Opposition research delivered as a structured, fully-sourced report — every claim one click from the document that proves it, every attack line shipped with the rebuttal it will get and the answer to that rebuttal. What costs traditional firms $30,000 and three months, Civly delivers in days.
A Research Book has seventeen sections, with every claim linked to its source and every finding graded. Click a section in the contents to read it.
Illustrative
Facts are stated neutrally in the body of the report. Persuasive framing is confined to the Attack Lines section — so nobody on your team has to guess which sentences are evidence and which are argument.
Guns split her from her own county. The county board — four of its five members from her own party — formally urged the governor to veto the school-carry bill, citing the shooting one county over by name. She was a co-sponsor. (See Vulnerability 3, Strong, High severity.)
Primary-voter appetite. No district-level polling exists for this seat, and the book says so. The closest hard data point is behavioral: the referendum she backed lost 57–43 statewide, and her own district held her under 53%.
More than a hundred distinct data sources run against every subject. The book names all of them and reports what each one returned, so you can tell a true null from a coverage gap.
A blank space in someone else’s report could mean nothing is there — or that nobody looked. This section is the difference, and it is why a clean subject still produces a useful book.
| # | Vulnerability | Severity | Confidence | Recommended use |
|---|---|---|---|---|
| 1 | Voted for the school-carry bill her own county board asked the governor to veto | High | Strong | Earned media and mail in the two suburban townships |
| 2 | Backed the constitutional referendum that failed 57–43, and has never revisited it | Medium | Strong | General-electorate contrast; pairs with 1 |
| 3 | About half of a $1.8M lifetime raise is capitol in-kind money | Low | Moderate | Texture under 1–2, not a standalone lead |
Source quality is named for every row, and “recommended use” is a real column. Some findings are television. Some are a press-conference question. Some are only a direction for more research — and the book tells you which is which instead of leaving your team to guess.
| Year | Event |
|---|---|
| 2019 | Wins the open seat with 58.4% in a district her party had held for two decades. |
| 2021 | One of three members of her caucus to vote against the emergency-powers bill — a genuine party break, and her likely defense. |
| 2022 | County board urges a veto of the school-carry bill, citing the shooting one county over; she is a co-sponsor. Re-elected with 52.9%. |
| 2023 | Votes for the constitutional referendum that fails statewide, 57–43. |
| 2025 | Votes for the budget carrying the stadium subsidy; margin falls to 51.6%. |
The timeline is assembled from roll calls and filings — so the sequence still holds when the other side challenges it.
The baseline nobody can afford to get wrong. Party-registration history is checked for switches. Credential claims are verified against the issuing institution and third-party records, and a discrepancy is reported as a discrepancy.
Anything genuinely unavailable is written as “not publicly available” rather than quietly omitted, and where a fact rests on an aggregator rather than a primary record, the sentence says so.
The bio also maps the people in the subject’s orbit — family, business partners, and anyone documented as a co-officer or co-director on public filings — because the useful finding is often one relationship away from the subject.
She was one of three members of her caucus to vote against the emergency-powers bill, and she will run on it. Against that record, she voted to table 73 of 91 minority floor amendments across two sessions.
Committee chairmanships, signature legislation, scorecard ratings, and ethics-filing status all live in this section, so the defensive picture sits on the same page as the offensive one.
Three cycles of certified results with the margin trend. Lifetime and cycle fundraising as the committees filed it, with the party in-kind share broken out by entity and every breakdown (donor size, in-state and in-district money, industry, ActBlue and WinRed share, top vendors) computed over all contributions. A notable-donor table with a flag on each — industry interest, leadership money, cross-cutting labor giving — and a plain note where an association is worth watching but a conflict is not established.
Every opponent gets a name, a party, a summary of how they are positioning, and a link to their campaign site. In states that publish a 2026 candidate list, the field comes from the election office’s own list, with filing dates and withdrawals cited to it. TV and radio ad buys are read from the stations’ FCC political files, with booked and invoiced money kept apart. In a federal race, each opponent’s filed receipts, spending, cash on hand, debts and self-funding sit beside the subject’s, with the same breakdowns and a citation on every figure. Historical challengers are included when their attack lines are still live.
