What sanctions the records show

Where a US dollar penalty is recorded (1,250 of 3,273 records) the median is $250,000, but ten records hold 69% of the penalty dollars and Stanford alone 35%. The median recorded sentence is 36 months (286 records). 30% of records show no money at all. SEBI's 462 rupee penalties (median INR 4.5 million) are kept out of every dollar figure.

Published 2026-10-04 · 17 min read

Where a civil penalty is recorded, the middle record carries about $250,000, and the middle half of penalties run from $75,000 to $1.2 million (1,250 records). That figure describes almost nobody who pays the large numbers. Ten records, under one percent of those with a penalty, hold 69% of every penalty dollar in the library, and the single largest, the Stanford Ponzi judgment, holds 35%. Half of the records with a penalty share 0.3% of the dollars between them. About seven in ten of the 3,273 records (2,281, 70%) carry some recorded money, 30% show none, and the sentence field is filled for only 286 records, with a median of 36 months. The seventh regulator, SEBI (India), adds 656 records whose penalties are in rupees; they appear in no dollar figure, and a separate section reports them in rupees.

This post sets out what the records say about sanctions, how little of it a single average can carry, and how much of any comparison between regulators, outcomes or respondents is explained by what each record happens to contain. It reports associations in this selection. It does not say that settling, cooperating, having a parallel criminal case or being a company causes any sanction to be smaller or larger.

The finding in five numbers

• Penalty, middle record: $250,000, with a middle half of $75,000 to $1.2 million (1,250 records with a US dollar penalty). Over a quarter (351, 28%) are $1 million or more and 29% (362) are under $100,000.
• Disgorgement, middle record: $354,000, middle half $66,000 to $1.9 million (1,052 records).
• Concentration: the largest 10 penalties are 69% of penalty dollars; with Stanford removed, the largest 10 of the remaining records are 56%, and the largest 10% are still 93%.
• Prison, middle record: 36 months, middle half 18 to 84 months (286 records with a sentence recorded, all of them among the 811 records with a criminal parallel).
• Nothing recorded: 992 records (30%) carry no penalty, disgorgement, interest or foreign-currency penalty. Of these 178 are records still recorded only as “filed”, 163 are dismissed (122 of them SEBI) and 194 are SEBI records (122 dismissals, 69 judgments, 3 settlements).
• SEBI, in rupees: a median penalty of INR 4.5 million (about 45 lakh), middle half INR 1.8 to 11.5 million, on 462 of its 656 records. Rupees are never added to dollars or converted.

How this was measured

Every figure comes from one script, scripts/analysis/sanctions.mjs, which reads the 3,273 case files and prints every number below. Five conventions matter.

