What's New

Political Autopsy — Why Did We Lose? (NEW, Early Access)

A race-centric post-election report that answers the question every losing campaign asks: why did we lose? Each Autopsy reconstructs the race from the data outward, with every number traceable to its source. Early access: contact us to commission one for your race →

• Itemized fundraising curves and the small-dollar mix
• Outside money, broken out by spender
• Prediction-market expectations over the final stretch, and certified results in district-historical context
• The paid air war
• Every claim has a receipt — a claim-by-claim Citation Map traces each fundraising figure, market-odds point, vote total, and ad-spend line to its original public source, generated mechanically from the same datasets that built the tables
• Narrative prose can only hyperlink sources from that same verified registry; anything else is stripped before delivery
• If a claim can’t be tied to a public source, it’s flagged for human resolution instead of published

Meta Ad Audiences — Your Poll Is Your Ad List (NEW)

Every segment a Synthetic Polling message test moves is also an ad audience. Any slice of a scored run now exports as a Meta Custom Audience, ready to upload to Facebook and Instagram — so the phone bank and the ad buy hit the same modeled voters, not two lists from two models. See how it works →

• Upload-ready to Meta’s own Customer List spec — every match key SHA‑256 hashed before the file leaves our systems — or seed a lookalike
• Every row carries its best-message tag, so the creative you run is the argument the model already said would move that segment
• You get the matchable count before you spend — reach is bounded by what’s on file, and we hand you that number first
• A “how a campaign runs it” playbook maps the run into persuade, reinforce, and expand audiences, each paired with the creative the test picked

Short Books Now Include Images

Research reports just got visual — and video visual analysis now runs by default on every new report with a YouTube presence.

• Subject photos, automatically selected and captioned — real candids pulled from the subject’s own public posts, chosen for probative value
• Verified video freeze-frames — stills from YouTube controversy moments, each deep-linked to the exact timestamp
• “Reach the subject” contact links — LinkedIn, X, Instagram, and Facebook, clickable right under the title

Deeper Money, Asset, and Business Research

A major push on following the money — what a subject controls, who they do business with, and what it connects to.

• Investment-adviser screening — subjects and their firms checked against the 23,000+ SEC and state-registered advisory firms added to the platform last week, including disciplinary disclosures
• Corporate-family mapping — global corporate-registry (GLEIF) ownership graphs now connect a subject’s businesses to their parent and subsidiary entities
• Controlled-entity portfolios fanned into government contracts, SEC Form D raises, nonprofit 990s, and PPP data
• Pay-to-play detection extends to state contracts — awards matched against donor employers across an 8.5-million-row warehouse (California, New York, Texas), alongside the existing federal-contract analysis
• Wealth mapping — aircraft and vessels attributed through LLCs and trusts
• A family-and-affiliation spine that anchors research on the right person from the start

Patent & Trademark Ownership History — A New Data Source

Civly research now includes the complete USPTO record of patent and trademark ownership transfers — roughly 12 million transactions since 1970, covering assignments, sales, mergers, licenses, and loans secured by IP.

• Why it matters: when an operating company moves assets to a holding company, both names appear on the transfer — exposing the network of LLCs and shell entities around a research subject that no other public source connects

Three New Business-Records Sources

Subject and controlled-entity research now checks:

• Federal contractor registrations (SAM.gov) — is the subject’s company registered to do business with the federal government, and under what codes?
• Motor-carrier records (FMCSA) — trucking and transport authority, fleet size, and safety registration tied to subject businesses
• New York State business registry — corporate filings for one of the highest-volume filing states, with the same identity guards as the rest of the platform, so a same-named stranger’s LLC never lands in your report

Property Held Through Companies

Real-estate research now follows the entities: properties held by companies a subject controls — not just properties in the subject’s own name — surface in the asset picture, connected back through the controlled-entity portfolio.

Sharper Identity Handling, Fewer False Positives

Research is only as good as its certainty that it’s reading about the right person.

