The Complete Guide to Evidence-Backed Opposition Research for Campaigns

Campaigns, party committees, PACs, and consultants use opposition research to prepare for a race with documents they can show a second person. OppIntell publishes this guide as a field manual for that job. The primary subject is opposition research software for campaigns, and the standard the rest of the page uses is the one already printed on the public site: every claim a reader is asked to repeat should tie back to a public record.

The public homepage introduces OppIntell as opposition research software for campaigns, party committees, PACs, and consultants, with the same profile fields for every candidate: votes, quotes, filings, and contradiction flags, each linked to a public record. This article walks through coverage, databases, public records, a repeatable pass, report contents, and responsible use. It stays inside those published pages.

What campaign opposition research covers

Opposition research, as the public opposition-research page defines it, is the collection and citation of public records about a candidate. Campaigns have kept files on opponents for as long as they have had opponents. What changed is the volume of the public record. Federal committees file with the Federal Election Commission. State candidates file with a secretary of state or a local election office. Personal financial disclosures, legislative votes, and court dockets are published on their own calendars. A modern file has to keep those streams attached to one person, or the campaign is working from a stack of documents nobody has time to re-read on a Tuesday night.

The same page says the work covers Democrats, Republicans, and everyone else on the ballot with the same record types. A file that applies one standard to every candidate in the race is something a campaign, a reporter, or a counsel can use. The short name inside campaign offices is often "oppo." The job under that name is the same: find the documented facts, check that they still hold, and put them where a communications team can use them with a source attached. Rumors, anonymous tips, and unsourced social posts can start a question. The opposition-research page's rule is that they stay out of the file as claims until a public record answers the question.

A campaign file answers what can be said about this candidate from the public record, with a source on each claim. The opposition-research page separates that job from a voter guide and from due diligence. Due diligence, litigation support, and watchdog research use the same filings for a different decision, and OppIntell describes that adjacent work as public-record intelligence.

The work also includes the campaign's own candidate. The opposition-research page calls this the mirror file: which sentences the same public records could support about your candidate, including an old quote against a current vote, a disclosure that omits a later-reported asset, a donor pattern, or a case that only shares a name. The homepage states the same idea. Every candidate is profiled at the same depth, so the database holds both the file on the rest of the ballot and a working copy of the file the other side can build.

Office changes which records exist. The opposition-research page says House, Senate, governor, mayor, legislature, judicial, and local contests draw on different mixes. A federal candidate's money is often in FEC filings. A mayor's filing is often at the city or county. A legislator's vote record exists only if they have served. The homepage browses those families, plus executive and statewide executive. A packet that assumes every office has votes and an FEC report will invent records the person does not have.

Candidate research databases

A candidate research database is the place those streams meet one identity. The methodology page describes the step as candidate-universe assembly. Records from upstream sources are normalized into a single universe. Deduplication is keyed on a combination of legal name, jurisdiction, office, and filing identifiers from the source registry, with a reliability score on each upstream source so a later query can prefer the more authoritative record when two sources disagree. Briefings, candidate profiles, and race pages read from that universe.

OppIntell publishes a canonical coverage figure for that index. The live llms.txt fact line, and the in-repo coverage claim it is built from, says: 24,000 U.S. political candidates as of 2026-10-03, defined as candidate profiles in the public candidate-research index, all parties, federal, state, and local, deduplicated and index-gated. The homepage meta description uses the same rounded claim as "24,000+ U.S. candidates, all parties." This guide treats that sentence, with its date and its definition, as the figure to cite when a single number is required.

The public homepage also renders a live counter from the research summary. On 2026-10-05 that counter read 24,130 total candidate profiles. The party lines on the same page read 8,177 Republican, 8,335 Democratic, 244 Libertarian, 72 Green, 938 Independent, 1,564 Nonpartisan, and 4,800 Other. Those party lines add to the 24,130 total. They are a point-in-time reading of the homepage, not a replacement for the 24,000 figure dated 2026-10-03. A later reader should check the live homepage and llms.txt rather than freeze either number as permanent.

The homepage also shows separate counts on race-family cards, including House, Senate, governors, mayors, and the other families listed above. Those cards are a browsing layout. This guide does not add them together and does not present them as a split of the 24,130 total. Using a race-card sum as "the database" would describe a different cut of the data than the total the page itself prints.

Public entry points are ordinary web pages. State hubs live at /states. Race hubs live at /races. Individual public profiles live under /candidates. The opposition-research page points at those hubs, including a race-and-state example at /races/california/mayor. llms.txt maps the same public pages and says the authenticated app, admin tools, API routes, webhook routes, and SEO operator surfaces are not public research content. The crawl note there says robots.txt excludes /api, /admin, /app, /auth, /profile, and /seo.

llms.txt also documents a Model Context Protocol server at /mcp for search, candidate profiles, controversies, voting records, quotes, issues, and race lists. The free tier it describes is a self-serve, rate-limited, attribution-required key. get_receipt returns provenance for a published claim (content hash, span locator, source URL, capture timestamp, and supersession status) and does not return claim text or the archived source. Receipt pages live at /receipt/{token}. The use-cases page names kinds of work for the wider records, including campaigns, litigation, due diligence, communications, watchdog research, and partner vetting. Those labels are not a customer list.

