GNB PRISMA OPPO™

Built to find it
before the story does.

PRISMA OPPO is GNB’s opposition research engine — it cross-references public filings (FEC, Senate EFD, FARA, the Federal Register) into a cited, dated, deterministically-scored dossier, instead of a confident-sounding AI paragraph with nothing underneath it. Every line traces back to an actual filing or article. Every claim carries a freshness verdict. When there’s no documented link, it says so — out loud.

Public RecordCross-Reference Cross-ReferenceCited Dossier

Proof, from the public record

Four real, already-public cases where PRISMA OPPO’s cross-reference method would have surfaced the finding well before it broke nationally. Subjects are redacted here — the underlying filings, dates, and outcomes are not.

Sen. ████████████
R–NC · Chair, Senate Intelligence Committee
COVID Insider Trading
3 days ahead of the story

PRISMA surfaces the trade the day the STOCK Act disclosure becomes searchable on Senate EFD — 3 days before the exposé reached the public.

Rep. ██████████
R–NY · NY-3 Representative-elect
Campaign Finance Fraud
65 days ahead of the story

PRISMA’s FEC cross-reference flags the shell-LLC loans the day Q3 filings go live — two months before the exposé revealed a fabricated candidacy.

Sen. ████████████
D–NJ · Chair, Senate Foreign Relations Committee
Foreign Bribery
10 months ahead of the story

PRISMA cross-references a cluster of NJ businessmen donors — all with Egypt-linked interests — against a USDA regulatory monopoly benefiting one donor’s company. The pattern surfaces 10 months before the case went public.

██████████
D–DE · Vice President, United States (2009–2017)
Family Foreign-Income Conflict
4 years ahead of the story

A family member’s foreign-company board seat was public for years before it became a national controversy. PRISMA cross-references the VP’s policy authority against that firm’s lobbying footprint — surfacing the conflict four years ahead.

Why most AI oppo research fails

Four real, demonstrated failure modes of AI-generated research — and how PRISMA OPPO is built to not make them.

Failure Mode · Staleness

Stale intelligence, presented as current

A real fact-check on an AI-generated vulnerability synthesis blended two current developments with two already-resolved ones and one unverifiable claim — all delivered with identical confidence and no dates attached.

PRISMA’s answer: a visible freshness badge — Current / Recent / Stale — on every dated claim. Nothing reaches the principal without a date and a staleness verdict beside it.

Failure Mode · Hallucination

Confident hallucination on relationships

A typical AI connection-finder will manufacture a plausible-sounding link between two people rather than admit none exists — because a confident answer is what the model is rewarded for.

PRISMA’s answer: Bridge explicitly returns “no documented connection found” when that’s the truth. An honest null is a trust feature, not a failure.

Failure Mode · Self-Contradiction

Scores that contradict their own labels

A real bug: a loyalty score of 42 was labeled “Critical” while a 68 on the same page read “High” — because the label was left to model discretion, decoupled from the number.

PRISMA’s answer: risk and status labels are computed deterministically from the score, in code. A lower number can never look safer than a higher one.

Failure Mode · Brittle Tooling

Brittle search on partial input

A government legislator-search API political staff rely on daily fails on partial name entry — a small but real, trust-eroding gap under deadline pressure.

PRISMA’s answer: a local, resilient prefix and multi-word search index — offline, with no external dependency, built the way staff actually type.

The engine

Public record in, a cited and scored dossier out — no step in between left to model guesswork.

Public Record
  • FEC filings
  • Senate EFD
  • FARA registry
  • Federal Register
  • GDELT tone
  • Dated press
Cross-Reference Engine
  • Donor clusters × regulatory actions
  • Family payments × policy authority
  • Deterministic scoring
Output
  • Sourced dossier
  • Deterministic score
  • Cited attack lines
  • Freshness-stamped

The arsenal

Six capabilities, each built to answer one of the failure modes above.

Capability 01 · Graph
Network
The relationship graph nobody else builds.

An interactive, force-directed relationship graph of the donors, PACs, organizations, and media tied to a subject — built from real public-record data. Every edge traces back to a real filing.

Capability 02 · Path
Bridge
The documented path — or an honest dead end.

Given any two people, Bridge finds the real documented connecting path through donors, orgs, and committees — or honestly reports that no documented link exists. Speculative links are shown dashed, never asserted as fact.

Capability 03 · Coalition
Constituency
A coalition mapped into blocs that can’t lie about themselves.

Maps a subject’s coalition into named voter and stakeholder blocs — each with a loyalty score, trend direction, and why they move. Risk tiers are computed deterministically, so a score can never contradict its own label.

Capability 04 · Moment
Zeitgeist
Alignment with the moment, dated to the day.

