Integrity GraphEarly access
$861M flagged for expert review. No model in the loop.
Integrity Graph joins the registers a review depends on into one structured graph, applies your rules, and flags related-party and governance risk with the evidence attached. On Alberta public records it joined fourteen datasets, ran 56 rules and flagged $861M across 16 individuals in 48 hours, with zero language-model calls at runtime.
BoundaryBuilt for the Government of Alberta national AI hackathon on public records. It was not a Government of Alberta customer deployment. Flags were prompts for expert review, not findings of wrongdoing.
The run
Published separately. Judged together.
The risk that matters is rarely inside any one dataset. It is the relationship between two of them. The rules judged them together.
- 01
Join the record
Registry filings, funding records and grant agreements, published separately and in different words, became one record per real person and organisation.
- 02
Derive what no source holds
Approved rules derived the connections that exist in no source database: who directs what, who funds whom, which parties are related.
- 03
Test every connection
Every derived connection was tested against the encoded risk conditions. Every record, every rule, every run. No sampling.
- 04
Flag with the filing trail
Each flag carries its derivation back to the source line, so a reviewer starts from evidence. A flag is a prompt for a person to look, never an allegation.
The rule hierarchy
A map is not a decision.
Most graph tools stop at the map. Integrity Graph carries the rules. Findings build in levels, and every level keeps the logic and evidence that produced it, so a reviewer can walk a flag back to its filing.
Direct signals. Read a fact from a filing or a public record.
Structural patterns. Test relationships across entities.
Composite findings. Combine supported signals.
Final outcomes. Route a supported finding for review.
See it, live
The demonstration is public and queryable.
Open the rules engine on the public-record demonstration. Flags are grouped by pattern: related parties, cross-domain funding and revolving-door connections. Every flag opens to its rule trace and evidence chain. Names come from public filings and can be viewed anonymised.
Open the live demonstrationEarly access
Your records, over the public graph.
Public records were the demonstration. Your records are the deployment. Early access is a cohort of five partners whose reviews depend on knowing who is related to whom.
What you bring
One register or record set your review depends on: corporate registry, charity and grant filings, procurement and contracts, or your own conflicts data.
What we build
The join, rules written with your team, a weekly refresh as registers update, and every flag with its evidence path.
How it is priced
Engagements start with a fixed-fee diagnosis sprint. Production runs on a monthly base with a metered volume of settled decisions.
The same layer runs underneath every Taskd workflow. See the Taskd platform
Bring one set of records that has to be defended.
We show you what the rules find, where the evidence leads, and what stays with a person to decide.
Book a demo