A life becomes data. The data becomes a score. The score becomes a decision no one admits the computer made.
Day 40 of 43 in The Punishment Machine
The screen contains rows. Name. Age. Offense.
Spreadsheets are excellent at rows and columns. Human beings remain stubbornly bad at fitting inside them. That does not make data useless. It does mean the neatness of the screen should never be confused with the messiness of a life.
Criminal history. Employment. Housing. Substance use.
Treatment. Violations. Program completion. Risk level.
The information looks orderly. A life rarely is.
The person may have survived poverty, addiction, trauma, unstable housing, prison, family loss, and years of trying to rebuild. The system translates that history into fields:
Yes.
No. Stable. Unstable. Low.
Moderate. High.
The officer may know the person. The judge may read the file. A prison case manager may conduct an interview. But the score arrives carrying a special authority.
It looks objective. Scientific. Consistent. Defensible.
The decision-maker can say:
That is what the assessment showed. The spreadsheet does not sign the order. It does not carry a badge. It does not sit behind the bench.
It can still influence how closely someone is watched, which programs are assigned, whether restrictions increase, and how readily progress is believed. The human being enters the system.
A classification comes out.
The Spreadsheet Is a Metaphor
The punishment machine does not always use a literal spreadsheet. It uses assessment instruments, case-management software, databases, algorithms, scoring sheets, checklists, and structured professional judgments. In federal community supervision, probation officers use the Post Conviction Risk Assessment, or PCRA, to help estimate who is more likely to fail on supervision or commit another crime, identify needs that may contribute to that risk, and direct supervision and treatment resources. (United States Courts)
Inside federal prisons, the Bureau of Prisons uses PATTERN—the Prisoner Assessment Tool Targeting Estimated Risk and Needs—to classify general and violent recidivism risk using static and dynamic factors. (Bureau of Prisons) These tools are more sophisticated than a column of numbers. But the title captures what they do.
They organize facts.
Weight them.
Compare one person with groups of other people.
Produce an output.
That output can affect liberty.
Data Can Improve Justice
Human intuition is not automatically fair. Judges, prison officials, and probation officers bring experience. They also bring fatigue, assumptions, fear, inconsistency, and personal bias. One officer sees a quiet person as stable.
Another sees evasiveness.
One judge sees employment as evidence of rehabilitation. Another dismisses it as merely expected. Structured assessment can force decision-makers to consider the same relevant factors, reduce arbitrary variation, identify treatable needs, and direct limited resources toward people who require more support. A reliable low-risk finding can also justify fewer restrictions.
Data can make justice more consistent. The danger begins when consistency is mistaken for truth. A score is not a neutral fact discovered in nature. People choose what the tool will predict.
They choose the data used to build it. They decide which factors count, how heavily each is weighted, where the categories begin and end, and what consequence follows from each category.
The calculation may be mathematical.
The design remains human.
The Database Records the Condition, Not Always the Cause
A case-management system may record: Employed: No. Housing stable: No. Treatment compliant: No.
Those entries may be accurate.
They may still conceal the most important facts.
Was the person unwilling to work—or rejected because of the criminal record?
Did he refuse treatment—or miss an appointment because the bus did not arrive?
Was housing unstable because he ignored available options—or because every landlord denied him?
Federal evidence-based-supervision guidance recognizes that treatment failures may result from transportation, childcare, housing, and other practical barriers rather than lack of motivation. (United States Courts) That distinction can disappear during data entry.
The system records the outcome.
It may not preserve the explanation.
Poverty becomes a risk factor.
The higher risk level produces more reporting, testing, and monitoring. Those demands make employment and housing harder to maintain. The tool begins measuring instability partly created by the system using the tool.
Precise Calculations Can Begin With Bad Information
A score may be calculated perfectly from incorrect data.
A dismissed charge appears unresolved.
A program completion is never entered. A legitimate job change becomes unstable employment. A missed appointment appears without the hospitalization that caused it. A provider reports poor progress after misunderstanding the person.
The software performs exactly as designed.
The result remains wrong.
A January 2026 Government Accountability Office review found that the Bureau of Prisons had not completed all required risk-and-needs assessments within required or internal timeframes, partly because of technology problems. GAO recommended improvements to data collection, and the Bureau agreed. (GAO) That does not prove every assessment is unreliable.
It demonstrates that even a validated tool depends upon accurate information, timely administration, trained users, and functioning technology.
The formula can be sound.
The implementation can still fail.
Bias Does Not Require a Race Field
A tool does not have to use race explicitly to produce different results among racial or ethnic groups. Other factors may reflect unequal histories:
Arrests.
Employment opportunities. Educational access. Neighborhood conditions. Housing stability.
Past supervision outcomes. Criminal-history accumulation.
