrote, but it has written none.

The 2 to 4 October 2026 audits then read the records against their documents. The results are in the corrected flag: the three prosecutor names no longer count by themselves, and a later judgment or release that reports a criminal case now counts even where the first release did not. The flag still does not say how the criminal case ended, who was charged, or whether it came before or after the civil one.

How it was measured

A script, scripts/analysis/civil-and-criminal-links.mjs, reads the case files and prints every number below. It loads the library’s grouping code (src/lib/matters.ts) for the matter counts. A second section reads the cached copy of each SEC and CFTC release, so it runs only where that cache exists; 26 of the 2,198 SEC and CFTC records have no readable cached copy and are left out of the release-text checks. Rates are shares of records, a record counts once under each technique tag it carries, and counts are given beside every percentage.

Figures are as of 4 October 2026, after the status and flag corrections. Since the first version of this post, 543 of the 733 records that had status “filed” were researched and updated; 190 are still “filed”, 1,190 settled, 987 judgment, 29 unknown, 25 dismissed and 7 appealed. The flag corrections are the ones described above, and the sentence count rose from 168 to 285 because the researched records add sentences stated in later documents. Cross-regulator matters were recomputed: the library’s grouping now links the Agridime matter.

The hand reading was done by a Claude AI agent, not by a lawyer. It covered the flagged records whose release held criminal wording but no explicit phrase, every flagged record that named only a prosecutor, a seeded random sample of 30 from the flagged records with an explicit phrase, every unflagged release whose text held conviction or charge wording, the ten longest sentences and a random 16 more, and every record of each cross-regulator matter together with the releases behind them. The other records were classified by the script’s word search and not read one by one. Headline counts were recomputed a second way, by counting flagged and sentenced files directly, and agreed.

What it shows

By regulator and route

The overall rate is 34% (816 of 2,428). By regulator it is 32% for the SEC (590 of 1,860), 32% for the CFTC (108 of 338), 27% for the FCA (13 of 48), 19% for the OSC (13 of 69) and 81% for ASIC (92 of 113). The library has no DOJ, FINRA or CIRO records. Because the three SEC routes behave so differently, the SEC figure hides most of the story: litigation releases 530 of 1,214 (44%), administrative law judge decisions 20 of 58 (34%), administrative orders 37 of 558 (7%). A litigation release reports a court action, and an administrative order is typically a settled order issued by the Commission itself. The data show the difference in rate; they do not show why it arises.

ASIC is not comparable. In an evenly spaced read of 12 flagged ASIC titles, nine announce a charge, a guilty plea, a conviction or a sentence, and the other three are a ban and two records titled “ASIC v.” Where ASIC brings a prosecution the record is the criminal matter, not a civil record with a criminal shadow. That read was made before the corrections and ASIC records changed little since. The FCA and OSC counts are small (13 flagged each), and the five FCA final notices in the matters below contain no criminal wording at all.

By technique

The chart uses SEC litigation releases only, because mixing in administrative orders and other regulators makes a technique’s rate depend on its route. Ponzi schemes (142 of 289), insider trading (181 of 391), pump and dump (48 of 104) and boiler rooms (23 of 52) sit between 44% and 49%; unregistered distributions (25 of 66) and undisclosed control blocks (19 of 50) at 38%; paid stock promotion (10 of 35) lowest. The gaps between the top four are a few records and are not a ranking. Among smaller groups, price manipulation is 10 of 21 and matched orders 14 of 25 (both under 30, so read cautiously); spoofing is 1 of 8 and Rule 105 offering shorts 0 of 3, both below 10 and too small to compare.

Across all routes, Rule 105 offering shorts (0 of 85) and naked short selling (0 of 35) never carry the flag. Seventy-seven of the 85 Rule 105 records and 29 of the 35 naked-short records are SEC administrative orders, which in general rarely carry the flag (7%), so the zero here is as much about route as about conduct. Neither the library nor this analysis can say whether a criminal case could have been brought for such conduct.