As of 7 October 2026, after the status and flag corrections, the French and Indian regulators, and the removal of 21 non-matter records. This post was first written on 4 October, before 543 of the 733 records then recorded as “filed” were researched and updated; 180 of the 720 of those that remain are still “filed”. The 543 mostly gained a status (settled or judgment), a resolution date and money fields, so the count of records with a penalty rose from 1,008 to 1,250 and the count with no money fell from 1,157 to 818; the 194 empty SEBI records added since, less the 20 non-matter records deleted, bring it to 992. Statuses are now 1,284 settled, 1,615 judgment, 180 filed, 163 dismissed, 24 unknown and 7 appealed. On 5 October the library added a sixth regulator, the French AMF, with 209 records of Commission des sanctions decisions from 2004 to 2026: 167 labelled judgment, 24 settled (homologated settlements, whose amount is a payment to the Treasury under the agreement, not styled a fine) and 18 dismissed. 189 carry a penalty in euros and none in US dollars. On 7 October it added a seventh, SEBI (India), with 489 records, and on 9 October 167 more read under a wider scope, 656 in all, of orders dated 2022 to 2026: 466 labelled judgment (adjudication and final orders), 122 dismissed and 68 settled. 462 carry a penalty in rupees and none in dollars. Like the AMF records, they enter no dollar figure; they raise the record totals, lower the share of records with a US dollar penalty, and add dismissals to the status mix. Five of the SEBI records are batch records, each grouping a year of near-identical adjudication orders on reversal trades in illiquid BSE stock options (1,577 orders in all); they carry no penalty figure and count here as five records, not 1,577 orders. The 466 “judgment” SEBI records and 167 AMF records are regulator decisions, not court judgments. The criminal-parallel flag was corrected on 128 records (113 set to false because the release states no criminal case, 15 set to true), then set to true on the records that carry a sentence and on others, leaving 811 flagged (six of them AMF records and one a SEBI record). The Stanford record was moved back to “sentenced” on 7 October from the Fifth Circuit’s opinion, so its sentence of 1,320 months is in the sentence statistics (286 records, one more than before). Twenty-one records that were not enforcement matters (16 SEC subpoena-enforcement applications, 3 CFTC umbrella announcements, a CFTC notice and an Ontario motion) were deleted on 7 October. I checked 20 of the 21 against the pre-deletion files: none carried a penalty, disgorgement, interest or sentence, so no dollar figure here moved; they lowered the record count and the number of “filed” records. Other corrections that touch this post: five CFTC benchmark orders, the five-bank 2014 order, the 2020 JPMorgan order, Lloyds and RP Martin moved from judgment to settled; the five-bank penalty was corrected to $1.475 billion; five ASIC records that held Australian-dollar amounts in US dollar fields were corrected; the money on the duplicate Fujinaga 2017, Bank of Nova Scotia 8221-20, Cartu 2020, Aequitas 2016, Agridime 2023, Tyche, Digitex 2022, Lek, Yin, Capital Financial, Illarramendi, Marquis, Churchville (two) and Apostelos records was removed; and the Sparkster and Woodward amounts were fixed. What moved in this post since 4 October: the median penalty rose from $204,000 to $250,000, the top-10 concentration fell from 75% to 69%, the gap between settled and judgment penalties narrowed (to between 1.6 and 3.9 times), and the criminal-parallel comparison no longer reverses in the last row. Adding SEBI and removing the 21 records moved none of the dollar medians, because neither group carries a dollar amount. Outcomes recorded for the 543 often cover only some of the defendants, so “settled” and “judgment” labels are per record, not per respondent.

Medians and quartiles, not means. Sanction amounts are so skewed that a mean would be a statement about Stanford, not about the records. “Middle half” means the first and third quartiles; “IQR” is the same range.

Per record is per matter, not per person. A record stores the sum of what every respondent owes. 1,670 records (51%) name two or more respondents, so a $2 million penalty may be one person or eight. For SEBI the sums were made by the AI agents reading an order that lists penalties noticee by noticee, so a SEBI total is only as good as that addition. The same applies to the sentence field: where a release reports several sentences, the field holds one figure. In the three records I opened to check (Simmons, Merrill and Kita) it is the longest term in the release, not a total. Where it matters I repeat a comparison using only the 1,596 single-respondent records.

What counts as money. Penalties and disgorgement are in US dollars. Penalties from the Canadian, Australian, British, French and Indian regulators are held in their own currency (Canadian and Australian dollars, pounds, euros, rupees) and are deliberately not converted, so they are in no dollar figure here, and amounts in different currencies are never added together. That leaves the ASIC, FCA, OSC, AMF-FR and SEBI with one US dollar penalty among 1,095 records (a 2013 FCA record). “No money at all” means no US dollar penalty, disgorgement or interest and no native-currency penalty; 748 records (462 SEBI in rupees, 189 AMF-FR in euros, 35 ASIC, 30 FCA, 32 OSC) with only a foreign-currency penalty are not counted as empty. A section below reports each currency by itself.

The “status” is a label, not a trial. The 1,615 “judgment” records are court judgments of any kind, plus the 167 French and 466 Indian regulator decisions. In 276 of them the summary mentions a consent or settlement, and in 130 a default. So “settled” against “judgment” is not the same as “settled” against “litigated”, and the 180 records still recorded as “filed” have no tracked outcome. SEBI’s labels mean something different again: “settled” is a settlement under SEBI’s scheme in which the applicant neither admits nor denies, “judgment” is an adjudication or final order, and “dismissed” is common (122 of 656, 19%) because SEBI orders often close a proceeding with no penalty.