Where Civly computed a figure from public filings, the inline citation points at the public source and the arithmetic goes in the appendix.
In a district where a mass shooting one county over remains a live civic wound, the incumbent’s position put her opposite her own party’s local establishment. The county board — four of its five members from her party — formally urged the governor to veto the school-carry bill, citing that shooting by name. She was a co-sponsor.
Context and mitigating factors. One caution before deployment: she chairs the chamber’s children-and-families committee and will reframe the bill as school safety. The attack is strongest when it stays on the documented facts — the ally split, the veto letter, and her own name on the bill.
Every attack line ships with the rebuttal it will draw, an honest read on how strong that rebuttal is, and the answer to it. Nobody on your team should hear the opponent’s best defense for the first time on a debate stage.
Before the book ships, Civly adversarially re-reads the sources the book itself cites — asking “what in here helps the subject?” Unlike a speculative rebuttal, every item below is a verbatim passage from a source we already used.
…the sponsors repeatedly invoked self-defense, but that’s a bit of a misnomer…
The article we cited as evidence for this framing explains the framing away. The finding was pulled from the attack lines and rebuilt around her roll call instead — before delivery, not after a reporter noticed.
…the idea came from a group of recent graduates as an extracurricular project…
Her defense can credibly call the bill constituent-driven. Better that you read it here than discover it in the mail piece written against you.
…the measure isn’t explicitly meant to address a school shooting…
The link to the shooting is inferential. Argue it — don’t assert it. The book marks the difference in the sentence itself.
Where a cited source has no full text captured for machine review, the book marks it “not machine-checked”.
A finding that fails the two-source rule doesn’t quietly disappear. It lands here, so your team can see it, decide whether it’s worth chasing, and know exactly why it isn’t in the deployable section.
When this section is empty, the book says it is empty as of the research cutoff — an affirmative result.
The subject plus their documented orbit — donation recipients, business and vendor associates, co-officers pulled from nonprofit filings, corporate registries and federal records, and family — screened against exclusion and enforcement lists and large public document corpora.
A clean screen is written up as a result: the book states how many names were screened and against what.
And a name match is described for exactly what it is — a potential same-name reference that must be identity-verified before anyone asserts it. Presence in a document corpus is never treated as evidence of wrongdoing.
Deletion candidates are identified across the archived post corpus and recovered through web archives, with the archive capture linked for every recovered post so you can see the original yourself.
The book reports the pattern honestly. Most deletions on most accounts are typo cleanup, and when that’s what the evidence shows, that’s what it says — with counts for what was recovered, what is no longer retrievable, and what remains unverified. Unrecoverable posts are never speculated about.
A claim based on deleted content cannot appear as a vulnerability without an archive URL, a quality outlet quoting it verbatim, or a timestamped capture. Paraphrase from memory is not evidence.
| Section | What was searched | What came back | Why a human should look |
|---|---|---|---|
| Campaign finance | The subject’s personal federal giving | A donor record with a matching name — employer and city don’t match the subject’s profile | Do not attribute without identity confirmation. A false attribution here is a material error. |
| Polling | District and issue polling for this seat | Nothing returned | Commission or locate district polling before finalizing the message hierarchy. |
| Voting record | A co-sponsorship claim on a second bill | Sourced only to a partisan scorecard | Verify against the official sponsor list before any headline use. |
This is the section that keeps a book honest. Null findings, identity-uncertain matches, and anything resting on a single partisan source route here — never into a numbered vulnerability or an attack line.
Government records and primary documents first, then local and national reporting, then aggregators and partisan outlets flagged for exactly what they are. The full bibliography ships as its own document alongside the book.
Before delivery, every URL in the book is machine-checked. Dead links, a chamber homepage standing in for a specific roll call, and display text that doesn’t match its destination are all treated as blockers — not as nits to fix later.
