A Right That Exists on Paper
Compassionate release, and the distance between a law and its use
Almost every state has a law that lets it release people who are dying, gravely ill, or old enough that the risk they once posed has passed. The laws have different names and different tests, but they share a premise: that a sentence can outlive its purpose, and that a system should be able to say so.
Almost none of them get used. That gap, between a right that exists and a right that operates, is the subject here.
This page is also an argument about method. Statutes and case decisions are qualitative data, and they are usually written up rather than charted. That habit costs something, because the questions people actually ask about a law are comparative: how many states require this, how far apart are the thresholds, what reasons do courts give, what changed and when. Those are chart questions. So each section below asks one of them, answers it with a form built for categorical and legal evidence, and carries a note on why that form won.
Nearly Everywhere on Paper
Start with the premise. When researchers last coded all fifty states, how many had any mechanism at all, and what did those mechanisms look like?
Forty-six states had a compassionate release mechanism; four did not (Illinois, Iowa, Nebraska, and Utah). Counts sum to more than fifty because a state can hold more than one kind of policy, so read these as counts of policy types, not as the fifty states sorted into piles. Source: Holland, Prost, Hoffmann, and Dickinson (2020), OMEGA 81(4), figure 1 at p. 613.
Two of those four have since acted. Illinois passed the Joe Coleman Medical Release Act in 2022, and Nebraska took up release for terminal illness and permanent incapacity the same year. That is the trouble with a decade-old map: it is the best one that exists, and it is already wrong.
Why this chart: a lollipop rather than a bar, because the quantity is a count of states and the bar area would add ink without adding information. The one category that is an absence, not a policy, is the only one in coral, so the eye finds the exception before it reads the labels.
What You Have to Prove
A law that exists still has to be satisfied. Across the states that had a policy, what did those policies require?
Source: Holland, Prost, Hoffmann, and Dickinson (2020), OMEGA 81(4), table 1 at p. 612. One caution carried from the source: the table reports age restrictions in nine states and the article text reports ten. The chart follows the table and flags the difference rather than quietly picking a side.
Why this chart: an icon array, the form risk communication research keeps reaching for when a share has to stay countable. Each dot is a state, so 45 of 46 reads as one state shy of everyone rather than as a long bar, and the denominator never leaves the picture. Sorting by prevalence rather than alphabetically turns a list of requirements into a finding.
How Old Is Old Enough
Where age appears in these laws, it appears as a number. States did not pick the same one.
Ten states used age as a factor at all, and in only three could age decide the question on its own. Wyoming’s policy set no number, requiring instead that a person be “incapacitated by age.” Source: Holland, Prost, Hoffmann, and Dickinson (2020), OMEGA 81(4), p. 614.
The same boundary problem shows up whenever a law puts a number on a fuzzy category, which is the subject of a short exercise on this site: How Old Is Old?
Why this chart: a span, not a bar chart of two values. The finding is the distance between the endpoints, so the distance is what gets drawn. Two bars would have invited the reader to compare their heights, which is not the point and would have implied that fifty-five is somehow less than seventy in a way that matters.
Eligible, Applied, Released
Now the use: following everyone the states themselves counted as eligible over three years, how many made it to a release?
Source: Holland, Prost, Hoffmann, and Dickinson (2021), Mortality 26(1), pp. 55 to 56. Denominators are reporting states only, so this is the funnel among systems that could produce the number at all. Note also what the last bar is not: the share who were refused. People leave this process by dying before a decision and by withdrawing, so the gap between applying and release is wider than any denial rate.
Why this chart: a funnel drawn as plain nested bars on a shared axis, because each stage is a subset of the one above it and the reader should be able to measure the drop, not just sense it. Tapered funnel graphics encode the same numbers as areas that no one can compare. The shading deepens as the pool narrows, the same part-in-whole grammar the federal chart below uses.
Does Anyone Have to Count?
Every number in the section above depends on a state having written something down. Most of the work of finding out whether a law functions is finding out whether anyone kept the record.
39 of 49
tracked how many people they released
32 of 49
tracked how many people applied
17 of 49
tracked how many people were eligible
2 of 50
required that incarcerated people be told the policy exists
Tracking counts are from Holland et al. (2021), Mortality 26(1), p. 55, across the forty-nine state systems that responded to a public records request. The notification figure is from Holland et al. (2020), OMEGA 81(4), p. 616: only Alabama and New Mexico required that people in their custody be informed that the policy existed.
Those two findings belong together. A right nobody is told about, in a system that does not count who was eligible for it, will look underused and unwanted in exactly the same way it would look if it were simply not offered.
Why this chart: tiles, because four numbers with different denominators cannot share an axis, and forcing them onto one would invent a comparison the data does not support. The last tile is the one that changes how the first three read, so it goes last.
The Federal Docket
Federal law runs on a different track: since 2018 people can file their own motions, and a judge rather than a prison official decides. Here are nearly four years of that docket, month by month.