Cross-checks. The Stanford penalty of $5,905,000,000 is the sum of $5.9 billion on Robert Allen Stanford and $5 million on James Davis in the SEC release; the same release gives disgorgement and interest only as combined figures in the billions, deemed satisfied by forfeiture, so the record carries no disgorgement for it. The Glencore penalty ($865.6 million) and disgorgement ($320.7 million), the Battoo penalty ($147 million) and the Cartu penalty ($153.5 million) matched their CFTC releases when I first read them. The last two are recorded as default judgments, which is a reminder that a large number on paper is not a large number collected.

After the refresh I checked the other top-ten penalty records against their record narratives (not the original releases, which I did not re-open). The five-bank record’s $1.475 billion equals the five penalties its summary lists ($310 million twice, $290 million twice, $275 million). JPMorgan ($436.4 million) is the penalty part of a $920.2 million total that also includes restitution the library has no field for. International Enterprise ($338.8 million) is a January 2025 default judgment, joint and several, against eight defendants, and the Control-Finance record ($429 million) is a 2021 default judgment against one individual in a case about roughly $147 million of bitcoin. Large figures in this top ten are therefore partly default judgments, not collected money. These records match what their own summaries say; I did not find a mismatch. The Quan record’s $80.6 million of disgorgement is the Minnesota court’s award restated in the SEC proceeding, which its narrative now says.

Matters that appear twice. The refresh removed the money from 15 records that restated another record’s figures: Fujinaga 2017, Bank of Nova Scotia 8221-20, Cartu 2020, Aequitas 2016, Agridime 2023, and then Tyche, Digitex 2022, Lek, Yin, Capital Financial, Illarramendi, Marquis, Churchville (two) and Apostelos (Fujinaga 2015 and 2017; Bank of Nova Scotia one and two; Cartu 2020; Aequitas 2016 and 2020; Agridime 2025 and 2023). The ten latest of these were identified from the records’ narratives and the matching amounts, not by re-reading every release, and I rely on that. The script no longer finds any pair of records of one agency that share an identical amount of $10 million or more and a respondent name, so no sensitivity run is needed: the median penalty is $250,000, total penalties $16.7 billion and total disgorgement $5.2 billion with the duplicates already gone. I have not edited any record myself.

What it shows

A few records are the money

Of $16.7 billion in recorded penalties, ten records are $11.6 billion. Nine of the ten are CFTC records: the five-bank 2014 foreign-exchange benchmark orders ($1.475 billion, one record summing five respondents), the Glencore price-manipulation order, Deutsche Bank ($800 million), Rabobank and Société Générale ($475 million each), JPMorgan, the Control-Finance default judgment ($429 million), Barclays ($400 million) and the binary-options default judgment. The tenth is Stanford. Strip out Stanford and the total falls to $10.8 billion, 35% lower, while the median penalty stays at $250,000. Strip out the ten largest and the remaining 1,240 records total $5.1 billion with a median of $248,000 and a middle half of $75,000 to $1.1 million. In other words the median is stable, and everything that grows or shrinks with outliers is the total.

The same shape holds for disgorgement: the largest 10 of 1,052 records are 43% of $5.2 billion, and the largest 105 (10%) are 83%. Counting penalty, disgorgement and interest together over 1,533 records with any dollar figure, the largest 10 hold 56% of $22.5 billion.

Who the large amounts belong to depends on the outlier. Before removing Stanford the SEC’s penalties ($7.8 billion over 1,005 records) and the CFTC’s ($8.9 billion over 244) look similar. After removing Stanford, the SEC is $1.9 billion and the CFTC 82% of recorded penalty dollars. The CFTC dollars are themselves concentrated, with its largest ten records 68% of its total.

Typical amounts differ by regulator, technique and family

The CFTC median is more than eight times the SEC’s. That says something about what each agency publishes: the CFTC records here are dominated by firm-level manipulation and benchmark orders, the SEC records by individuals. It does not say the CFTC penalises the same conduct harder, because the conduct is not the same. The ASIC, FCA, OSC, AMF-FR and SEBI have too few US dollar penalties to compare; their sanctions are in other currencies. The next section reports them one currency at a time, and nothing in this table says whether a rupee, euro or pound penalty is larger or smaller than a dollar one.

