Datasets

Products and open releases, clearly separated.

Commercial data products carry explicit access and permitted-use terms. Open research releases keep their individual licences, provenance, validation and limitations.

Commercial data products

Licensed for operational use.

Availability, territory, update cadence, delivery and rights are set at product and contract level.

Featured product · SAI-01

Seismic Asset Intelligence

Pilot

Asset-level earthquake exposure, event impact and damage intelligence for insurers, corporates, governments and investors.

Coverage
Initial design-partner territories
Cadence
Release-based · scope agreed per pilot
Sample
Evaluation sample available on request
Access
Design partners · commercial evaluation
Data APIResolve APIModel APIBulk export
Explore product → Discuss access
Open research releases

Methods and evidence in public.

Each release states its own licence. Open status does not extend to commercial products.

AvailableCC-BY 4.0
Safeguardingv1.0

Police Sexual-Offence Outcomes — Structured + Modelled

A structured, modelled derivative of Home Office police-recorded-crime open data: roughly two million source rows (every offence outcome in England and Wales in 2023/24, the 2005/06–2013/14 sanctions series, and Community-Safety-Partnership recorded volumes for the decade to 2024) turned into tidy force-by-offence-by-outcome tables and calibrated statistical models. It quantifies the justice gap (3.5% of recorded rapes charged; ~50% closed by victim withdrawal), the near-fourfold force "postcode lottery" (18 of 44 forces outside funnel control limits), and the drivers of a charge (offence type first, police force second). Aggregate counts only — no personal data; the system's response is the unit, never a person's guilt or a complainant's credibility.

human-validated · provenance documented

2,045,000 records Datasheet + limitationsData ↗
AvailableCC-BY 4.0
Safeguardingv1.0

Safeguarding Institutional-Failure Network Dataset

A coded, anonymised, schema-versioned dataset derived from 11 public statutory reviews and inquiries into child sexual abuse and related deaths. Each case is coded into controlled vocabularies — agencies, ordered decisions, risk indicators, and the record-integrity failures the reviews themselves identify — so cases can be represented as networks and compared. Extraction was LLM-assisted and human-validated; inter-rater agreement, a hallucination audit, and the limitations are reported in the datasheet. It contains no complainant data and supports record-integrity analysis only — never any judgement of an individual's credibility.

AI-assisted · human-validated · provenance documented

11 records Datasheet + limitationsData ↗
Research previewCC-BY 4.0
Safeguardingv1.0

Safeguarding Reviews — Scaled Model-Coded Corpus

A 61-review corpus compiled from the public record (22 publishers, 2014–2026: Child Safeguarding Practice Reviews, Domestic Homicide Reviews, Serious Case Reviews, IICSA volumes, council inquiries, and a Northern Ireland review). Each report was fetched from its official source (provenance manifest with hashes) and coded into the controlled-vocabulary schema by a local large language model (liquid/lfm2.5-1.2b) — so this release is AI-coded and NOT human-validated. We publish only the derived structural codes plus provenance links; the report text is not redistributed (per-document copyright). Read the quality note: the agency structure is sound at scale, but the small model's interpretive (failure) coding is unreliable — use the agency/source structure, not the model-coded flags.

AI-assisted · provenance documented

61 records Datasheet + limitationsData ↗