Public-record research

The methodology page states the intake rule in one line: every research record begins with a public source. It names three production intakes. Federal Election Commission filings cover candidate committees, principal campaign committees, independent expenditures, and Super PAC reports. Secretary of State candidate filings come from 47 states. County-level candidate registrations come from the four county adapters currently in production. Ingestion runs on a schedule. The page's reason is decay: filings change, committees terminate, address records update, and a one-shot scrape goes stale within days.

The opposition-research page separates the documents a desk actually opens. FEC filings identify the committee, the treasurer, the reporting period, and itemized transactions above the disclosure threshold: who gave, when the report was filed, and whether a later amendment changed the totals. A state or local candidate may never appear in an FEC report. Disclosed money for that candidate lives in a state or local system. Personal financial disclosures are a different form. Some offices require a filing that lists income, assets, and liabilities in ranges set by the rules. A sentence that mixes that form with the committee's FEC report cites the wrong document.

Donor lines sit inside those finance filings. The opposition-research page says itemized contributions, PAC support, and recurring employers can be grouped so a reader sees concentration by industry, family, or PAC. That map supports a sentence about the reports. A sentence about why a donor gave, or what was promised in private, needs its own public source. A gift can sit next to a vote. The file states that both documents exist, and stops there.

A vote line belongs when a legislative record exists for that person and that bill. A quote stores the words that were said, with the date and the place. A newspaper summary, a staff recollection, and a clipped caption are descriptions by someone else, and the opposition-research page says to label them that way. A court record belongs when it matches the person. Two candidates in one state can share a surname, and a case caption is not a biography.

On a published briefing, the homepage and the opposition-research page describe the same citation. Each claim carries a source URL and a source type, such as an FEC filing, a Secretary of State portal, a court record, an official campaign site, or a government press release. The methodology page names the check citation_validation_status: whether the document still exists at that URL and still contains the cited language. A claim that fails stays out of the public post.

Where the record supports a direct fact, the methodology page says the briefing states it and cites the record. Where the record only opens a question, the briefing uses posture language such as "may," "could," "would examine," and "appears consistent with," so a reader can separate a documented fact from an open thread.

A repeatable workflow

The opposition-research page prints a working order a desk can run on paper. Profiles already organize several of the steps for votes, statements, finance, court records, and packets ready for communications review.

Start by fixing identity. Write down the legal name, the office sought, the jurisdiction, the cycle, and anyone else who shares the name. Secretary of state filings and FEC committee names are the anchor, because a ballot name and a committee name are often different strings. A file that starts from a nickname or a headline can attach the wrong vote or the wrong donor report. If two candidates in the same state share a surname, the first page should carry a middle name, a city, or a committee ID. Those are the same keys the methodology page uses for deduplication: legal name, jurisdiction, office, and filing identifiers.

Pull the public records that match that identity. For a federal candidate, read receipts, disbursements, and debts, and cite the report that is current after amendments. For a state or local candidate, use the state or local disclosure system and the Secretary of State or county registration the methodology page names. Read votes only where a legislative record exists. Read financial disclosures as their own form. Keep a court record out of the packet until it matches the identity block.

Attach a source before a line leaves the desk: a URL, a source type, and a validation status that the document is still there and still says what the line says. If the link dies or the bill number is wrong, the line comes out until it is fixed. The methodology page says a broken citation flips citation_validation_status and holds the public post back until the source is re-resolved or replaced.

Run that same pass on the campaign's own candidate before the packet is written. The opposition-research page treats a file that covers only the opponent as a miss. The homepage calls the result pre-empting with receipts: response strategy uses the same citations required on an opponent.

Then review. The homepage's spine is Capture, Govern, Analyze, Review, Export. Review names analyst, senior, legal, and communications. Export names receipts, citations, and watermarking. A new FEC report can arrive while quotes are still being matched to votes. Identity stays fixed, and a line waits for a source. A mail piece written first is how a desk learns late that the document does not say what the draft assumed.

Cadence is part of the workflow. The methodology page says blog briefings regenerate hourly, candidate research records refresh daily, and sitemap and indexing signals refresh continuously so a briefing that clears the publish guard can be discovered by crawlers within minutes. When a candidate withdraws, a committee terminates, or a filing is amended, the corresponding research record is reprocessed on the next ingestion run. A file that was accurate in March is a draft again after an amended report.

