STRIPED DONKEY / Worked examples

Start with a question.
Follow the evidence.

Three ways to try connected synthetic data. Inspect a reference result, explore its records and reproduce it with the included code.

AI evaluation · synthetic sample

Which obligations remain operative after an appeal?

Nairobi · source case 86587 → appeal 191198

Reference answer · as of 3 February 2022

The original resolution is superseded. The appeal judgment is current. This extract does not establish replacement operative obligations.

KES 2,796.25 original obligation − KES 1,901.45 recorded payments = KES 894.80 arithmetic difference. That difference is not established current debt.

Inspect the supporting events

Original obligation · E9, E10, E11

Payment obligation 169681 is linked to source resolution 2536: KES 2,796.25.

E9 · 2021-08-13: RESOLUTION_RECORDED
E10 · 2021-08-13: REMEDY_RECORDED · KES 2,796.25
E11 · 2021-08-13: OBLIGATION_EFFECTIVE · KES 2,796.25

Payment is not the same as full compliance · E15, E17, E22

Recorded payments total KES 1,901.45. The KES 894.80 arithmetic difference is not asserted as current debt. FINAL_PAYMENT is the supplied event label; the later source checkpoint remains PARTIALLY_COMPLIED.

E15 · 2021-09-01: OBLIGATION_EFFECT · KES 627.48
E17 · 2021-10-06: OBLIGATION_EFFECT · KES 1,273.97
E22 · 2021-10-26: COMPLIANCE_CHECKPOINT

Enforcement was stayed · E23, E24

Stay 33 is GRANTED for ENFORCEMENT, effective 14 November 2021 to 3 February 2022. A stay request alone does not grant a stay or extinguish an obligation.

E23 · 2021-10-27: APPEAL_STAY_REQUESTED
E24 · 2021-11-14: APPEAL_STAY_GRANTED

Source resolution superseded · E25, E27

On 3 February 2022 the source resolution ceased, and appeal judgment 50671 recorded VARIED with source effect SUPERSEDED. The summary field named CurrentSourceResolutionId still points to 2536; its lifecycle is SUPERSEDED.

E25 · 2022-02-03: RESOLUTION_CEASED
E27 · 2022-02-03: RESOLUTION_RECORDED · VARIED / SUPERSEDED

Replacement obligation not established · E27, E28

The appeal judgment is CURRENT and records a KES 2,125.15 cost remedy. AppealObligationCount is 0: a remedy is not an explicit obligation. Its summary compliance code is FULLY_COMPLIED, but no appeal obligation/effect rows are supplied. Do not invent a new amount owed or transfer old payments to the appeal.

E27 · 2022-02-03: RESOLUTION_RECORDED · VARIED / SUPERSEDED
E28 · 2022-02-03: REMEDY_RECORDED · KES 2,125.15

Explore the interactive records →Download source JSON ↓

This is a selected synthetic case, not legal advice or a measured AI result. The full-extract lifecycle labels are not automatically safe as pre-outcome features.

Reproduce this result

Your first result, locally.

Download and extract the examples. Open a terminal in the extracted folder and run:

python run.py --example law

Python 3 · no extra packages · no account or API key.

Download examples + runnable code ↓

Includes the original sample files, script, expected results and source references. The script writes results.json and checks it against the packaged reference.

Inspect the expected output →

These are deterministic record checks. No AI model has been run or scored by this example.

Need data for your own question?

Tell us the cities, period, cohort, fields and intended use. We’ll confirm the available scope and quote.

Request data for this question →

AI grounding · synthetic sample

What can a care summary say—and which records support it?

Nairobi · synthetic citizen 33793 · visit 14762260

A summary supported by the supplied visit

Visit 14762260 records an inpatient admission on 1 July 2025 at Ngando Health Facility 8 and discharge on 5 July 2025.

The same visit row records diagnosis code J45.9 and Salbutamol Inhaler. These fields alone do not establish the medication’s clinical indication.

Inspect the supporting fields

Encounter dates and location · Visit 14762260

VisitId: 14762260
VisitDate: 2025-07-01
VisitType: IPD
VisitFacilityName: Ngando Health Facility 8
AdmissionDate: 2025-07-01
DischargeDate: 2025-07-05

Diagnosis and medication · values, not an inferred link

DiagnosisICD10Code: J45.9
MedicationName: Salbutamol Inhaler

The example preserves the original code. It does not infer clinical appropriateness or a diagnosis-to-medication link.

Keep the time boundary explicit

The discharge date is a later outcome relative to admission. The patient’s snapshot-registered facility is separate from the facility of this visit. Gaps between the 32 supplied visits do not establish an absence of care.

Supplied visits: 32
First supplied visit: 1986-05-23
Last supplied visit: 2026-03-04

Explore the interactive records →Download source JSON ↓

This curated history does not establish population patterns or clinical correctness. A diagnosis and medicine appearing together do not establish why the medicine was prescribed.

Reproduce this result

Your first result, locally.

Download and extract the examples. Open a terminal in the extracted folder and run:

python run.py --example health

Python 3 · no extra packages · no account or API key.

Download examples + runnable code ↓

Includes the original sample files, script, expected results and source references. The script writes results.json and checks it against the packaged reference.

Inspect the expected output →

These are deterministic record checks. No AI model has been run or scored by this example.

Need data for your own question?

Tell us the cities, period, cohort, fields and intended use. We’ll confirm the available scope and quote.

Request data for this question →

Research workflow · synthetic sample

How do one resident’s education and employment records line up over time?