• A unified subject-identity anchor now threads through search, deep research, and social classification — a major accuracy upgrade for common-name subjects
• Extremist-affiliation screening against published watchlists (SPLC Hate Map 2025)
• Fallback checks rescue real findings earlier passes marked empty, and research-scope metrics now report captured-versus-attempted honestly

Common-Name Accuracy

The unified subject-identity anchor above just got sharper teeth for the hardest case: subjects who share a name with hundreds of strangers. Business affiliations derived from nonprofit officer records are now geography-gated — an organization only attaches to your subject if its location, or a verified employer tie, actually matches.

• In one validation case, a common-name subject went from 276 spurious investment-adviser matches to zero, with nearly 200 same-named organizations nationwide correctly excluded and the legitimate handful kept — every keep/exclude decision logged with its reason

Voting Records and the Courts

Deeper coverage of what legislators and judges have actually done.

• Significant-votes ranking — thousands of votes analyzed per subject so the ten that matter surface first (one recent report: 3,024 analyzed, top 30 ranked)
• Court-opinion search now runs against the full text of each opinion — and is available through the Civly Intelligence API
• Judicial ideology scoring — a calibrated, factor-based scoring API for judicial subjects

Compliance Engine Upgrades

Donation screening now catches more of what a treasurer needs to know.

• Lobbyist awareness — contributions from registered lobbyists are flagged for your review, so you know before the press does
• Earmark detection — contributions routed through conduits with earmark language are identified and traced to their true source
• In-kind valuation — donated goods and services are valued and checked against limits the same way cash contributions are

AI Narrative Keeps Growing

Voters increasingly ask AI who your candidate is — the AI Narrative audit keeps widening what it checks.

• Audits now cover Grok and Google AI Overviews alongside ChatGPT, Gemini, Claude, and Perplexity
• Earned-media integration brings press coverage into the narrative picture
• A candidate-website health audit — page speed and Core Web Vitals, plus whether AI crawlers can actually read your content, with a specific fix for each finding
• Reports now render in a rich grid format: per-model report cards, drivers behind each narrative, earned-media context, site-health factors, and recommended actions — organized so a comms team can act on it in one read

Down-Ballot Bundle — $999, Now on the Homepage

Full-stack research sized for down-ballot races, now featured on the homepage. See the Down-Ballot Bundle →

• Order research plus dialer-ready contact exports for a whole slate of down-ballot candidates in one request

Civly in Slack (Early Testing)

We’ve started testing a Civly assistant that lives right in Slack. Our own team is running it daily now. Want early access when it opens up? Get in touch and we’ll put you on the list →

• Mention the bot to kick off a Quick Vet or an AI Narrative probe, and approve the run with one click
• The finished report is delivered back into the same thread when it’s ready
• A condensed daily briefing, posted to your team channel

Clip Catalog — A New Opposition-Research Product (NEW)

Point Clip Catalog at a public figure’s podcasts, YouTube videos, and recorded floor speeches, and get back a fact-checked, click-to-play catalog of their most newsworthy moments — with the right words attributed to the right speaker.

• Transcribes the audio and separates speakers, so a host’s words are never mistakenly pinned on a guest — a common failure of ordinary transcription
• Screens for controversial moments, checks the surrounding context, and confirms the statement was actually theirs before flagging it
• Captures the exact still frame and timestamp, and delivers an analyst-ready Word catalog where every finding links straight to the clip
• Runs unattended at scale and resumes where it left off, so a full channel or feed processes in one pass

Seven New Research Data Sources

Seven new public datasets are now loaded, indexed, and searchable in the platform — several wired straight into people-and-business matching.

• Doctor-and-hospital industry payments (CMS Open Payments) and investment-adviser registrations (SEC Form ADV)
• Trucking-company records (FMCSA), business lien filings (UCC), and the federal contractor registry (SAM.gov)
• State contract awards for California, New York, and Texas, plus a new Ohio campaign-finance tab

Open Payments Newsroom — A Free Public Data Tool

On top of loading the doctor-and-hospital payments data, we built a free, public tool aimed at journalists and the public. Explore the Open Payments tool →

• Pull payment totals by state, county, and congressional district
• Filter by year and payment category, and see how a state’s totals have changed year over year
• A simple public interface with pre-computed totals, so results come back fast

Entity Backbone — Connecting Companies, People, and Donors

A new canonical “entity backbone” links business records to the people behind them and bridges those people to donor records — so research connects a company, its leadership, and their political giving automatically instead of by hand.