Any figure Civly computed itself — a fundraising aggregate, a margin swing, a floor-vote pattern, a deleted-post recovery rate — appears in an appendix table alongside the public source it was built from.
You should be able to reproduce our numbers. So the numbers arrive with their inputs, and the inline citation points at the public record.
Research Books are delivered in days. Every claim is linked to a source, with analyst-reviewed findings and material your campaign can use.
Name the subject — a single opponent or a full slate — and the angle: an opposition book, a defensive book on your own candidate, or a neutral due-diligence profile.
More than a hundred data sources run against the subject and their documented orbit: roll calls, hearing transcripts and filings, campaign finance, court and property records, contracts and nonprofits, news across hundreds of outlets including local and Spanish-language press, published fact checks, and the full social archive including deleted posts.
Every claim that will appear in the book is pinned to the document that proves it before any prose is written. A claim that can’t be pinned doesn’t get written.
Our research team reviews each finding against the qualification gates, grades it for confidence and severity, then adversarially re-reads our own sources for anything that helps the subject.
The book ships with attack lines, rebuttals and counter-rebuttals, appendices, and a full bibliography — then gets refreshed on demand as the record grows and the cycle moves. Ask for the Brief and the same research comes back as six pages you can hand to a principal: every line a quotation, every quotation carrying a link to the page it came from, and an excerpt that cannot be found word for word in the source it cites dropped rather than smoothed into place. Send back your reviewer’s marked-up Word file and the book returns revised, each comment researched, fixed and answered beside the edit.
Most research products are graded on how much they found. A Research Book is graded on how much survives — because five airtight attacks beat ten that fall apart on contact. The filtering starts before anyone reads a word: across 29 client news-archive searches, 49% of what a name search returned was a duplicate or a different person. The News Article Cleaner now strips the reprints from any Lexis export you upload before a researcher opens it.
More than 100 data sources run across the subject and their documented orbit.
A fact isn’t a vulnerability unless it harms the campaign — financially, reputationally, with a named voter segment, or as a contrast. An interesting biographical detail is not a finding.
At least two independent Tier 1 or Tier 2 sources on the same fact, or one Tier 1 source plus a primary document — a roll call, a court filing, an official record.
Positions that are mainstream with the electorate you’re actually facing get cut, not counted. A primary and a general are different books.
Absence of evidence is not evidence. “No record found” and “has never said” route to Manual Review — never into an attack line.
What’s left carries a confidence badge, a severity rating, a source-quality note, and a recommended use.
Nothing that fails a gate is deleted. It routes to Rumors & Allegations or Manual Review Required, labeled with the reason — so your team sees the whole picture and makes the call itself.
The checks behind each report: source verification, identity matching, link checks, and review of evidence that supports the subject.
Citations hyperlink the claim itself, not a source name in parentheses at the end. Click the sentence, see the roll call. A reviewer can audit any line in the book without asking us for anything.
Two independent quality sources on the same fact, or one plus a primary document. A single partisan write-up is a lead, not a finding — and it gets labeled as one.
Independent means a separate publisher: two sites owned by the same company, or a syndicated copy of one story, count once. On a podcast, a quote is attributed to a named speaker only when the episode’s own evidence puts that person in the conversation — someone the hosts discussed did not speak.
The most common failure in automated research is a different person with the same name. Committees, contribution records, licensing hits and market listings belonging to somebody else are identified, excluded, and the exclusion is documented in the book.
A records search returns rows that match a name, so before anything is written every records lane — federal and state campaign finance, nonprofit filings, court opinions and dockets, federal loans, property, UCC filings, business registries — is tested for a fact that ties it to your subject beyond the name. A lane with no such anchor is set aside rather than written up, and the count leads with what was verified. A dated record from before your subject turned eighteen is moved out of the money, loan and docket lanes and marked rejected, because a filing from someone’s childhood is somebody else’s.