Across the full period, courts decided 13,614 motions and granted 1,866 of them, or 13.7 percent. Monthly volume fell from 584 in October 2021 to 207 in June 2025, while the share granted rose from 9.2 percent to 15.9 percent. Source: U.S. Sentencing Commission, Compassionate Release Data Report, 3rd Quarter FY2025, table 1. Fiscal year 2025 figures are preliminary and cumulative through the third quarter.
Why this chart: one bar nested inside another, rather than two lines on two axes. Grants are a part of motions decided, not a separate series, and drawing a part inside its whole keeps the reader from reading a rising rate as a rising count. They moved in opposite directions here, which is exactly the confusion a second axis would have manufactured.
Why Courts Say Yes, and Why They Say No
Courts record their reasons, and the reasons are the most direct evidence there is of what this law has come to mean in practice. Grouped into themes, what does the federal docket say?
Among granted motions, reasons about sentence length and changes in law accounted for 33.7 percent of citations and medical reasons for 25.2 percent. Among denials, the largest theme by far was the general sentencing factors at section 3553(a), a discretionary weighing rather than a finding about health. Courts may cite several reasons per case: 457 reasons across 299 granted cases, and 3,703 across 1,766 denied. Source: U.S. Sentencing Commission, Compassionate Release Data Report, 3rd Quarter FY2025, tables 10 and 11.
One procedural line is worth reading on its own. Failure to exhaust administrative remedies was cited 325 times, which is a paperwork step rather than a judgment about whether release was warranted.
Why this chart: small multiples sharing one x scale, so grants and denials are read side by side without either being subtracted from the other. They are not complements: a case can be denied for four reasons and granted for two, so a single diverging bar would have implied an arithmetic that does not exist.
From Statute to Data
The chart above depends on a step that is easy to skip past. The Sentencing Commission publishes reason codes; the themes are mine. Here is the whole assignment, so the grouping can be argued with.
| Outcome | Theme (my coding) | Reason code (the Commission’s) | Cited |
|---|---|---|---|
| Denied | Discretionary weighing | Section 3553(a) sentencing factors | 926 |
| Denied | Discretionary weighing | Rehabilitation insufficient | 371 |
| Denied | Discretionary weighing | Protection of the public | 115 |
| Denied | Discretionary weighing | Post-sentencing conduct | 40 |
| Denied | Family | Other care available for minor child | 101 |
| Denied | Family | Other care available for spouse or partner | 16 |
| Denied | Medical proof | Insufficient proof of serious medical condition | 263 |
| Denied | Medical proof | Able to provide self-care in prison | 177 |
| Denied | Medical proof | Not at risk from COVID-19 or pandemic | 157 |
| Denied | Medical proof | Insufficient proof of deteriorating health | 29 |
| Denied | Medical proof | Insufficient proof of functional or cognitive impairment | 11 |
| Denied | Other | Other | 625 |
| Denied | Procedural | Failure to exhaust administrative remedies | 325 |
| Denied | Threshold not met | No extraordinary and compelling reason provided | 407 |
| Denied | Threshold not met | Does not meet age or time-served requirement | 81 |
| Denied | Threshold not met | Mandatory minimum penalty | 31 |
| Denied | Threshold not met | Binding plea agreement | 4 |
| Denied | Unspecified | No reason recorded | 24 |
| Granted | Conduct and rehabilitation | Rehabilitation | 73 |
| Granted | Family | Family circumstances: care for parent | 18 |
| Granted | Family | Family circumstances: care for minor or disabled child | 15 |
| Granted | Family | Family circumstances: care for other immediate family | 8 |
| Granted | Family | Family circumstances: care for spouse or partner | 2 |
| Granted | Medical | Serious physical or medical condition | 57 |
| Granted | Medical | Terminal illness | 25 |
| Granted | Medical | Deteriorating health due to aging | 10 |
| Granted | Medical | Age 65 with deteriorating health and time served | 9 |
| Granted | Medical | Prison failure to provide treatment | 8 |
| Granted | Medical | Serious functional or cognitive impairment | 3 |
| Granted | Medical | COVID-19 or pandemic | 3 |
| Granted | Other | Other | 23 |
| Granted | Other | Abuse by a corrections officer or contractor | 2 |
| Granted | Sentence and law change | Unusually long sentence and change in law | 56 |
| Granted | Sentence and law change | Multiple section 924(c) penalties | 34 |
| Granted | Sentence and law change | Other mandatory minimum or long sentence | 26 |
| Granted | Sentence and law change | Young age at time of offense | 13 |
| Granted | Sentence and law change | Section 851 enhanced drug penalties | 12 |
| Granted | Sentence and law change | Career offender issues | 11 |
| Granted | Sentence and law change | Armed Career Criminal Act issues | 2 |
| Granted | Unspecified | Not recorded | 38 |
| Granted | Unspecified | Extraordinary and compelling (unspecified) | 6 |
| Granted | Unspecified | Nearly meets the policy statement | 3 |
Two judgment calls in there are worth naming. “Young age at time of offense” sits under sentence and law change rather than under anything medical, because it is an argument about whether the original sentence still fits. “Rehabilitation” is kept apart from both health and sentence length, because the Commission notes it never appears alone in a granted case: it always arrives alongside another reason.