• Company-to-donor connections scaled up to full coverage
• A production matching engine that finds the right record even when names are spelled differently or abbreviated

Synthetic Polling — A Maryland 2026 Forecast, Now Multi-Model

Our synthetic-polling engine added a new race and a reliability upgrade. See Synthetic Polling →

• A Maryland 2026 primary forecast with incumbency and self-funder corrections, delivered as a Word document and backed by a six-race accuracy backtest
• The engine and its calibration can now run on multiple AI models, so results can be cross-checked head-to-head

Social Search — Telegram and Mastodon Added

Our unified Social Search now reaches two more networks.

• Telegram and Mastodon join the existing networks, so a single search covers more of where a subject actually posts
• Ready-to-use demo presets for both, so you can try them on a known account in one click

Reliability & Performance

A large share of the week went into making everything faster and steadier.

• Core donor datasets now rebuild in the background and swap in with zero downtime, on a much lighter, safer process
• Hardened the bulk import for huge, messy data files, so a single malformed row or an oversized field can’t sink a multi-gigabyte load
• Scheduled refreshes now run on their own — Ohio campaign finance weekly, and the federal contractor registry monthly

Research War Room — Rebuilt for Speed and Memory

The Research War Room got a ground-up rebuild — a faster workspace that opens straight into your most recent session, so you pick up exactly where you left off.

• A new searchable, grouped history sidebar with auto-named sessions — find and reopen past work in seconds instead of starting over
• Deep Research and People Finder are now built-in tabs, alongside the AI Narrative tab added last week
• Every run, including AI Narrative, is saved to a Recent Analyses history and to your Research Vault

Voter Files Explorer — Registration & Turnout, Just Ask

Explore voter registration and turnout the same way you already explore campaign finance — now with a built-in AI chat, so a question gets you an answer without writing a query.

• Type a question like “turnout among under-35 unaffiliated voters last cycle” and get a result, not a research project
• Break registration and turnout down by party, age, and vote history — the cuts you’d normally pull a file to get
• Michigan is now live, joining the states already covered

Synthetic Focus Group — Hear Why Voters Move, in Their Words

Message testing tells you how far a message moves the vote; the new Synthetic Focus Group tells you why. Put a simulated panel of real-voter stand-ins in a room with your competing messages and read how they react — before you spend on a live focus group. Read the Kansas focus group →

• The same simulated electorate from a message-test run reacts to each message in its own words — in the Kansas pilot, 292 persuadable voters produced 4,380 reactions in minutes
• Each voice is a stand-in for a real registered voter, tagged by party and age — not an invented persona
• Surfaces the language and themes that actually land, so you can pick and sharpen a framing before validating the finalist with real voters
• Directional by design: like any synthetic instrument it points you to the right message — it doesn’t replace a live test

Deeper Legislative Research — Federal Bills & an Effectiveness Score

Federal bill records are now in the platform, so you can research a legislator’s actual record — not just what they say about it.

• Ask about a bill’s full history — sponsors, votes, and vetoes — in plain language
• A new in-house legislator effectiveness score — an internal reproduction of the Center for Effective Lawmaking model — ranks a member against their peers by percentile
• Structured bill data comes first, with web search as a fallback, so the record is grounded in the official source rather than a summary of it

Steadier Social Media Analysis — and a Polished Partner Deliverable

A batch of reliability and output improvements to social media analysis, so reports finish more consistently and read more cleanly.

• Instagram analysis now uses a more robust collection method with automatic fallback, so a single hiccup no longer stalls a report
• Analyses return faster, with the heavy work running in the background
• For partners and API: social analysis now delivers a polished Word document with a full transcript appendix

Expanded Data Coverage

New public-record sources added this week, so more of who a subject is surfaces automatically.

• Nevada state campaign finance
• Michigan voter registration and turnout
• Federal legislation and bill records

Accuracy & Reliability

A few guards that make research more precise and deliverables more trustworthy.