Every URL is machine-verified before delivery. A dead link, a chamber homepage standing in for a specific vote, or display text that doesn’t match its destination is a blocker, not a nit. Nonprofit filings cite ProPublica's Nonprofit Explorer and roll calls cite the number the House Clerk assigns, because a citation is only evidence if a reader can still open it. A link that refuses an automated visitor is resolved against a record of real reads — read and confirmed, read and contradicted, or unread and reported as unread — so a refusal is never recorded as a reading.
“We found no record” is a research outcome, not a vulnerability. It routes to Manual Review with a note on what a human should check — instead of quietly becoming “has never disclosed.”
We re-read our own sources hunting for what helps the subject, and we print what we find. You should never be surprised by a fact sitting in your own bibliography.
Everything a Research Books engagement hands your team, and how fast each piece moves.
Executive summary, graded vulnerabilities, attack lines with rebuttals and counter-rebuttals, counter-evidence, screening, and manual-review notes — in Word, ready for your team to mark up.
Days, not weeks
Everything the pipeline gathered, not only what made the report — so a follow-up question in October doesn’t mean starting the research over.
Retained and re-queryable
A standalone bibliography, a voting-record appendix, a podcast transcript appendix so a quote can be read with the question it was answering, the complete social-media archive, and the raw data behind every figure we computed rather than quoted. Every document opens with a clickable contents list.
Ships with the book
A finished book can be scored against a structured trait inventory as a repeatable second step. Fifteen traits are scored separately and never collapsed into one number, because the groups run in opposite directions and a single score lets them cancel out. A trait with no evidence is left blank with a stated reason rather than scored low, and no source without speaker-level attribution can raise a score. Every output states plainly that there is no cut score and no comparison cohort.
On a completed book
A book on a prospective candidate can be rated across nine factors, from seat availability to fundraising, for a go / no-go on the race. Add a book on the likely opponent to rate their vulnerability too. See the nine factors ↓
On a completed book
News breaks and scandals emerge. The book is refreshed against the current record instead of rewritten from scratch, and it carries a research cutoff so you always know how current it is.
Anytime, all cycle
Should this race be run? A finished book on a prospective candidate becomes a rated assessment of the candidacy, optionally against a second book on the likely opponent. It is delivered as a branded PDF.
An open seat, a retirement, a term-limited exit, or an entrenched incumbent.
Favorable · Neutral · Unfavorable
Whether the seat’s partisan fundamentals favor the candidate’s party.
Favorable · Neutral · Unfavorable
How exploitable the incumbent or leading opponent is, rated from the opponent’s book.
High · Moderate · Low
How well the candidate’s biography and record match the district’s priorities.
Strong · Moderate · Weak
How much usable material an opponent’s researchers would find on the candidate.
Low · Moderate · High
Whether the candidate can raise what past winners of the seat have spent.
Strong · Viable · Challenging
Who else is running or likely to run, and whether the field helps or hurts.
Clear · Contested primary · Crowded
Whether the cycle favors the party in this kind of seat.
Tailwind · Neutral · Headwind
What it takes to get on the ballot. Rated for independent and third-party candidates only.
Routine · Demanding · Prohibitive
Same sources, same gates — a different report depending on who the subject is and what you need to know.
The standard opposition book, written to an angle. Vulnerabilities graded for deployment, attack lines built for debates, mail, and rapid response, and the counter-arguments waiting on the other side of each one.
A defensive book runs the same sources and the same gates against your own candidate, so you learn what the other side will find before they find it. The gates don’t loosen for a friendly subject — an inflated count helps nobody.
A neutral due-diligence profile with no attack lines and no electoral framing: subject profile, career and ventures, track record, financial profile, and areas of scrutiny. For vetting a hire, a partner, an endorsement, or an investment.
“In a 24-hour news cycle, a 3-month research window isn’t a strategy, it’s an autopsy.”
Tell us the subject and the angle, and we’ll walk you through a real book — including the sections most research products don’t have. Ask to see a sample and check the depth for yourself.
Starting at $7,500
Single subject or a full slate — pricing scales with scope and turnaround.
Contact Us