By technique the median runs from $88,000 for paid stock promotion to $12.4 million for price manipulation, a 141-fold spread; insider trading, the largest group, is $89,000. The two ends are different populations: insider trading records are mostly one person’s trades; the price-manipulation records are firms. A technique tag is the library’s own classification and a record can carry several, so these groups overlap. Spoofing is a case in point: the 71 records with a spoofing penalty have a middle of $700,000 and a middle half from $275,000 to $2.0 million. Benchmark-submission rigging, not in the chart because only 18 records carry a penalty, has a median of $110 million, the highest of any technique; its upper quartile is $381 million. Counted by the broader families, the median penalty is $750,000 for the order-book family (175 records), $230,000 for information (141), $150,000 for issuer and structural (138) and $195,000 for related-but-distinct (554, mostly insider trading and Ponzi). The “benchmark and cross-market” family ($208,000, 108 records) mixes benchmark rigging with the 79 Rule 105 records, so its median describes neither. The 434 records with no technique tag (85 of them SEBI) have a median of $470,000, higher than most tagged groups, and 85 of the 195 with a penalty name only entities; the audit left untagged the records whose document did not support a technique, so this group is a mixed remainder.

Penalties in other currencies, one currency at a time

748 records carry a penalty only in the regulator’s own currency. They cannot be compared with the dollar figures above without an exchange rate and a date, which this library does not hold, and they cannot be compared with each other either, because the currencies differ and because each regulator’s remedies differ. Within each currency the records are:

The FCA’s middle half is wide because its 30 penalties include a few very large firm penalties. These columns say how large each regulator’s own figures are in its own money; they do not say that one regulator penalises more than another.

SEBI: orders, dismissals and rupee penalties

SEBI’s 656 records are Indian securities-law orders dated 2022 to 2026. Most are under its anti-fraud and insider-trading regulations (PFUTP, PIT) and the SEBI Act, and the scope includes issuer fraud, fund diversion and broker fraud on clients where the order makes PFUTP findings; 85 of them carry no technique tag. They are not all findings of manipulation: 122 (19%) are dismissals, in which SEBI closed the proceeding without a penalty, and the 466 recorded as judgment include adjudication orders that remand or limit a case. 462 records (70%) carry a penalty in rupees; none carries a US dollar amount, a prison sentence (SEBI imposes penalties and market restraints, it does not convict) or a disgorgement figure in the dollar fields. 194 records carry no money at all: all 122 dismissals, 69 judgments (five of them the batch records, which carry no figure) and 3 settlements.

• Size. The middle record’s penalty is INR 4.5 million (45 lakh), the middle half runs from INR 1.7 to 14.9 million, and the ninetieth percentile is INR 81.8 million. 143 records (31% of those with a penalty) are INR 10 million (1 crore) or more. The largest, INR 6.43 billion (643 crore), is a single settled record with 10 respondents and no technique tag; it is 23% of the INR 28.0 billion total, and the largest ten records hold 65%. The next, INR 6.24 billion on 12 respondents, is recorded as misleading issuer disclosure. The shape is the same as in dollars: a few records are most of the money.
• Settled against judgment runs the other way. The median penalty on the 65 settled SEBI records with a penalty is INR 10.5 million against INR 3.2 million on the 397 judgment records. A settlement under SEBI’s scheme is typically paid by several applicants together, and the record stores the sum, while a judgment record is often one noticee’s penalty. So the gap reflects how the records are built, not that SEBI settlements are costlier.
• Respondents. 367 of the 462 records with a rupee penalty name several noticees (median INR 5.5 million), and 94 name one (median INR 1.25 million). 492 of the 656 records name more than one respondent, and 330 name both entities and individuals.
• Restraints. 164 SEBI records (25%) list a bar-type remedy (a restraint from the securities market, a directorship or managerial ban, a debarment, a registration cancellation or a disgorgement or refund direction): 32% of judgment records and 22% of settled ones, and none of the dismissals.

By technique, the medians run from INR 3.0 million for price manipulation and matched orders to INR 28 million for social media ramps (13 records). Misleading issuer disclosure, 140 records with a penalty, has the widest spread, INR 2.2 to 78.6 million, because the INR 6.24 billion record sits in it. These tags are the library’s own classification of what each order charges, and a record can carry several. 85 SEBI records have no tag, including the largest penalty.

