Skip to main content
lab-notes.aiPublic-service AI playbooks
Opportunity 11/17Justice and educationJustice and community safety

Learning Support in Custodial Settings

Explore what a suggestion about a learner's next course would need to know when enrolments and completions are not published by course.

Inspect starter dataSet up the build partner
Published sources
2
Starter data
20 synthetic records
Domain build partner
build-offender-learning
Last reviewed

From the draft strategy

Opportunity

The draft strategy names AI support for education in justice settings as a potential public-service application, linking personalised material with rehabilitation, without saying what protects the learner, what the curriculum is, or what would count as a good outcome.

Where the draft says it

Table 2 — potential public-service applications

Read the draft strategy source (opens in a new tab)

Source register

Research already done

Real published source

We reviewed these published sources so you do not have to start from zero. Each source shows what it covers, how it can be accessed, and why it matters.

2 sources

  1. Department of Justice

    Open

    Learning and skills

    What it covers
    How learning in Northern Ireland's prisons is organised: delivery by two further education colleges under agreement with the Prison Service, essential skills in literacy and numeracy, higher-level study by distance learning, and personal development plans used to schedule what each person does.
    Why it matters here
    It is where this playbook's course types come from, and it publishes no counts at all, which is why the numbers in the synthetic file have nothing published to check them against.

    Open the source (opens in a new tab)

  2. Department of Justice

    Open

    NI Prison Service statistics

    What it covers
    The Prison Service's official statistics publications and the notes on how they are compiled and what they can be used for.
    Why it matters here
    It is where prison figures do get published, and comparing it with the learning pages shows that course-level activity is not among them.

    Open the source (opens in a new tab)

Starter data

Starter dataset

Synthetic working data

Use 20 synthetic quarterly rows across five course types to explore next-resource choices without learner records.

AI-assisted research helped identify and interpret the published sources. We then created a small, non-sensitive synthetic dataset shaped by the information those sources expose. It is for exploration—not evidence, training, or operational decisions.

Records
20
Fields
5

Field preview

table view
  • id

    SYN-OL-01 · SYN-OL-02 · SYN-OL-03

  • course

    Essential skills: literacy · Essential skills: numeracy · Digital skills

  • quarter

    2025-Q1 · 2025-Q2 · 2025-Q3

  • enrolmentsBand

    50-100 · 20-50 · over 100

  • completionBand

    40-60% · 60-80% · under 40%

Inspect all 20 recordsView repository file (opens in a new tab)Download JSON (opens in a new tab)

How it was prepared

AI authored fictional banded enrolment counts and completion rates for each course type.

Limitations

  • Enrolments and completions by course are not published, so the two most useful columns here have no published counterpart at all.
  • The bands are invented, so nothing here describes learning in any prison.
  • A course-level table cannot hold what actually decides things in custody: what is running this week, who is allowed to attend, and whether a sentence is about to end.

Clone and build

Domain build partner

Build with a domain-aware coding agent

Clone the repository, then ask your coding agent to load the skill that ships with it. The skill brings the opportunity, sources, starter data, known unknowns and constraints into the conversation before anything is proposed. It follows the open Agent Skills standard, so it works in Claude Code, Codex and any other agent that reads skills.

01 / Clone

git clone https://github.com/Hypership-Software/lab-notes.ai.git

02 / Invoke

Claude Code
/build-offender-learning
Codex
$build-offender-learning

Conditions, not a checklist

Before you build

These are not footnotes. They are conditions that any responsible prototype would need to address.

  • Consent is different in custody

    Education, justice, disability, and behaviour information sit close together in custody, and consent means something different in a place people cannot leave.

  • Learning data can become surveillance

    Anything collected to support learning can be experienced as surveillance, or later used for a decision about discipline or release, which is why this file holds no learner records.

  • Partial histories can narrow opportunity

    A suggestion built from a partial history can quietly narrow what someone is ever offered, and in custody there is rarely another route to the same course.