Data documentation

A criminal case reaches this dashboard only if it has a filing date recorded for a U.S. District Court. Some prosecutions handled entirely in a magistrate court may have dates that appear “late” or “misdated.” These cases are excluded from dashboard charts but will appear in the Case Look-Up feature. Any analysis or chart built on the dashboard should not be categorized as reflective of all federal criminal prosecutions, given these differences.

In fiscal year there are approximately prosecutions excluded from the visualization, of which are illegal entry cases under 8 U.S.C. 1325(a)(1), many in the southern and western districts of Texas, in Arizona and in New Mexico.

There’s over criminal case records in this dataset but (or ), have no U.S. District Court filing and therefore are not visualized on the dashboard. In fiscal year this dashboard counts criminal cases filed and defendants.

Declined matters and matters that never went to court are, similarly, not visualized. Declined matters have their own dashboard.

By counting only district court filings, this dashboard is comparable with the figures that DOJ publishes at the end of the year. Using this smaller case count, our dashboard visualizes of DOJ’s published count of criminal cases filed and of its published defendants. The wider count, taking in magistrate court prosecutions, matches no published DOJ table at all, making data validation nearly impossible.

The word "filed" has different meanings on the dashboard versus the Case Look-Up function.

The dashboard dates a case by its earliest filing in a U.S. District Court across cases that reached that court. Case Look-Up dates a case by its earliest filing in any court, across every record with a court proceeding, which includes magistrate court prosecutions and proceedings that happen after a case has been decided. Therefore Case Look-Up always returns a larger number. Both are accurate.

For example, if you filter for criminal cases filed in fiscal year , the dashboard shows and Case Look-Up shows . Because the two count different populations, a Case Look-Up search is not the detail behind a dashboard figure and the two will not reconcile. For more detailed breakdowns of the dashboard, use the table feature at the bottom of that page.

In the process of completing data validation, it became clear that the DOJ’s annual reports do not always align with their own data from one year to the next.

In each of the five annual reports for fiscal years 2021 to 2025, the civil caseload table gives a caseload pending at the start of the year, the cases filed, the cases terminated and the caseload pending at the end. However, the four figures do not agree with one another: the published end of year figure is between and smaller than the other three imply. The criminal tables in the same five reports are inconsistent in the same way. The reports do not say why and it’s not clear if the discrepancy is due to a reporting error, a change in methodology or some other reason.

DOJ publishes a comparable count once a year in its annual report; the criminal case dashboard does not reproduce it exactly.

For fiscal year this dashboard counts criminal cases filed, compared to that DOJ reports, or of DOJ’s figure. Regarding the number of defendants, the dashboard counts against DOJ’s , which is .

Note this analysis is not replicable across other years. Data validation shows our dashboard matches, on average, 98% of the DOJ’s total reported criminal cases and 103% of their total defendants.

DOJ publishes the civil caseload once a year. Compared against its five annual reports for fiscal years 2021 to 2025, this site’s counts differ from DOJ’s in every one of the five years, on both counts.

The reason has not been identified, and the difference is not the same size in each year, so it is not a fixed amount that can be added to or taken off. Part of it is that our most recent years are still being reported and will move; that cannot be the whole of it, because the oldest years here have long since settled and still differ.

The total dashboard count differs from DOJ’s published count, sometimes due to a greater number of cases, sometimes fewer.

Cases pending is a count taken at the end of each month of the civil cases that had a court record by that date and no recorded ending. It is not a running balance of arrivals minus departures, and it does not include events that happened after the published date.

Part of the discrepancies can be accounted for by later data that will retroactively change the count of cases pending at the end of a given month. One trend we’ve noticed is the U.S. as a plaintiff accounts for almost none of the difference; why is not clear.

Matters pending is a count taken at the end of each month of the civil matters that had not yet reached court and had not been closed, using only what LIONS had recorded by the end of that day. It is not a running balance of arrivals minus departures.

We count a matter as closed when LIONS gives it a closing date, or when LIONS records an outcome for it before it reached court, such as a declination or a settlement. DOJ does not say which of these it uses, so this is our reading of the data and not DOJ’s published method.

Because it counts only what had been recorded by each date, the figure moves with the timing of data entry as well as with the number of matters open. Measured on this site’s July 2026 data, some single months move sharply, when an office enters a batch of matters after the month they belong to: for U.S. as defendant, the national count for August 2022 sits about 8% below the average of the months either side of it for this reason. District figures move far more than national ones. Read a change over a single month with caution, and a change in one district’s figure with more.

Our matters pending calculation starts in . Earlier years were left out because we were unable to match the numbers in the DOJ annual statistical reports.

The table below sets this site’s count at 30 September of each year beside the figure DOJ published in its annual statistical report for the same date.

Because each month’s count uses only what LIONS had recorded by the end of that month, later releases of the data have changed past months very little, so the difference from DOJ’s figures in the table is not a matter of recent months still being reported.

Do not add matters pending to cases pending to get a total. The two are counted in different ways, and a matter that reached court shortly before a month’s end can be counted in both.

We caution against using a single month’s matters pending to signify events in that month.

The Criminal Cases dashboard and the Referring Agency dashboard are built from the same records by two different builds, and their criminal totals differ by a handful of cases.

For fiscal year the Referring Agency dashboard counts criminal cases filed.

We have yet to establish a cause but the difference is small enough that it will not change any reading of a trend, and large enough that two figures quoted side by side will not match.