By outcome

By status the flag appears on 434 of 987 judgment records (44%), 71 of 190 “filed” records (37%), 14 of 29 “unknown” (48%) and 288 of 1,190 settled records (24%). Dismissed (9 of 25) is small, and appealed (0 of 7) is too small to compare. That settled figure is mostly route. Of the 1,190 settled records, 522 are SEC administrative orders, and those carry the flag 5% of the time (28 of 522). Among SEC litigation releases alone the gap disappears: judgment 242 of 547 (44%), settled 217 of 486 (45%) and “filed” 60 of 157 (38%). Status “filed” means the outcome was not found or the case is still running, so “filed” is not a stage.

The flag is also unstable within one matter. Of the 301 matters the library groups into more than one record, in 137 every record is flagged, in 124 none is, and in 40 the records disagree. Records in multi-record matters carry the flag more often (368 of 678, 54%) than single-record matters (448 of 1,750, 26%). Longer matters give a release more chances to mention a criminal case; that is a reading, and the data cannot test it.

What the corrected flag rests on

Of the 696 flagged SEC and CFTC records whose release text was read:

The 95 in the third row are mostly later judgments that treat civil money as satisfied by criminal restitution or forfeiture, or later releases that report a plea or sentence. For example the SEC’s case against Manuel Romero was flagged on a prosecutor’s thanks in the first release and is no longer. We did not re-read those later documents here; the row shows only that the record’s own text refers to a criminal step.

Across all 816 flagged records, 779 have a criminal word in the record’s own summary or narrative. Of the other 37, 34 are SEC or CFTC records whose release states a case. Three have neither, and we could not check them: two FCA final notices (Danziger, Shah) and one SEC order whose release is not cached (Hannelius).

What the flag still cannot show

Missed cases. Of 1,476 unflagged SEC and CFTC records whose release was read, 50 hold conviction or charge wording. Reading them: eight are follow-on orders that rest on the respondent’s own earlier conviction or plea; three name a criminal case about the same scheme or issuer (Kaplan, Voutsas, Bandimere) and are arguably flaggable; the other 39 mention a conviction as background, a founder’s past record, or case law. Another 142 unflagged records have a criminal word somewhere in their summary or narrative; 40 of them still carry the case-page template sentence about a parallel criminal proceeding, which no longer matches the flag. We did not read those one by one. The earlier hand reading found the real misses, such as Dawn Bennett and Rudy Avila, and they are now flagged.

Same conduct or another person’s. The flag says a criminal case is connected to the matter. It does not say the criminal case is against the civil respondent: a co-defendant’s or an earlier scheme’s case counts the same.

Silent releases. The flag comes from what a release or a later document said. A civil release that is silent about a criminal case is not evidence there was none.

On this measure the SEC and CFTC rate is 32.0% (696 of 2,172 records whose release text was read), and 27.7% (601 of 2,172) where the first release itself states the case.

Sentences

A custodial sentence in months is recorded on 285 records, all flagged (a record with a sentence was treated as having a criminal case), over a third of the 816. A sentence is recorded where a release or a later document states one, so it covers only what a document says. By agency the share of flagged records with a sentence is 65% for ASIC (60 of 92), 38% for the CFTC (41 of 108) and 30% for the SEC (179 of 590). The 285 skew toward cases at the end of the process, because a document usually mentions a sentence only after one is imposed.