Nairobi · synthetic resident 61 · Kiplangat Kipchoge

One resident. Two connected histories.

The supplied profile contains six school enrollment records, dated 1960–1967, followed by a recorded Sales Assistant contract from 1996 to 2017.

Education and work records for the intervening years are not supplied. This gap does not establish unemployment or the date of graduation.

Four reference points

First and last supplied enrollment, then contract start and end. Expand below for all eight dated entries from seven source records; the contract contributes both a start and an end.

5 Jan 1960
Education
Enrollment recorded
Class 5 · Westlands Covenant Primary School
Inspect source record
Record
61:education:0
Date field
/education/0/enrollmentDate
Original value
1960-01-05
Open the original synthetic profile →
5 Jan 1967
Education
Enrollment recorded
Form 5 · Huruma Faith Boys High School
Inspect source record
Record
61:education:5
Date field
/education/5/enrollmentDate
Original value
1967-01-05
Open the original synthetic profile →
1 Jan 1996
Employment
Contract started
Sales Assistant · Nakumatt Holdings Limited
Inspect source record
Record
61:contract:0
Date field
/employment/contracts/0/startDate
Original value
1996-01-01
Open the original synthetic profile →
3 Oct 2017
Employment
Contract ended
Sales Assistant · Nakumatt Holdings Limited
Inspect source record
Record
61:contract:0
Date field
/employment/contracts/0/endDate
Original value
2017-10-03
Open the original synthetic profile →

Full chronology · 8 dated entries

5 Jan 1960
Enrollment recorded
Class 5 · Westlands Covenant Primary School
61:education:0 · enrollment_date
5 Jan 1962
Enrollment recorded
Class 7 · Kibera Government School
61:education:1 · enrollment_date
5 Jan 1963
Enrollment recorded
Form 1 · Karen Model Boys High School
61:education:2 · enrollment_date
5 Jan 1965
Enrollment recorded
Form 3 · Gikomba Community Boys High School
61:education:3 · enrollment_date
5 Jan 1966
Enrollment recorded
Form 4 · Huruma Faith Boys High School
61:education:4 · enrollment_date
5 Jan 1967
Enrollment recorded
Form 5 · Huruma Faith Boys High School
61:education:5 · enrollment_date
1 Jan 1996
Contract started
Sales Assistant · Nakumatt Holdings Limited
61:contract:0 · start_date
3 Oct 2017
Contract ended
Sales Assistant · Nakumatt Holdings Limited
61:contract:0 · end_date

Download timeline CSV ↓Download timeline JSON ↓

How this was reconstructed

Match education and contract rows on citizen_id = 61, retain each supplied date, then sort the events chronologically. Blank exit dates remain unknown. An enrollment is not a graduation, and the supplied contract is not necessarily the resident’s first job.

Resident: 61 · Kiplangat Kipchoge
School records: 6
Employment contracts: 1
Dated entries: 8
Original source: source/61.json

Explore the wider connected sample

32 residents connect to 161 enrollment records and 32 employment contracts. Count each related table separately before joining to avoid multiplying rows. This selected teaching sample does not establish population-level education or earnings effects.

residents.csv: 32
education.csv: 161
contracts.csv: 32
visits.csv: 288

Explore the interactive records →Download source JSON ↓

The wider download includes 32 selected residents. This example describes resident 61’s supplied records; it does not estimate citywide education or employment outcomes.

Reproduce this result

Your first result, locally.

Download and extract the examples. Open a terminal in the extracted folder and run:

python run.py --example pathways

Python 3 · no extra packages · no account or API key.

Download examples + runnable code ↓

Includes the original sample files, script, expected results and source references. The script writes results.json and checks it against the packaged reference.

Inspect the expected output →

These are deterministic record checks. No AI model has been run or scored by this example.

Need data for your own question?

Tell us the cities, period, cohort, fields and intended use. We’ll confirm the available scope and quote.

Request data for this question →

Make your evaluation repeatable

Test your own workflow against these records.

Use nine public reference questions across Law, Health and Education/work. Keep the inputs and scoring rules fixed, record your model and prompt, and inspect the answers against the source.

Download examples + evaluation kit ↓

What the kit checks

The local scorer checks exact structured values and required source locators. It reports every question, including omissions. Prose quality and whether citations support every claim need human review.

The download includes the questions, input records, answer key, blank run template, scorer and a pilot-report template. These are public reference exercises; no model score or customer outcome is claimed.

python evaluate.py my-run.json

Read the evaluation protocol →Inspect the task inputs →

From a free sample to your own study.

1. Try the records

Download the evaluation sample, inspect its evidence and run the example under the published sample licence.

2. Define your scope

Choose a focused marketplace dataset or broader domain access. Specify cities, cohort, period, fields and intended use.

3. Review the quote

We confirm available records, release evidence, licence, price and delivery. An enquiry does not place an order.

Find a focused dataset →Build a domain package →Open the complete Python + SQL walkthrough →

Start with a question.Follow the evidence.

Which obligations remain operative after an appeal?

Reference answer · as of 3 February 2022

Inspect the supporting events

What can a care summary say—and which records support it?

A summary supported by the supplied visit

Inspect the supporting fields

How do one resident’s education and employment records line up over time?

One resident. Two connected histories.

Four reference points

Enrollment recorded

Enrollment recorded

Contract started

Contract ended

Test your own workflow against these records.

From a free sample to your own study.

1. Try the records

2. Define your scope

3. Review the quote

Start with a question.
Follow the evidence.