• A new namesake guard blocks fabricated connections between a subject and an unrelated person who happens to share their name
• Numeric and quoted claims now bind to the exact page that contains them, so every citation links to the source of the claim — not a generic index
• Dead source links auto-repoint to an archived copy during the pre-delivery audit, so citations keep working

Civly iOS App — Now Live on the App Store

The Civly iOS app is out of testing and live on the Apple App Store — anyone can download it straight to their iPhone, no TestFlight invite required. The platform that started in the browser now ships in your pocket, built for the people doing the work in the field. Download Civly on the App Store →

• Launch a canvassing shift in one tap from the bottom nav, built for field volunteers
• Step count auto-syncs from Apple Health during a shift, so admins can see a route was actually walked — not just marked complete

National Voter File (L2) — Verified Voter Records, Built In

We’ve integrated L2’s national voter file, so Civly can now match a person to their verified voter record — registration, vote history, and cell phone — and bring it straight into your research.

• For any research subject, Civly automatically pulls voter records for the person, their household, and related donors
• Every match is graded for confidence, and only the highest-confidence matches are accepted automatically — when a match is uncertain we skip it rather than guess, so you’re never shown the wrong person’s record
• Missing donor cell phones can be filled in from the voter file, used only for the highest-confidence matches so a number is safe to dial
• Vendor-modeled fields are labeled as modeled, and records we can’t confidently match are flagged plainly rather than guessed at

AI Narrative — See What AI Assistants Say About a Subject

Voters increasingly ask AI chatbots about candidates — so we now show you what they hear. AI Narrative asks the major AI assistants about a subject and surfaces their answers, making it easy to spot misinformation before it spreads.

• Captures what ChatGPT, Gemini, Claude, and Perplexity say about a subject, then grades each answer and fact-checks it against the verified facts already in the book
• Available both as a Big Book section and as a dedicated AI Narrative tab in the Research War Room
• Try it on any subject from an inline box, with a one-click demo, and save results to the Vault

Synthetic Polling Is Now a Core Product — with Message Testing & Voter Targeting

Synthetic Polling has graduated from research study to core product — you’ll now find it under Product in the site menu. And it goes well past predicting a race: the same engine now measures which messages move your voters, and turns every poll into a ranked contact list. See the live Kansas results →

• Message testing: the same simulated electorate is re-polled under each message, and movement is measured per voter against their own baseline — in the Kansas pilot (2,000 real Johnson County voters, five messages), opposition framings out-punched supportive ones about four to one, and undecided voters consistently moved most
• Voter targeting: every poll doubles as a targeting run — undecided, unstable, and cross-pressured voters identified from the voter file, each tagged with the message that moved them most, exportable for calls, mail, and canvass, with counts that re-aggregate exactly to the published topline
• A validation report card on the page: blind reproductions of published experiments — question-wording gaps, the attacks-beat-positives asymmetry, candidate-scandal ordering — plus a no-message control showing real messages move voters far above re-ask noise
• An honest-limits box, on the page and on every deliverable: rankings are the validated signal, effect sizes ship as ranges, cross-party persuasion is flagged as under-detected, and the secret ballot means per-voter scores are an ordering — not a prophecy

Synthetic Polling — Candidate Races Now Anchored to Real Past Results

The candidate-race correction now starts from each county’s own certified result in the previous presidential election — public, pre-election information, the same starting point every serious forecast uses — and the AI engine measures the shift from there. Re-scored on the same 234-county, three-state test, the average county error fell from 7.6 points to 1.7. See the updated numbers →

• Average county error on the 2024 presidential margin: 7.6 → 1.7 points across all 234 counties in North Carolina, Florida, and Pennsylvania — every county still predicted out of sample
• The statewide margin lands within 0.4 to 0.7 points in all three states, and the worst county in the study improves from 46 points off (Robeson, N.C.) to about 10 (Miami-Dade, Fla.)
• Labeled honestly: these are 2024 presidential backtest numbers — on a same-office test the anchor itself does much of the work — and cross-race validation (the 2024 N.C. governor’s race, the Pa. and Fla. Senate races) is the next test to ship
• With the tail of big misses gone, the average no longer has anything to hide — so every accuracy figure on the page is back to a plain average, including the unchanged ballot-measure results

Civly Intelligence — Our Research Data, Now Available via API

Partner organizations can now access Civly’s research data and reports programmatically through the new Civly Intelligence API.

