How long enforcement cases take

Across 619 records with a measurable span, the median from filing to recorded resolution is about 1.6 years. SEBI's 656 records are excluded: its filing date and resolution date are the same order date. Of 733 records once marked "filed", 540 of the 720 remaining now have an outcome and 180 stay filed, a ceiling on truly pending cases.

Published 2026-10-04 · 10 min read

After the 4 October 2026 status rounds, 619 of the library’s non-SEBI case records carry two different dates that let anyone time a case from filing to a recorded end. Among them the median is 1.62 years, about 19 months, with the middle half running from 0.83 to 2.83 years. The library’s 656 SEBI (India) records are left out of this and every other duration on this page: a SEBI record’s filing date is the date of the order and its resolution date is the same date, so there is no span to measure (see below). The first version of this post found only 85 such records and a median of 0.24 years, because 733 records then had no outcome at all. A hand-checked random sample of 100 of those 733 had suggested most had in fact ended; two full research rounds have since confirmed it: 540 of the 720 of those 733 that remain in the library (75%) now have a settled, judgment or dismissed status and 180 remain “filed” (13 of the 733 were deleted on 7 October as non-matter records). The 180 is an upper bound on truly pending cases: 116 were checked and no outcome was recorded (some because pages could not be fetched) and 64 were not reached.

How it was measured

A Node script, scripts/analysis/time-to-resolution.mjs, reads every case file and prints every number below. It runs in two parts, and the figures are as of 2026-10-07, after the two status rounds on 4 October, the flag corrections, the deletion of 21 non-matter records and the addition of the French AMF (5 October) and SEBI (7 to 9 October) records. The 209 AMF (France) records are all recorded as decided and resolved on the decision date, so they carry no filing-to-decision span and none is in the timeable set.

Why SEBI is excluded. SEBI is the library’s seventh regulator, with 656 records dated 2022 to 2026. 651 of them have a resolution date equal to the filing date, because each rests on one order and the library dates it by that order. The other 5 are batch records, each grouping a year of near-identical adjudication orders on reversal trades in illiquid BSE stock options (1,577 orders across the five); their resolution date is the last order of the year and their filing date the first, so the gap is the spread of the batch, not the length of any case. SEBI’s investigation, show-cause notice and hearing come before the order and none of those dates is in the record. A duration for SEBI would therefore be zero for 651 records and meaningless for 5, so no SEBI record enters any figure, chart or percentage here. This is a limit of the dates the library holds, not a finding that SEBI cases are quick.

What changed in the inputs. Earlier on 4 October, 733 records had status “filed”. AI agents then researched them in two rounds and set a status for 543 (295 settled, 232 judgment, 16 dismissed), a resolution date, and often money fields and timeline events. The other 190 remained “filed”. On 7 October 13 of the 733 (3 of the 543 and 10 of the 190) were deleted as non-matter records (SEC subpoena-enforcement applications), leaving 720: 295 settled, 231 judgment, 14 dismissed and 180 filed. Outcomes often cover only some defendants, and the record narratives say so. The original list of 733 slugs is frozen in scripts/analysis/time-to-resolution-filed-733.json.

Part 1, the records that can be timed. Of the 2,617 records from the other six regulators, 1,656 have a resolution date equal to the filing date (mostly SEC orders instituted and settled the same day, all of the AMF and FCA), 332 have no resolution date, and 629 have a resolution date that differs. Ten of those 629, all SEC litigation releases, show a resolution a few days before the recorded filing date. In those records the “filed” date appears to be the release date, so they were left out, leaving 619. The status rounds supplied the date for 535 of them; 84 were already dated. A second measure, filing to the latest dated event in a record’s timeline, covers 670 records with two or more events. It agrees closely (median 1.58 years) and 614 records appear in both, so it is not an independent check. The figures use calendar days divided by 365.25.

Part 2, the “filed” records, in two stages. The first stage is the original test and is kept as the account of how this was found. The script sorts the 733 slugs, shuffles them with a fixed seed (20261004) and takes the first 100, printing the list so anyone can redraw it. Five Claude AI agents each took 20 and looked for how the matter ended, first in the cached regulator documents and then on the regulator’s own site or a court record. The allowed answers were settled, judgment, dismissed, still pending, or could not find. I opened three of the sources myself (CFTC release 8467-21, the SEC Lek Securities release and the Darvasi final judgment) and all three said what the agent reported. No lawyer reviewed any of this. Where a matter had several defendants the outcome is the lead defendant’s. Intervals are Wilson 95% intervals. One of the 100 sampled records was among the 13 later deleted, so the first-stage counts below are the test as it was run, and the comparison with today’s data covers the 99 that remain. The second stage is the full round described above, whose result is read straight from the case files.

What it shows

Timing, among the records that can be timed. Half took under 1.62 years, a quarter took more than 2.83 years, and the longest, an SEC insider-trading case, 10.6 years. 192 records (31%) were resolved within a year and 39 (6%) took over five years. Without the slowest tenth (62 records) the median is 1.49 years and the mean falls from 2.04 to 1.61.