The median is 36 months (middle half 18 to 84; mean 63). Forty-nine records are 12 months or less, 47 over 120 months. The two longest are Keith Simmons at 480 months (40 years) and Gary McDuff at 300; without both, the median is still 36 and the mean falls to 61. By tag with at least 10 records, Ponzi schemes have 104 (median 84; also 84 without Simmons), insider trading 86 (median 18), price manipulation 21 (median 21) and pump and dump 18 (median 45). Within insider trading the SEC’s 49 records have a median of 14 months and ASIC’s 36 have 18. Records of one matter often repeat the same sentence: the three Merrill records each carry 264 months and two Adams records 234. Counted by matter there are 204 sentences, median 36.5 months (middle half 18 to 84), and 67 of those matters have more than one sentenced record. We checked the ten longest values and 16 random ones against the cached releases when 168 records carried a sentence. All matched a stated sentence for a person named in the record, except that multi-defendant records carry one defendant’s figure, and the Merrill record carries the longest of three (22 years) while its release headline concerns a different defendant’s two-year sentence. The records added since were not re-checked here. A total of 111 of the 285 records have more than one defendant, so a reader comparing sentences by case should not read a sentence as the case’s.

Matters that span regulators

What the library can and cannot count

The library’s grouping code links records of one scheme when they share a lead defendant name of at least two words, overlapping technique tags and a date window, share a court docket, or cite each other’s release number or address. On 4 October 2026 it puts the 2,428 records into 2,051 matters. Of these, 301 hold two or more records (678 records altogether), and seven span more than one regulator: five CFTC and FCA pairs and two CFTC and SEC matters.

No matter has a DOJ record, because the library has none. The DOJ ingest reads only a rolling feed of recent press releases, and the sources page says historical DOJ items are reached only through the civil action’s page. No record names a justice.gov address in its release link, document list or timeline. So the measure asked for, the days between a civil record and a DOJ record for the same matter, has zero pairs. The nearest substitute is what each civil release says about the criminal case, below.

Eleven matters, read by hand

The grouping rules prefer a missed link to a false one, so we looked for misses. We searched for any defendant name shared across regulators within three years, then for CFTC releases that say a related SEC or FCA action exists, and read both records and both releases for every candidate. Eleven are the same matter: in ten, a CFTC release itself names the other regulator’s action, and for the eleventh (Barclays, FX) the two releases cover the same bank, the same conduct and the same day. The 2013 Royal Bank of Scotland notice was issued by the FSA, the FCA’s predecessor, and the library files it under FCA.

The four unlinked matters have one of two causes: a lead name of one word, which the rule ignores (Citibank in the FX order), or a different lead defendant in each regulator’s record (Coscia against his firm, JPMorgan’s holding company against its securities arm, RP Martin against Martin Brokers). The Goliath, Rabobank and JPMorgan matters are the ones we would point a reader to first, because the later release names the parallel action. Eleven is a floor, not a count: our search was by name and by CFTC cross-reference, and an SEC release that names a CFTC action in a way the search missed would not appear.

Two cautions on the table. The gap is between the library’s dateFiled values, which are mostly release dates but not consistently: the SEC’s Agridime release is dated 18 December 2023 though the complaint was filed on 11 December, so the true gap is 155 days there, and the four Agridime records run over 652 days. And eight of the eleven were released the same day because they were coordinated settlements; the dates show announcement timing, not which regulator moved first.

The corrected flag disagrees across the records of five of the eleven matters: Rabobank, Deutsche Bank, Royal Bank of Scotland, Goliath and JPMorgan. In each the CFTC record is flagged and the other regulator’s is not; in the other six no record is flagged. The five FCA notices we searched contain no criminal wording. The corrections matter here: the Rabobank, Deutsche Bank and Royal Bank of Scotland orders describe Justice Department deferred prosecution or plea arrangements and stay flagged, while the ICAP and Barclays FX orders only thank the Department and are now unflagged.