• A managed API gateway with a simple access-request flow
• Monthly usage quotas per product (for example, books or social analyses), adjustable per key
• A new read-only Data Access option for bulk data export, with an auto-generated data dictionary so partners can see exactly which fields they receive
• Social analysis now returns in seconds, with the heavy work running in the background

Verified On-Air TV Quotes — Verbatim, Sourced, Click-to-Play

A new Big Book section captures what a subject actually said on television — verbatim, with the receipts.

• Screens TV-news coverage for verbatim on-air quotes and renders each with the station, show, and date
• Every quote carries a severity rating and a click-to-play archive link, so you can watch the moment yourself
• When nothing notable turns up, the report still documents what was screened

Prediction-Market Odds — Polymarket & Kalshi, In Your Research

Big Books now include live prediction-market odds for a subject’s races, drawn from Polymarket and Kalshi.

• Daily odds snapshots with history backfill, so you can see where the market has moved
• A built-in guard means a thin trading history never produces a misleading “trend”
• Also available as a research tool in the War Room

A Major Expansion of Research Data

We added a large batch of new public-record and compliance sources this week, so more of who a subject is surfaces automatically.

• SEC Form D filings — private-placement roles (promoter, director, executive) and the economics of the offering
• Public-employee salaries — agency, title, and pay, matched to donors, starting with Florida and Texas
• U.S. Coast Guard documented vessels — boat ownership as a luxury-asset signal
• USDA farm subsidies, plus corporate-family data (GLEIF/LEI) that rolls an employer up to its parent company
• More state and local campaign finance: New Mexico (full pipeline), California city and county via NetFile (~200 agencies), and Washington (now with city-race filtering)
• Google political-ad spend and advertisers, now included as a research signal
• Broadened compliance screening — donor checks now run against the full U.S. Consolidated Screening List (not just the Treasury SDN list), with SAM.gov federal exclusions as a second check and a safeguard against false matches on a shared name

Reliability & Quality

A batch of improvements to make books more complete and deliverables cleaner.

• Court-record results are no longer capped — books now report the full count (one book that had shown 100 court records actually had more than 1,100)
• Sharply fewer false-positive citation flags in Short Books, while live PDF and podcast citations are kept
• More reliable discovery of social handles and campaign sites, with government domains excluded from campaign-site matches
• A single empty data section now fails on its own instead of affecting the rest of the book
• Fixed an issue that could blank out ordinary words in a report

Donor Dossier — Redesigned

The Donor Dossier got a ground-up redesign — a cleaner, six-section format that gives you a far richer, better-organized picture of who a donor is and how they give.

• An executive summary up front puts the headline first, so you don’t have to dig for it
• A richer giving profile: contributions are grouped by jurisdiction, attributed to the named committees behind them, and de-duplicated for a truer total
• Now also pulls in the subject’s relevant social posts, disambiguated news coverage, and an AI-written profile
• Every organization and source is hyperlinked, with a consolidated Links & Sources appendix — and a pre-delivery audit verifies every link before the document goes out

California Behested Payments — A New Research Signal

We’ve added California “behested payment” filings to our research data — a layer of influence and relationship data that standard contribution records miss.

• Behested payments are made to organizations at the request of a public official, surfacing the official → donor → charity money nexus
• Drawn from California Form 803 filings, tied to the subject and linked back to the original record

On-Screen Video Analysis — Big Book

Our YouTube controversy analysis can now examine what’s on screen, not just what’s said — a deeper read on the moments that matter in a clip.

• An optional visual pass flags noteworthy on-screen imagery alongside spoken-word findings, each with a timestamped, clickable source link
• Building on last week’s podcast coverage, notable statements from a subject’s podcast appearances are now surfaced and deep-linked to the exact moment in the audio — and when nothing notable turns up, the report still documents what was screened

More Accurate Donor & Contribution Data

A batch of accuracy improvements across donor profiles and contribution data, so what you see reflects reality more precisely.

• Cleaner donor profiles: we’re better at recognizing when multiple records belong to the same person, now with a safeguard against incorrectly merging similar names — for example, a parent and child who share one — for fewer duplicates and a truer picture of each donor
• More accurate contribution totals: we removed duplicate records from several states’ campaign-finance data, so totals and giving histories are more precise
• Delaware voter registration added, improving identity and address verification for donors and contacts in the state
• Upgraded phone and identity matching for more reliable contact-confidence signals when verifying how to reach a donor

Reliability & Deliverable Quality

Plus a couple of improvements to keep reports running smoothly and client documents clean.