Caveats specific to SEBI. Every SEBI record was written from one AI reading of the order. The first 489 had a sampled second read of 60 on 8 October that agreed on every field for 56 (93%); the misses were penalty amounts, so rupee figures remain the weakest part. The 167 added on 9 October have no second read. The sums on multi-noticee orders were made by the reading agents from tables in the order. The five batch records (368, 860, 61, 136 and 152 adjudication orders for 2022 to 2026 respectively) summarise orders read from a sample, not all of them. Orders are dated by the order, so a SEBI “year” is a year of orders, often on conduct a decade earlier.

Where the money is missing

Of all 3,273 records, 992 (30%) show no money. Under a fifth of that (178, 18%) is the status. For those filed records the library records the filing and does not follow the outcome, so the empty field means “not tracked”, not “no sanction”. 163 more are dismissals, which by definition carry no sanction (122 of them SEBI’s). Of the 3,093 records that are not “filed”, 814 (26%) still have none. Among 1,615 judgments, 384 do (24%; 32% before the AMF and SEBI records, which mostly carry foreign-currency amounts), as do 245 of 1,284 settled records (19%). The three records I opened that had a prison sentence and a civil judgment (Gupta, Kita and McClatchey) show why: disgorgement ordered but deemed satisfied by criminal forfeiture, with no civil penalty on top. A sanction can exist and not be a number the record carries.

Settled and judgment

In every row the median penalty is smaller in records recorded as settled than in records recorded as judgment, by roughly 1.6 to 3.9 times (the CFTC gap is the narrowest). Across all records the settled median is $150,000 and the judgment median $657,000. Judgment records also carry more disgorgement (SEC median $730,000 against $189,000) and more bars (39% against 27% in the SEC records). Both groups still have wide overlaps: the middle half of SEC settled penalties ($50,000 to $425,000) and judgment penalties ($150,000 to $1.5 million) overlap across much of their range.

What this supports is an association in this selection: the records labelled judgment are, on average, the bigger matters. What it does not support is that settling lowers a penalty. A judgment record, as noted above, is often a consent or default judgment; the cases that are litigated to judgment are more likely to be the larger and the more contested ones before a penalty is set; 317 of the 630 SEC judgment records show no penalty, which cannot be told apart from “not recorded” (and many of those were updated only on 4 October, with outcomes for some defendants). The data cannot say what the same respondent would have paid on the other route.

Entities and individuals

Seven records list no respondent type and are not in the table. The record counts, no-money and bar columns include the SEBI (248 individuals only, 77 entities only, 330 both) and AMF records, while the penalty columns are US dollars and so cover only the SEC and CFTC (and one FCA record). Restricting to records with a single respondent, so the figure is per respondent, an entity’s median penalty is $400,000 (267 records, middle half $125,000 to $2.5 million) and an individual’s is $100,000 (449 records, $36,000 to $298,000). Entity records are far more likely to show money (81% of single-entity records against 64% of single-individual records) and almost never record a bar: 31 of 490 entity-only records. Individuals, not entities, receive prison sentences. None of the 490 entity-only records has a sentence recorded.

Prison

Only the criminal court sets a prison term, and the records carry it only when the regulator’s document mentions it. Of the 811 records with a parallel criminal matter, 286 (35%) carry a sentence, and 198 of those 286 are records recorded as judgment; the field is empty for most criminal parallels, which means “not stated”, not “no prison”. Across the 286 records with a sentence the median is 36 months, the middle half 18 to 84 months. A little over a third (107, 37%) are 24 months or less, 38% (109) are over five years, and 17% (48) are over ten. SEBI records carry no sentence: SEBI does not prosecute.

The FCA (3 records) and OSC (2) are too few to compare, and so is any other technique below ten. Stanford’s 1,320 months sets the maximum for the SEC and for Ponzi records but moves no median. The range runs from two months (Gupta, an insider-trading plea) to 1,320 months, 110 years, in the Stanford record, restored on 7 October from the Fifth Circuit’s opinion (the next longest is 480 months in Simmons, a forex Ponzi-style scheme with five respondents, where the release gives Simmons 40 years and a co-defendant 4.5). The gap between Ponzi and insider-trading records is large, but the records do not carry the criminal charge or the size of the fraud, so the data cannot say why. The record for Merrill shows 264 months, the 22 years given to one of three people; the person the release is about, a co-defendant sentenced to two years, is not what the field holds. The two records I suspected of a wrong flag, Bekkedam and CR Intrinsic, are now flagged as having a criminal parallel, and every record with a sentence now carries the flag.

