Skip to main content
STRIPED DONKEYWe do the heavy lifting.

Artificial Intelligence · AI-20

Grounding & Hallucination Evaluation

Evaluate unsupported claims, missing-evidence handling and identity consistency across connected synthetic life records.

  • Synthetic records
  • Longitudinal benchmark
  • Nairobi + Lagos

Scope & purpose

Connected scope

Use connected resident histories to assess factual support, unsupported inference and answer discipline.

  • Demographics
  • Health
  • Employment
  • Households

Assess identity, dates and factual support against the supplied records. Agree evaluation references before delivery.

Possible uses

Task prototyping · Evidence tracing · Error analysis · Controlled evaluation

Who it is for

AI safety teams · LLM evaluators · research labs · red-team groups

Preview type
Illustrative format preview
Shows example structure and fields. It is not a sample extracted from the paid release.
Commercial availability
Confirmed in your written quote
The listed cities describe catalogue coverage. We confirm the exact release and available extract before agreement.
Starting-price basis
USD per selected city
Cohort size and period are specified in the quote; the listed rate does not define a fixed-size package. Discount and tax are shown separately.
Your agreed delivery
Scope and supporting artifacts listed in the quote
Confirm entities or rows, observation period, fields, formats, documentation, evidence, known limits and usage rights.

Paid dataset delivery

What you receive

SD delivery standard
Published cities
Nairobi + Lagos
Record scope
Resident + linked events
Structure
Longitudinal benchmark

Striped Donkey delivery package

  • data/Agreed extract and formats
  • documentation/Schema, fields and relationships
  • evidence/Agreed release checks and limits
  • examples/Agreed loading and join examples
  • LICENSEApplicable rights
  • MANIFESTRelease and file inventory
  • CHECKSUMSSHA-256 fingerprints
  • DELIVERY_NOTEScope and handoff record

Your written quote confirms selected cities, period, formats, commercial rights, detailed artifacts and secure delivery. The complete paid package is prepared privately after agreement.

Free evaluation preview

Inspect before you request

15 illustrative rows · 11 fields · CSV

Download preview CSV

Illustrative rows show the format. Your quote confirms delivered records, fields and time coverage.

Download all preview rows. City selections apply to your data request.

Loading preview…

Fields and record structure

Listed features

Resident + linked events · Longitudinal benchmark

  • age / life stage
  • household relationships
  • dated demographic events
  • dated encounters
  • care history
  • health-state events
  • employment spells

Study question

Evaluation design

Can an AI distinguish supported, contradicted and unanswerable claims across connected synthetic life records while preserving identity and temporal context?

Build reviewed claim labels with missing and conflicting evidence cases; report unsupported-claim rate, contradiction detection and abstention accuracy by error type.

Explore study questions →

Evidence you can inspect

Source & limitations

Review release-specific methods and published checks. Preview rows show format and scope; they do not certify a dataset.

Supporting references

Request detailed artifacts →

Clear rights for your work

Licence & delivery

Free preview

Use free samples under the public-use policy or their dataset-specific licence. Attribution is required for publication and research outputs.

Free sample policy →

Purchased dataset

Your agreement defines the release, licensed organization, permitted uses and duration. Commercial model training, external applications and redistribution require explicit agreement.

Data licensing agreement →
Delivery & purchase terms →