Settled and judgment records are close: 1.57 years against 1.70, so the earlier finding that settlements ran a week against five months for judgments no longer holds. It came from a set dominated by Ontario consents. The SEC (1.78, n = 436), CFTC (1.77, n = 81) and ASIC (1.62, n = 34) medians are close to each other. Records with a criminal parallel (292) took longer than those without (327): 2.01 against 1.34 years. Both groups mix short and long cases, and the data cannot show cause. Dismissals (16, median 1.78) are few, and appeals (1) and unknowns (2) too few to say anything. SEBI’s heavy use of dismissals (122 of its 656 records) cannot be set beside these, because its dismissals carry no span.

Why the first figure was so low. 68 of the 619 are Ontario, with a median of 0.21 years; excluding them the median is 1.75 years (n = 551). Before the status rounds, 84 records had a later resolution date, and their median was 0.25 years, 67 of them Ontario. The 535 dated by the rounds have a median of 1.75 (middle half 1.00 to 2.97). The earlier “0.24 years” was a statement about the Ontario records, not about US or UK enforcement. The data still skews: of the SEC’s 1,844 records, 1,181 have a same-day resolution, 217 none and 446 a different date, and all 48 FCA and all 209 AMF records are same-day, so none can be timed. The same is true of 651 of the 656 SEBI records, which are the reason they are left out.

Is the resolution date really a resolution date? Not always. For the 535 records dated by the rounds with a later resolution date, I read the timeline event on that date. In 53 (10%) the wording is of an announcement or publication (“CFTC announces a consent order”, “published 13 January 2022”), 28 CFTC, 23 SEC and 2 ASIC; in 18 (3%) it gives only the announcement and no entry or order wording. Since the order or judgment usually precedes its announcement, these dates may overstate the span by the gap, typically days to weeks but occasionally longer. Those 53 have a median of 1.65 years against 1.78 for the other 482, so they do not drive the result. Some other records, where an event text does not say, may also carry an announcement date; I could not tell. So treat 10% as the share I could see and the true share as uncertain. The multi-defendant effect points the other way: the date is often that of the last defendant’s outcome, which lengthens the span.

What happens to the “filed” records. First the sample of 100, hand-checked:

The 71 that had ended are 71% of the sample (interval 62% to 79%). Scaled to 733, that suggested roughly 450 to 580 with an out-of-date status, about 520 at the midpoint. Among the 77 records where something was found, 92% had ended (interval 84% to 96%). Six were pending and 23 untraced. By agency the 99 sampled records still in the library are 80 SEC, 15 CFTC and 4 ASIC records, which follows the filed group’s mix and is too thin to compare agencies.

Then the two full rounds, which updated the data itself:

540 of the 720 (75%) are now resolved, inside the sample’s interval (62% to 79%), and 180 (25%) remain filed: 147 SEC, 30 CFTC and 3 ASIC. By agency, 147 of 576 SEC records (26%), 30 of 107 CFTC (28%) and 3 of 36 ASIC (8%) are still filed; the one Ontario record is resolved. Of the 180, 116 carry the 4 October review date, meaning the rounds looked at them and recorded no outcome (the status-resolution agents’ notes say this was mostly blocked fetching or nothing found, which I did not re-audit), and 64 were not reached. So 180 is an upper bound on the truly pending, not a count of live cases. I have not checked how many of them have in fact ended.

Cross-check against the sample: 75 of the 99 sampled records still in the library are no longer filed. Of the 71 the hand check found ended, 70 are now resolved and 1 is still filed in the data. Of the 22 the hand check could not trace that remain, 5 now have an outcome. Among the 70, the status label matched the hand-found outcome in 63.

What this does not show

• Selection. These are cases regulators chose to bring and publish. Matters closed without a public step, or never announced, are not here, and the speed of a published case says nothing about enforcement in general.
• Timing is still biased toward finished, documented cases. A record needs a second date to be timed, so a case still open, or never followed up, is invisible to Part 1. With 180 records untimed because still filed, the 1.62-year median is not an estimate of how long a typical case takes; open cases, being longer so far, would probably raise it.
• Three of the seven regulators cannot be timed or barely can. SEBI (656 records), the FCA (48) and the AMF France (209) give no filing-to-resolution span, because their records hold one date. The medians describe the SEC, CFTC, ASIC and Ontario records.
• The resolution date is not always a resolution date. It can be an announcement date (53 records show that wording, 10% of the 535), a release date (10 SEC records have it before filing), or the last defendant’s outcome in a multi-defendant matter.
• A sample of 100, then an unaudited round. The sample’s shares carry wide intervals and the agent-found outcomes are only as good as the pages opened. The full round was not independently checked either, and “filed” for 180 means “no outcome recorded”, not “pending”.
• Outcomes often cover only some defendants. Where the status says settled or judgment, other defendants may have ended differently or not been checked. The two Lek Securities records in the library describe one case.
• Checked by AI agents, not lawyers. The library’s original 2,428 records were read by AI agents on 2 and 3 October 2026 and had sampled second reads; the 543 records updated in the status rounds (the sample and the full rounds) were researched on 4 October the same way, with no sampled re-read of the round. The 209 French AMF records had a single reading and a 60-record sampled second read on 7 October; the first 489 of the 656 SEBI records had a single reading by one AI agent, followed by a 60-record sampled second read on 8 October agreed on every field for 56 (93%, interval 84% to 97%), the four disagreements all being penalty amounts; 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). A court outcome found on a regulator’s page is a regulator’s report of it, not the docket itself. No lawyer reviewed any of it.

If you find a record whose status is wrong, use the error link on its page.

Cases referenced

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