The most recent months in each data update from LIONS are always incomplete but the variance is not steady. Every monthly data upload is re-reported more completely in later updates, which we are referring to as “settling” the data. The most recent months on any chart are the least complete, and they are marked on the chart, in the table and in the download as "provisional."

The scope of missing data is not a constant factor, so we can’t provide an estimation of what’s missing. For example, between the May and June 2026 updates, the newest month of criminal terminations grew by a factor of . Between the June and July 2026 updates the same measurement gave . Critically, we still do not believe that this is the final figure, just one step closer.

Users should note that a sharp decline at the most recent month on the chart is not evidence of a decline, but rather another indicator of incomplete filing data.

All provisional data is shaded on charts, denoting that they are incomplete. The scope of provisional data depends on the dataset: three months for criminal filings, four for civil filings, six for anything related to terminations, dispositions, declinations or pending cases, and one month for pending matters. Note that figures still move outside of the provisional range, but this is smaller and smaller as the months pass. We’ve noticed that civil caseload data continues to move for cases two years or older, but by smaller than 1 percentage point.

Each monthly update does not simply add new figures for the most recent month, but reports the data from all previous months more completely, so a figure for a month years in the past can still change. If you use a figure from this dashboard, include the dashboard visit date.

A criminal case can carry several program category codes and is counted under every one of them, so adding all categories will result in an overinflated total. To see total cases over a period of time, see the "all categories" figure.

Over the whole period, adding the umbrella categories together gives criminal cases filed against a true total of , which is too many. Within fiscal year the same sum is too many. Expanding every umbrella and adding all of the specific categories instead gives too many over the whole period and within fiscal year .

Another option is to filter by "Primary only", which counts each case once under its first entered code.

Similarly, there may be multiple referring agencies for a single case, so agencies should not be added up to reach a total. Instead, use "All agencies," or filter by "Lead agency," to see a reflective sum.

It is possible for the dashboard to return no data if you filter for one district, one cause of action and one role in the Civil Matters and Cases dashboard or the Cases by Referring Agency dashboard.

These months are not errors, even if its pending cases, it just shows that a caseload was empty, not that a month is empty. The one exception is for provisional data, which may be empty because they’re incomplete. For more information about provisional data, read below.

The four dashboards provide some general topline calculations based on the filtered data. All displayed percentages are analyzed from the whole sample, even if data is visualized in quarters or years, never averaging the monthly data.

The data table and the CSV download have fixed columns that are not affected by the filters you select. The table and download will always show the same columns, even if you have filtered for a single district or cause of action. However, the data you download will be filtered, so there may be empty columns in the table.

Additionally, the total number displayed is only the total of the filtered data, not the entire dataset, unless you have selected all districts, causes of action, and roles.

As noted above, Case Look-Up data is organized by the earliest filing date in any real court, so there may be cases that date earlier than the dashboard dates or outside of your filtered range.

Another discrepancy is the court type; court type is the first court that a record appeared in. For example, a case might show Magistrate Court but still have a U.S. District court filing date later

In the most recent full fiscal year that affects about of the criminal cases a search returns.

In Case Look-Up, the disposition column is only recorded for matters that reached a court and were declined. However, it is not representative of all declined cases, because many of them do not reach a court. In addition, some matters are declined and later prosecuted, and these would not be visible in the data.

To see declined matters only, you can filter by "declination" and set "declined matters only." To visualize and filter immediate case declinations terminated by U.S. Attorney’s Offices, visit our declination dashboard.

The three record types.

Case - A case is when there is at least one proceeding in a real court that is not post-disposition. These records carry a filed date and a court type.

Matter - When the USAO records no real-court proceeding. Filed and court type are blank.

Post-disposition Proceeding - The only real-court activity USAO records for this case is post-disposition - activity recorded after a case has already been disposed of. There is no filing that opens a case, so filed, court type, court no. and U.S. District Court filing date are blank on every one of these records. A disposition may still be shown: it is the outcome recorded on the post-disposition proceeding itself, not the outcome of a prosecution.

Some caveats about record types. The name describes where the record sits in the court history. It is not a description of what the proceeding was about: if LIONS shows that the activity is post-disposition, it does not record what kind of proceeding it was. Most of these records carry a program category or a lead charge, some carry neither, and we have not labeled them beyond what is recorded.

Why court type is blank on some records. Court type is the first court a record appeared in, so it is blank wherever there is no real-court proceeding to name. The Court type list no longer offers the post-disposition values. Use Record type to separate those records instead: choose Case to exclude post-disposition proceedings, or Post-disposition Proceeding to see only them.

Differences in dashboard display and data download. Justice Connection did some data labeling to make the dashboard easier to read so there may be some slight changes if you choose to download data as a CSV. One example: dates are MM/DD/YYYY on screen and YYYY-MM-DD in the original data. Another change is the CSV’s column headings match the underlying data field names rather than our table headings - for example dc_filed_date for “U.S. District Court filing date”.

Disposition records. Disposition is the last outcome recorded on a proceeding in a real court. Where a case had several court proceedings, it is the outcome of the latest one.

A matter that was declined and later charged and prosecuted shows the outcome of that prosecution, so it can read guilty. To find declined matters use the declination filter, which is built from the declination itself.

The “Declined – …” values. The disposition list also carries several values beginning declined –, such as “declined – agency request”. Those are civil outcomes from debt-collection and bankruptcy work. They are not criminal declinations.