Tracks a subject’s alignment with the current cultural and political moment across a real 12-month trajectory, grounded in real media-tone data, with a freshness indicator on every claim.

Capability 05 · Forecast
Prediction
Getting ahead of the story, productized.

A 30/60/90-day forecast of where the moment is heading versus a subject’s current messaging — a deterministic statistical extrapolation of real historical data, not model imagination.

Capability 06 · Read
Pulse
A continuous read on where the coalition is soft.

A continuously-updated read on constituency psychology — which messages are landing, which are backfiring, and where an opponent’s coalition has an opening worth pressing.

The deliverable

Same engine, two outputs: research an opponent, or audit your own candidate before someone else does.

6 minto first draft Dozensof sources cited Zerounsourced claims
Sample Output · Illustrative Subject · Offense
Oppo Dossier — J. Doe (D)
Opponent · State Senate · District 14
71 High Vulnerability
FECUndisclosed LLC donor cluster tied to a zoning vote three weeks prior.
FARASpouse’s consulting firm registered as a foreign agent in 2021.
VOTE RECVoted against a bill she publicly co-sponsored eight months earlier.

Every line above is dated, sourced, and clickable back to the public record.

The field

Competitor names below are anonymized; the capability gaps are real and drawn from each product’s own public claims.

Competitor 1
Live social listening

Live monitoring across a dozen social platforms with real-time alerts and a post-to-attack-ad pipeline in minutes. Adds synthetic polling and a built-in CRM — but its attack-ad scripts ship with no citation on any claim, no relationship graph, and no staleness indicator.

Competitor 2
Narrative intelligence

Real-time alerts and in-platform drafting of “sourced” talking points — with no per-claim citation actually attached, no donor or regulatory analysis, and no deterministic scoring on real historical data.

Competitor 3
Vulnerability discovery

Automated, tier-rated vulnerability discovery with messaging angles generated in seconds — but a tier rating with nothing to click through to, no fact-check step, and no relationship or coalition mapping.

Competitor 4
Rapid content

AI rapid-response video and audio generation, alert-to-clip in roughly 2–4 hours — where a false claim moves through the pipeline exactly as fast as a true one, with nothing to cross-reference it against first.

Competitor 5
Detection only

Cross-platform bot and deepfake detection — strong at surfacing coordinated activity, but produces no dossier, no score, and no attack lines on any named subject.

Competitor 6
GOTV / outreach

Outreach and content generation at scale — a volume-and-reach play with no per-claim verification layer and no monitoring or research claim at all.

Competitor 7
Content copilot

A pure content-generation copilot for fundraising and campaign copy — it writes, it doesn’t investigate, and has no verified source of its own to write from.

Competitor 8
Content copilot

Same category as Competitor 7 — ad-copy generation only, with no verification, scoring, or relationship analysis of any kind behind it.

CapabilityThe FieldPRISMA OPPO
Live multi-platform social monitoring
Rapid content generation
AI hallucination safeguards
Citation-level verifiability
Staleness detection
Honest null results
Relationship-graph analysis
Coalition mapping
Trend forecasting on real history

Live multi-platform social monitoring is the one cell we honestly leave unlit — PRISMA OPPO doesn’t do it, and we say so.

The Core Differentiator

Fast and wrong beats no one.

The Field

  • AI writes the narrative, the score, the pitch
  • Can hallucinate a quote, a link, a record
  • Zero citation trail to catch it
  • Ships fast, unchecked

PRISMA OPPO

  • Every claim traces to a source
  • Fabrications get caught before they ship
  • Honest “no connection found” beats a made-up one
  • Verified, then fast

A wrong attack launched fast is worse than a verified one surfaced with rigor — depth beats scrape speed when the principal’s name is on it.

The coverage map

Live in the full app

A 50-state-plus-DC board of where every tracked race stands, repainted continuously as the public record moves. Click any tile to zoom in.

50 statesone board Repaintsas the record moves Zoomto any state, instantly
Vulnerable — exposed right now; a live, actionable finding sits in the public record.
Borderline — one filing or one news cycle away from red. Worth a standing watch.
Stale — quiet today; nothing new since the last scan. Can flip fast.

Click any state to zoom in.

LIVE CAPABILITY · SAMPLE COVERAGE MAP — this board updates continuously from the public record inside the full app. Statuses shown on this public preview are illustrative, not a live read on any current race.

Days to yearsahead of the story Every claimcited Zerounsourced attack lines

Give us one race and half an hour, two days, or one week — whatever your timeline demands — and we’ll hand you a dossier you can put your name on: every claim dated, cited, and clickable back to the public record.

Request a Confidential Briefing →
By visiting this site you acknowledge our and . This site uses no tracking cookies.