What belongs in a report

A report is the packet a second person can defend. The opposition-research page's list of what belongs in a working file is the right checklist. Identity and office come first: legal name, office sought, jurisdiction, cycle, and the disambiguators that keep this person separate from a namesake. Then votes and position history where a voting record exists. Then public statements, stored as the words that were said. Then campaign finance filings, financial disclosures, and donor information taken from those filings. Then court records that match the person. Every claim the campaign intends to repeat carries a source.

A useful report also shows posture. Direct sentences cite the record that supports them. Open questions stay in posture language from the methodology page, so a communications writer can see that the record has not closed the point. Contradiction flags, as the homepage and the opposition-research page use the term, mark places where a statement and another public record sit in tension. The flag is an invitation to read both documents. It is not, by itself, a finding about motive.

Receipts are how a line stays checkable after it leaves the desk. llms.txt describes a public receipt page for a single published finding: the claim, a source link, a capture timestamp, a content hash, and supersession history. The free-tier get_receipt tool returns provenance only. A campaign packet can follow the same shape even when it never touches that API: the sentence, the URL, the date the record was read, and a note when a later amendment supersedes it.

Some material stays out. An unsourced social post can be logged as a question and becomes a claim when a public record answers it. A donor's gift stays a reported transaction. A personal financial disclosure stays separate from the committee's FEC totals. The methodology page says anonymous tips and unverifiable sourcing are outside the pipeline, and that records trace to a named public source.

The homepage also lists product tools around the file: an AI research assistant, vulnerability scoring, electability predictions, donor-network mapping, social intelligence, and opposition briefings, plus a side-by-side comparison of campaign finance, voting records, policy positions, and controversy timelines. The comparison block on that page is labeled as an illustrative example, with a pointer to the real comparison tool for any two candidates in the database. This guide reports those labels as homepage copy. It does not add accuracy rates, win-rate results, or examples of a score applied to a named candidate.

Responsible research

Responsible research is the same standard applied in public. The opposition-research page's ethics section says campaigns in the United States collect opposition research from public records as part of contested elections, and that the page is a research description. It is legal advice for no one. A campaign should ask its own counsel before it collects or publishes anything beyond the public record.

The edges the page names are concrete. Recording a conversation, entering a system that is not public, using personal data that a statute treats differently from a campaign filing, or presenting a record so that it implies a fact the record does not contain are separate questions from reading an FEC report. They depend on the state, the method, and the document. They are why the file shows its sources and why a packet goes through review before it is said out loud. The homepage's TrackerOps band points at the same boundary: jurisdiction-aware recording rules and encrypted uploads are described as live in the evidence backend, and the mobile app is labeled shipping next. Recording rules stay a question for counsel.

All-party coverage is part of the standard. The opposition-research page says citation rules apply to every party. The homepage repeats it: Democrat, Republican, or otherwise, the profile shape is the same.

Publication on OppIntell's own site runs through a database-layer publish guard. The methodology page says posts clear that guard and citation validation without a per-post human review. The guard checks content length, FAQ count on cluster pages, source-backed signals on candidate-level briefings, structured data, and the absence of a broken citation or a source whose reliability score has dropped below threshold. When a cited source disappears, the post is held back until the source is re-resolved or replaced. Corrections land on the next regeneration. The longer account is /about/methodology.

What is live, and what the homepage still marks in build

Readers should be able to tell a public page from a workspace surface the homepage has not finished. The collection section says the candidate universe is live today, and that field capture, media processing, and live response are in build for workspaces. TrackerOps mobile, media intelligence, and war room mode are labeled in build. The TrackerOps band says the lawful-recording engine and the encrypted upload pipeline are live in the backend, and that the mobile app is shipping next. The War Room band says tracker and clip data are live in the backend, that analyst, legal, and communications routing is modeled, and that the live dashboard is shipping next.

The intelligence section says a claim and contradiction engine exists in the backend, with the workspace UI coming soon, and it labels narrative intelligence as in build. Team workflow labels the analyst workbench as in build and lists committee workspaces beside it. The Research Assistant band has no "in build" label. The homepage describes it as a copilot for candidates, evidence, money, narrative drift, and contradictions: summarize evidence, find vulnerabilities, draft research questions, and turn one fact into cited next steps. Public summaries come from the candidate universe. Analyst workflows, exports, and receipts stay inside the authenticated app. Those status lines are homepage copy, not customer results. The public index, the methodology page, and the opposition-research page are the surfaces this guide treats as published now.

How to read this against the live site

The pages behind this guide are public: /opposition-research for the definition and ethics section, /about/methodology for ingestion and the publish guard, /public-record-intelligence for due diligence and watchdog use of the same records, /states and /races before a profile under /candidates, and /llms.txt for the coverage sentence dated 2026-10-03. llms.txt points access requests to /public-record-intelligence#contact and /opposition-research#contact. The contact address in the product configuration is [email protected]. The file this guide describes has one identity, public records, a source on every line that ships, the same standard on both sides of the race, and counsel before the work leaves the public record.