What the records say about order

The library cannot time civil against DOJ records, but a civil release can state when the criminal step happened. Where it does, the order is whatever the dates show:

• Goliath Ventures. The CFTC release says that in June 2026 Christopher Delgado pleaded guilty in a case brought by the US Attorney’s Office for the Middle District of Florida, and that the CFTC and the SEC each filed civil actions on 11 August 2026. The plea comes some six to ten weeks before both. The SEC release does not mention the criminal case, so the SEC record is not flagged.
• Ward Onsa. The SEC administrative law judge’s decision of 17 June 2013 and the CFTC’s judgment release of 31 March 2014 (287 days later) both refer to the same criminal case in the Eastern District of New York, and both records are flagged. The CFTC’s order directs that payments first satisfy Onsa’s criminal restitution. The two civil records concern different proceedings against one person, so we count them as linked through a criminal case, not as one civil matter. Both records now carry a 78-month sentence; we have not re-checked that figure against the decision here.
• JPMorgan, 29 September 2020. The CFTC release reports that the same day the Justice Department announced a deferred prosecution agreement with the holding company and the SEC announced its own order. No date separates them.
• A sample of 30. From the flagged SEC and CFTC records whose text held an explicit criminal phrase (528 when drawn, 533 now), we drew a seeded random sample of 30 and read each. All 30 are still flagged and still hold the phrase. In 17 the release reports the criminal step as already done, such as a plea, a sentence, a forfeiture or an indictment with an earlier date. In 6 it was announced the same day. In 6 a parallel case is mentioned with no date or is still under way, and in 1 a criminal step is still to come. In this sample, then, the criminal step is more often earlier than later. Part of that is built in: judgment and settlement releases, which are later stages, naturally report earlier criminal steps, and a release reports only what is known when it is written.

None of this shows that a criminal case caused a civil one or the reverse. The releases show the order of stated events and nothing about why.

What this does not show

Selection. These are cases regulators chose to bring and publish. A matter that was never brought, or that a regulator did not announce, is not here. The rates describe the library’s records, not enforcement at large or the share of offenders who face criminal charges.

A missing flag may be an unreported parallel. A civil release that is silent about a criminal case is not evidence there was none; of 12 CFTC releases that name a parallel SEC action, 9 have no SEC record in the library, so the cross-regulator coverage is thin in the other direction too.

What the flag still cannot show. Of the 696 flagged SEC and CFTC records read, 95 rest on later documents we did not re-read, and three flagged records have no criminal wording in the record or the release. Three unflagged records name a criminal case about the same scheme, 50 unflagged releases hold conviction wording that was read, and 142 unflagged records have criminal words in their own text that were not read one by one. The 26 uncached releases and every ASIC, FCA and OSC record were not release-checked, and ASIC’s rule measures something else. The flag does not say who was charged or how the case ended.

Sentences are partial. They exist only where a document states one, cover over a third of flagged records, repeat across records of one matter, and give one defendant’s figure on 111 multi-defendant records. Records added since our spot check were not re-verified. The set is small by technique, so the medians for groups under 30 are indicative.

The grouping and the dates. The grouping is the library’s and misses single-word and mismatched lead names; our eleven is a floor. dateFiled is not defined consistently between release date and filing date. No DOJ record exists, so no civil-to-criminal gap in days was measured, and the order in the matters above comes from the text of civil releases.

Checked by AI agents, not lawyers. The readings behind the classifications and the flag corrections were made by Claude AI agents with the lead session re-reading disputed cases. No lawyer reviewed them. The case pages say, in the library’s own words, that the grouping and the flag are the library’s, not the regulators’. If you find one that is wrong, use the error link on that record’s page or write to [email protected] with the slug and a link to the document; it will be logged on the corrections page.

Techniques referenced

• Insider trading — Insider trading is dealing in securities on material non-public information in breach of a duty of trust or confidence, which exploits a true signal others cannot see rather than injecting a false one into the price.
• Ponzi schemes — A Ponzi scheme pays returns to earlier investors from later investors' money rather than from genuine profits, which is investment fraud with no market involvement and no price to manipulate.
• Benchmark submission rigging — Benchmark submission rigging is skewing the figures a panel member reports into a reference rate calculation, so that the published benchmark moves in favour of positions that settle against it.
• Spoofing — Spoofing is placing orders a trader intends to cancel before execution, to create a false impression of supply or demand and move the price toward a smaller genuine order on the other side.

Cases referenced

Reviewed October 4, 2026. Spotted an error? Tell us.