Criminal parallel or not

811 records (25%) are flagged as having a parallel criminal matter. The two groups differ in what the civil record shows.

Before the flag corrections the last row reversed the pooled gap; it no longer does. The criminal-parallel penalty is larger in every row, by 2.5 times pooled and 1.1 to 2.1 times within a status, and the SEC judgment gap is so small, with overlapping middle halves ($150,000 to $1.5 million against $130,000 to $1.5 million), that I would not call it a difference. The pooled number partly reflects which kinds of record sit in each group: 23% of the CFTC’s penalty records (56 of 244) are flagged against 12% of the SEC’s (117 of 1,005), 49% of Ponzi-tagged records are flagged (231 of 474), and the CFTC’s flagged penalties have a median of $5.9 million against $1.0 million for its unflagged. Stanford’s record is now itself flagged, which lifts the flagged SEC total but barely moves any median. Treat the size of the gap as unstable.

A steadier pattern is the missing money. Only 173 of the 811 flagged records (21%) carry a US dollar penalty, against 1,077 of 2,462 (44%) without. The unflagged group includes 748 records whose only penalty is in a foreign currency (the AMF’s, SEBI’s and others), so the cleaner comparison is the SEC and CFTC alone: 173 of 691 flagged records (25%) carry a dollar penalty against 1,076 of 1,487 unflagged (72%). Of the 2,899 records recorded as settled or judgment, 318 of the 718 with a criminal parallel (44%) show no money, against 311 of the 2,181 without (14%); among the SEC and CFTC alone, 236 of 605 (39%) against 205 of 1,360 (15%). The three records I read (Gupta, Kita, McClatchey) say why in those cases: the civil disgorgement was deemed satisfied by criminal forfeiture and no civil penalty was set. Whether that explains the whole gap of 24 to 30 points I cannot say from the data.

Bars

881 records (27%) list at least one bar or injunction, and 812 list a bar of some kind, mostly of the kinds the SEC uses: a penny stock bar (243 records), an officer-and-director bar (226), a registration bar (158) or a trading ban (116), and 152 records, nearly all SEBI’s, list a restraint from the securities market, a debarment or a directorship ban; 149 list a conduct-based injunction, and 69 of those list nothing else. A record can list several. The share varies by family: 41% of information records (228 of 558) and 45% of issuer-and-structural records (121 of 269), against 20% for the order book and 1% for benchmark and cross-market. By respondent, 30% of records naming individuals only (535 of 1,771) list one, against 6% of entity-only records (31 of 490). By agency 31% of SEC records list a bar, 30% of the CFTC’s, 36% of the OSC’s (25 of 69), 25% of SEBI’s (164 of 656; 32% of its judgments and 22% of its settlements, none of its dismissals), 6% of the FCA’s (3 of 48), 9% of ASIC’s (10 of 113) and 2% of the AMF’s (4 of 209). Among SEC records, 39% of judgments list one against 27% of settlements. The field is free text taken from the release; a bar that a release does not mention is missing, and the regulators’ own remedies differ, so cross-agency comparisons say as much about what each publishes as what each imposes.

Penalty against gain, a small sample

Only 267 records carry an alleged gain, and 145 of those also carry a penalty (141 SEC, 4 CFTC). That is a small sample drawn almost entirely from one regulator, and it should not be read as a market-wide pattern. In it the median penalty is 1.0 times the alleged gain, the middle half runs from 0.50 to 1.5 times, 57 records (39%) are below 1 and 10 above 3. The highest ratio, 10, is the International Capital Group reverse-merger settlement, whose order was not in the cache and which I could not check; the lowest, 0.003, is Tobin, which sets a $165 million alleged gain against penalties of $460,928, so the gain is a scheme-wide figure and not the respondent’s. The Woodward gain, previously misread, now equals its $21,326 penalty.

