04 / Data visualization · Public records
Carny Connected
A complex public record. A clearer way through it. An interactive explorer connects public disclosures with dated corporate records, keeping the evidence close to every connection.
60-second summary
A dense Miro board made relationships easy to see, but difficult to interpret. Historical roles, proposed transactions and ownership claims could appear equivalent.
Research · Evidence modeling · Interaction design · Responsive web
Verify the evidence before drawing the connection, then offer spatial, chronological and documentary ways to follow the same record.
A responsive explorer with Explore, Timeline and Records, contextual search, source details and a documented exclusion register. Developed independently with Codex; research and implementation were AI-assisted.

Actual frozen v1 screenshot. Circles identify records; their size does not represent investment value. Select any screenshot to view it full-size.
Decisions and system thinking
Make a connection interesting to follow—and specific enough to question
The Miro board supplied research leads. Original documents established what each connection could say. The design then made those distinctions easier to inspect.
Let the source set the visual rules.
A historical role is different from a direct financial interest. An agreement is different from a completed acquisition. Each included relationship retains its source, locator, date and qualifications. Unsupported claims stay out; narrower verified context can appear while an unresolved ownership claim remains withheld.
Follow the record without starting over.
Explore offers a contained neighborhood of six entries, with eight on wide canvases. Timeline separates events from document dates; Records separates disclosure entries from supporting documents. Selection and the exact relationship travel between views. Breadcrumbs, visible paging and contextual search help readers recover their place.
Keep the evidence within reach.
Hover and keyboard focus preview a connection; selection pins it with a solid ring and stronger line. Smooth motion supports orientation, with a reduced-motion control. On mobile, a dedicated detail view replaces the side panel. Portraits and matched logos aid recognition without becoming proof of ownership.





Exploration workflow
Follow a question back to its source
- 01Choose a circle or use contextual search suggestions
- 02Read the relationship, its date and its limits
- 03Explore documented connections one neighborhood at a time
- 04Switch to Timeline or Records with the selection retained
- 05Open the cited source, or inspect the remaining-gap register
Evidence, scope and remaining gaps
A frozen first version. An open research boundary.
The frozen corpus contains 679 documentary entry rows, 1,243 sourced relationships and 212 primary graph sources. These are record counts, not unique companies, current investments or wealth totals. The primary disclosure is dated July 10, 2025. Screening inclusion and historical roles do not establish current personal ownership. Annex C states no managing role and no direct financial interest; third-party-selected managed securities were divested into a blind trust.
For example, the 49% Cameco and 51% Brookfield investor-group interests describe the November 7, 2023 Westinghouse closing snapshot. The collective group share is not allocated among participants or attributed personally to Carney. Saved originals, readable extracts and publisher links retain their different labels. Two additional hosted extracts reproduce the exact cited PDF pages; larger originals remain linked at their publishers and preserved in the unchanged local archive.
55 lead groups remain withheld: 53 await original evidence, one has a user-excluded access route, and one is excluded on primary counterevidence. Riverside’s exact Parkwood investee, EverWind’s upstream parents and National Realty’s full current shareholder structure remain unresolved. Isle of Man registration and payment remain excluded. The bounded Miro audit deferred 1,997 visual candidates; separately, 3,805 saved labelled objects received no new visual disposition. These are different units, not missing-investment counts. Full-board verification remains incomplete.
The final interface review recorded 34 passing checks across 37 observed states and six browser widths, from 320 to 1600px. These checks establish specific interface behavior, not faster comprehension. No user study, adoption, conversion or measured impact is claimed. The next validation would ask readers to find a source, distinguish an agreement from a closing and return to a selected record after following a deeper branch.
Design contribution
What I contributed across the product workflow
- Original-source research and evidence modeling, assisted by Codex
- Relationship types, date scope, source locators and exclusion tracking
- Connected Explore, Timeline and Records workflows
- Contextual autosuggest and keyboard navigation
- Preview, pinned selection and recoverable map navigation
- Responsive evidence panels and reduced-motion behavior
- Manually matched identity assets and visible attribution
- Frozen v1 packaging, integrity checks and documented research limits
Explore more public datasets through interactive visual experiences.