Why this chart: a table, which is the right chart when the reader’s question is “what exactly did you do.” A chart of a coding scheme shows its shape; the scheme itself has to be legible line by line, or nobody can tell you that you put a code in the wrong pile.
One State, Thirty Years
National counts flatten a law into a yes or a no. Any single state’s version arrived in pieces, and Wisconsin’s is a clear case: a statute, a rollback, a rule, a set of forms, a policy.
Sources are the instruments themselves: 1997 Wis. Act 283, 2001 Wis. Act 109, 2009 Wis. Act 28 creating Wis. Stat. sec. 302.113(9g), 2011 Wis. Act 38, Wis. Admin. Code DOC sec. 302.41, Wisconsin Court System forms CR-254 through CR-256, DAI Policy 302.00.13, and a 2023 amendment to the notice provision. Today all three layers operate together and the sentencing court makes the final call.
Notice the 2011 entry. A release mechanism does not only get built, it gets narrowed, and a chart that only recorded expansions would have shown a straight line up.
Why this chart: a timeline set vertically, events alternating sides of the spine, because these labels are phrases and phrases need horizontal room; along a horizontal axis each label gets only the width of its own gap, and eight of them collide. Color marks the kind of instrument, which is the point most easily missed: what looks like one law is a statute, a rule, a form, and an internal policy, each changeable by a different body on a different schedule.
What This Cannot Tell You
Everything above is the law as written and the decisions as recorded. Both are real, and both stop short of the question people actually care about, which is why a law that exists so widely operates so rarely.
The records cannot answer it because the deciding happens before anything gets written down. A prison official weighs whether to forward a petition. A physician decides how confidently to put a prognosis in writing. A board or a judge weighs a small, quiet benefit against a rare, loud risk. Those are judgments made under uncertainty, and the file records the outcome, not the reasoning that produced it.
That is where I think the useful work is. A few questions I would want answered before claiming to know why these laws sit unused:
- Does what the policy is called change who applies and who is approved? The federal statute says “extraordinary and compelling.” Most states have moved to medical or geriatric parole, medical release, or conditional medical release. Only New Jersey and Utah still use compassionate release as the exclusive title, and the words carry different promises: mercy in one, clinical necessity in the other.
- Does telling people the policy exists change the number who apply? Two states required it as of 2016, which is close enough to a natural experiment to be worth running.
- Do clearer medical criteria move decisions, or only move where the argument happens?
- When a deadline is written into the statute, does the decision arrive sooner, and does the effect survive the first few years?
None of those are answered here, and none of them can be answered from statutes alone. They need the decisions, the timing, and the people who made them.
Where the Numbers Come From
Every figure on this page is drawn from a public document, and each data file carries its own citation on every row.
Federal decisions. U.S. Sentencing Commission, Compassionate Release Data Report, 3rd Quarter FY2025 (tables 1, 10, and 11). Fiscal year 2025 data is preliminary and cumulative through the third quarter. These are federal court decisions under 18 U.S.C. sec. 3582(c)(1)(A), which since the First Step Act of 2018 may be sought by a sentenced person directly.
State policy content. Holland, Prost, Hoffmann, and Dickinson (2020), “U.S. Department of Corrections Compassionate Release Policies: A Content Analysis and Call to Action,” OMEGA: Journal of Death and Dying 81(4), 607 to 626. Their coding covers all fifty states from policies collected in 2016, which makes it the most complete state-level source available and, by now, a decade old. Counts here are cited to it; the figures are drawn fresh rather than reproduced.
State utilization. Holland, Prost, Hoffmann, and Dickinson (2021), “Access and Utilization of Compassionate Release in State Departments of Corrections,” Mortality 26(1), 49 to 65, built from public records requests covering 2013 to 2015.
Wisconsin. The statutes, administrative rule, court forms, and department policy named in that section.
Two things this page does not do. It does not mix the federal and state figures into a single national rate, because they count different things over different periods under different law. And it does not report a denial rate, because the published sources do not support one: people leave these processes by dying and by withdrawing, not only by being refused.
The state layer is the weak link, and it is weak in a specific way: it is the best fifty-state coding anyone has published, and it describes a world two of its four empty states have already left. Re-coding it from current statutes is the next piece of work.
Related on this site: How Old Is Old?, a short exercise on putting numbers to a fuzzy category, and Maturity Gap, which asks the same kind of question at the other end of a life.