Districts, precincts, municipalities
Boundaries, registration, turnout, election results, comparison, and field-planning context.
Case study 01 · Statewide political intelligence
A political-intelligence system built to show its work.
A statewide campaign research and strategy workspace that connects Rhode Island public records, elections, geography, voter rolls, candidate dossiers, money, government action, and agent-driven analysis.
- Role
- Sole creator: product, data, engineering, agent direction
- Domain
- Campaign strategy, elections, public records, civic intelligence
- Adoption
- Prepared for use by multiple candidates in upcoming Rhode Island state elections
- Boundary
- Evidence and association never become unsupported identity, intent, or misconduct claims
Statewide scope
Municipalities, districts, precincts, offices, elections, candidates, and public decisions.
Source-native
Records keep their original identifiers, time, grain, citations, and caveats.
Geography first
Maps connect turnout, registration, candidate context, field planning, and place.
Agent legible
Typed tools return evidence, counterevidence, gaps, and deep links into the workspace.
The problem
Campaign questions cross systems that were never designed to meet.
A strategic question about a candidate can touch contribution filings, public statements, legislative votes, ethics matters, office history, municipal decisions, election results, district geography, registration, and turnout. Each source uses a different identifier, time period, record grain, and standard of proof.
The system preserves those differences while making them usable together. A matching name is a candidate link, not a person identity. A contribution is a financial event, not proof of influence. A complaint, finding, dismissal, and unresolved search gap remain different states. The agent inherits those rules from the tools instead of relying on a warning buried in prose.
Product thesis: the most valuable political AI is not the system that produces the strongest-sounding answer. It is the one that makes a complex answer fast to inspect, challenge, and revise.
The project workspace
Map the electorate. Interrogate the record.
The map is a screenshot from the actual Rhode Island project, showing its coastline and municipal boundaries. The analyst example below uses a synthetic candidate to illustrate how an answer keeps evidence and interpretation separate.
Strategy analystRepresentative agent trace · Synthetic candidate
Evidence modeCandidate dossierCampaign moneyEthicsVotesPublic statements
Where does Candidate A's affordability message require closer review against contributions, ethics records, and recorded votes?
The evidence supports a review, not a verdict. Candidate A received reportable contributions connected to the development sector[1], publicly emphasized affordability[2], and cast recorded votes on measures with housing-cost consequences[3]. Those facts may create a message vulnerability, but they do not establish motive or improper influence. The ethics matter must be described by its official disposition rather than flattened into “a violation.” The analysis should test bill text, timing, stated rationale, counterexamples, and the complete contribution denominator before drawing a strategic conclusion.[4]
01 · FilingCampaign-finance record
Source-issued account, transaction, date, amount, and filing locator.
02 · StatementPublic affordability message
Archived artifact, exact quote, speaker, publication date, and context.
03 · VoteRecorded legislative action
Bill version, roll call, officeholder, date, and measure summary.
04 · DispositionOfficial ethics record
Complaint, finding, dismissal, warning, or open gap kept as distinct states.
INTERPRETATION LIMIT · Money, timing, message tension, or a graph path can justify a question. None alone proves influence, intent, misconduct, or causation.
System scope
One workspace, many evidence grains.
The interface connects domains without pretending they are one universal table.
Candidates and officeholders
Temporal roles, races, public positions, official acts, coverage gaps, and candidate-link review.
Campaign finance and public money
Accounts, transactions, donors, payees, filings, contracts, procurement, payments, and outcomes.
Meetings, votes, decisions
Agendas, legislation, votes, recusals, implementation events, statements, and exact source locators.
Ethics, legal matters, counterevidence
Claims, official findings, denials, dismissals, corrections, and unresolved searches kept distinct.
Questions, scenarios, field routes
Precinct intelligence, vote goals, issue salience, candidate comparisons, routing, and evidence-backed briefs.
Architecture
Source truth stays local. Strategy composes above it.
A federated evidence spine makes new products possible without erasing the identifiers, timing, or uncertainty of the original records.
- 01 / Sources
Public records and retained artifacts
Elections, filings, government records, geography, documents, and lawful voter-roll snapshots.
- 02 / Authority
Source-local stores
Immutable bytes, exact keys, temporal scope, manifests, typed locators, and coverage state.
- 03 / Spine
Identity and relation contracts
Exact joins, reviewed correspondences, discovery candidates, and own-grain relations.
- 04 / Tools
Campaign task capabilities
Dossiers, electorate comparison, money paths, timelines, gaps, evidence packets, and routing.
- 05 / Product
Map, console, and agent
A visible workspace where retrieved evidence becomes a map, table, profile, timeline, or cited answer.
Ownership and outcome
From corpus to campaign product.
Arthur originated the product vision and built the research program, data architecture, acquisition and validation workflows, geospatial surfaces, agent and tool contracts, campaign-analysis products, evaluation controls, and operating handoffs. Agent fleets accelerated research and implementation; Arthur set their scope, adjudicated their evidence, and owned the final system.
The system is being prepared for use by multiple candidates in upcoming Rhode Island state elections. That planned adoption is the outcome boundary today: it demonstrates a product moving from a personal research program toward real campaign use without claiming election results, persuasion effects, or field outcomes that have not happened.
A separate demonstration page runs the interaction model against a synthetic state with twelve fictional districts and seven prepared questions, each answer carrying its evidence, its counterevidence, and its interpretation limit. Its records are fictional and its answers are prepared; the project screenshot above shows actual Rhode Island geography.