A 2024 PATTERN revalidation found that the tool remained predictive across the racial and ethnic groups examined. It also found evidence that the general-recidivism instruments overpredicted risk for Black, Hispanic, and Asian men and women relative to White individuals. (National Institute of Justice) That does not establish intentional discrimination.
It shows why validation cannot stop with the question:
Does the tool predict something? The system must also ask: For whom is it accurate? Whose risk does it exaggerate?
Whose risk does it underestimate? What happens when it is wrong?
An overprediction may produce greater surveillance, fewer opportunities, delayed release, or more restrictive placement. An underprediction may leave the public insufficiently protected.
Both errors matter.
They do not impose the same harm upon the same person.
Reassessment Must Change Something
A dynamic assessment should respond when the person changes. Federal research has found that many people initially classified in the highest PCRA risk levels later moved into lower categories. Declining risk characteristics and scores were associated with lower recidivism than unchanged or rising scores. (United States Courts)
That is encouraging.
It means the tool can recognize improvement rather than preserve the original label forever. But reassessment matters only when it occurs on time, uses current information, and changes practice. A lower score should lead, when safety permits, to:
Fewer unnecessary contacts. Less intrusive monitoring. Reduced testing. Narrower intervention.
Greater independence.
An assessment that records progress without reducing control merely converts success into another field.
The person improves.
The system congratulates him.
The restrictions remain.
The Score Can Become a Shield
Decision-makers may use a score as one source of information. They may also use it to avoid responsibility.
Grant relief and something later goes wrong?
The official may be blamed.
Deny relief and preserve existing restrictions?
The burden falls privately upon the person. A high score offers an easy explanation for caution. A low score does not always provide equal permission for trust. The warning is treated as powerful.
The reassurance is treated as advisory.
The score becomes a ratchet.
Risk increases control.
Reduced risk does not necessarily reduce it. Federal supervision policy rejects that one-way approach by directing officers to respond to improvement and reduce intrusion for stable, compliant people. (United States Courts) The practice should match the principle. Data should be allowed to release pressure, not merely increase it.
The Person Should Be Able to See the Inputs
A person whose liberty is affected by an assessment should be able to understand it.
What information was used? Which facts increased the score? Which can change? Which cannot?
Is the information accurate? When will reassessment occur? What does the category actually change? How can an error be corrected?
What can the person do to reduce the classification?
Transparency does not require everyone to become a statistician. It requires explanation in ordinary language: You were classified at this level because of these factors.
This information can change.
This information cannot.
This is what the category means. This is when it will be reviewed. This is how you may challenge an error. Without that explanation, the person receives a verdict from a machine he cannot question.
Human Judgment Must Remain Human
A risk assessment should inform judgment.
It should not impersonate it.
The officer may know facts the tool cannot capture:
A new caregiver responsibility. A job offer. A deteriorating medical condition. A supportive relationship.
A local housing crisis. A provider mismatch.
A relapse followed by an immediate request for help. An act of courage or responsibility.
The score processes variables.
The officer investigates meaning.
Human judgment also requires accountability.
If the official departs from the score, the reason should be explained. If the official follows it despite important contrary evidence, that reason should be explained too. A person should not lose liberty because “the computer said so.” Nor should a validated low-risk finding be ignored because an official has a feeling.
The best decision combines data, professional judgment, present facts, and written reasons.
A Better Spreadsheet
The answer is not to return to intuition alone. It is to make data systems worthy of their power. Tools should be independently tested and regularly revalidated. Their definitions and major factors should be public.
Accuracy should be examined across groups.
Underlying information should be correctable.
Static factors should not overwhelm genuine evidence of change. Barriers created by poverty should be distinguished from deliberate refusal.
Assessments should occur on schedule.
Lower risk should produce lower supervision when safety permits. No major liberty decision should rest upon a score alone. Every person should receive enough information to understand how the classification was reached and what it means.
Data can improve justice.
Secret, unreviewable, or mechanically applied data can hide injustice beneath decimal points.
The Spreadsheet That Decides Your Freedom
The spreadsheet does not hate.
It does not forgive.
It does not know remorse.
It does not know fear.
It does not know whether a missed appointment resulted from defiance or a broken transmission. It does not know whether “unemployed” means unwilling to work or rejected forty times. It does not know whether the person has become someone different from the person described in the original file.
It knows the fields it was given.
That is useful.
It is not enough.
The danger begins when the system forgets that the score is a tool and starts treating it as the person. A human life can be translated into data. It cannot be fully contained there.
Use the spreadsheet. Check the boxes. Run the score. Then look up.
The person standing there is not the printout. Data should discipline human judgment, not replace it. When freedom depends on a number, the least the system can do is show where the number came from and leave room for facts that do not fit neatly in the cell.