Two warnings make the ratio weaker still. First, in 75 of the 159 records that carry both a gain and a disgorgement figure the two are within one percent of each other, so the “gain” field often restates the disgorgement. Where the two differ or there is no disgorgement (82 records with a penalty) the median is still 1.0 (middle half 0.27 to 1.6). Second, a scheme-wide gain set against one respondent’s penalty is not a ratio: the Bauer record sets a $145 million scheme against $1 million of penalty across the defendants, and Roda sets one respondent’s $560,762 profit against the other’s $5,602 penalty. A larger sample is the penalty against disgorgement, 771 records with both: the median is 1.0, the middle half 0.5 to 1.5, 205 (27%) within one percent of 1.0, 313 below and 253 above, 86 (11%) more than three times. Insider-trading records hold the sharpest pattern, 148 of 260 (57%) at about one times disgorgement; Ponzi records sit well below, with a median of 0.5 (108 records); CFTC records median 2.1 (56 records). These are what the records show, not what any rule requires.

What this does not show

It does not show what regulators sanction in general. These are the matters that regulators chose to bring and publish, and that this library ingested and read. Many are the large, the clear or the settled. Cases not published, not brought or not pursued are not here, so a median penalty here is a median of selected matters.

It does not show what settling does. The settled-judgment gap is large and holds across the cuts I tried, but a judgment record includes consent and default judgments, a matter that goes to judgment differs from one that settles in ways the records do not capture, and nobody observes what the same respondent would have paid by the other route. That the two groups differ is a fact about these records. Why is not something these records can answer.

Money that is missing is not money that was not ordered. 30% of records show no money; under a fifth of those are records still tracked only as “filed”, and 163 are dismissals, 122 of them SEBI’s. Disgorgement deemed satisfied by a criminal forfeiture, as in Stanford (where the disgorgement figure is missing and the penalty is the only dollar amount), does not show as money. Collected amounts are not in the data at all: an ordered penalty against a defendant in default, like Battoo, Cartu or Control-Finance, is a judgment, not a payment. The 543 records updated on 4 October often report outcomes for only some of their defendants, so a status or amount may describe part of a matter.

Other currencies are reported separately, not compared. Of the ASIC, FCA, OSC, AMF-FR and SEBI records, 748 have a penalty only in rupees, euros, Australian or Canadian dollars or pounds and are excluded from every dollar figure, so regulators outside the US are absent from the dollar comparisons. The section on other currencies gives each currency’s own median, but no conversion is made and no regulator can be ranked against another on penalty size. Their sentence figures are in months and are included (65 records, mostly ASIC).

Per record, not per person. 1,670 records sum several respondents’ amounts, a sentence figure is the longest term in the release, and a matter can appear in more than one record (the duplicates found so far are nulled, others may remain undetected). Five SEBI records each stand for a year of hundreds of orders. The 3,273 records are not 3,273 independent matters and not 3,273 respondents.

The tags and the fields were read by AI agents, not lawyers. The library’s original 2,428 records were compared with their primary documents on 2 and 3 October 2026 by Claude agents following written instructions, and they had sampled second reads; no lawyer reviewed them (see what reading every record found). The 543 records updated in the status rounds were researched on 4 October the same way. Each of the 209 French AMF records, added on 5 October, was written by an AI agent reading the French decision itself and was read once; a sampled second read on 7 October of 60 of the AMF records agreed on every field for 54 (90%, 95% interval 80% to 95%) and on defendants, status, amount and date for 59, and the disagreements were corrected. The 656 SEBI records, added on 7 to 9 October, were each written from a single AI reading of the order; for the first 489 a sampled second read of 60 on 8 October agreed on every field for 56 (93%) and the four misses were penalty amounts, so the rupee figures here are the weakest part, and for the 167 added on 9 October a sampled second read of 50 of them on 9 October agreed on every field for 46 (92%, interval 81% to 97%; the misses were two dissolved-company dismissals counted as out of scope and two technique tags, since corrected). The figures in this post were checked by recomputing them a second way and by reading the record narratives for the largest values; several orders in the cache, mostly SEC administrative PDFs, were not available, so the figures that rest on them are unverified. If you find a figure that is wrong, use the error link on the record’s page and 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.
• 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.
• 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.
• Price manipulation — Price manipulation is trading intended to create an artificial price — one that does not reflect genuine supply and demand — and it is the general charge a regulator brings when the conduct does not fit a more specific named technique.

Cases referenced

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