• Internal QC notes are now stripped from exported deliverables, so client-facing documents stay clean
• A per-source timeout keeps a single slow data source from stalling an entire report

Synthetic Polling — The Research Is Live

We published an interactive research page showing how Synthetic Polling stacks up against reality — predicted vs. certified 2024 results across North Carolina, Florida, and Pennsylvania, plus all six Florida ballot measures, county by county, with the full methodology and sources. Explore the study →

Synthetic Polling (New) — Predicted Poll Answers in Hours, Not Weeks

A real district-level poll can cost tens of thousands of dollars and take weeks — so most campaigns only ever test one or two questions. Synthetic Polling gives you predicted answers to the same kinds of questions in hours, for a fraction of the cost, so you can test twenty message variations instead of one. It’s built to complement traditional polling, not replace it — for asking more questions, faster, than a real poll can affordably cover.

• Ask vote-choice, candidate-favorability, and policy-position questions for a specific congressional district, a state, or nationally
• Each simulated voter is grounded in real public-record behavior — actual voting history, campaign donations, and household makeup, not demographics alone — fully anonymized from data Civly already has
• Every voter is sampled multiple times, producing a per-voter confidence rating alongside the population result — the simulator reports a distribution, not false certainty
• A calibration layer, trained on published survey responses, corrects for systematic model tendencies, so policy and favorability questions come back trustworthy — not just head-to-head races
• Hyper-specific by design: test a single message framing, a niche policy angle, or a sub-district segment that nobody else has bothered to poll
• Benchmarked against real 2024 election outcomes in tight, competitive races

Mobile Experience — Redesigned for the Field

The mobile experience got a ground-up redesign — a cleaner, dark, phone-first layout built for the people doing the work from their phone. The tools you reach for in the field are now one tap away.

• Refreshed Social Search with a cleaner, dark, phone-optimized layout
• New Chat tab in the bottom navigation — one-tap access to the War Room AI research assistant
• Canvassing moved into the bottom nav, so you can jump to door-knocking from anywhere in the app
• Social Alerts and the Briefing Room now fit properly on phones, with readable pagination, filters, and trending links
• Continued groundwork for the native iOS app, building on last week’s TestFlight launch

Canvassing — North Carolina Now Live

Full North Carolina support is live for door-knocking teams — voter file, voter history, and address data all in place — and multi-state teams can now move between turf without leaving the page.

• Complete North Carolina coverage: voter file, voter-history, and address data
• Switch states from a dropdown in the canvassing header — manage every race from one place
• Step and visit verification is now visible to both canvassers and admins, for cleaner accountability on whether a route was actually walked
• Smoother maps on mobile and faster loading across large areas

Donor & Business Research — Smarter Profiles, Fresher Data

Donor research gains a new signal and more complete profiles this week, plus faster state campaign-finance data that keeps itself current.

• New: business-reputation lookups, with automatic location matching so you get the right result with less input
• More complete donor profiles — professional-licensing records across many states, with missing occupation and employer details filled in automatically (which also helps with FEC compliance)
• Fresher, faster state campaign-finance data, with contributions kept up to date automatically

Deeper Research Coverage — Video, Podcasts, and New Public Registries

Big Book research now reaches into audio and video and pulls from a wave of new public registries, so more of who a subject is surfaces automatically.

• Research now pulls from video and podcast sources, with accurate, speaker-attributed transcripts so quotes are tied to the right person
• New public registries added: FAA pilot and aircraft registrations, and FCC license records
• Expanded background data, including Arkansas voter registration
• Structured identity and date-of-birth matching helps pin down the right person earlier in research
• Local-newspaper milestones — engagements, weddings, anniversaries, obituaries — now surface the biographical detail that never makes national coverage

Reliability & Accuracy

A batch of accuracy and speed improvements across the platform.

• More accurate social media analysis, with better handle matching and fewer incomplete runs
• A faster, more accurate Savings Calculator — results now load in the background, with improved committee matching
• Research runs now recover on their own from interruptions, so books finish without manual intervention
• Various data-quality and speed improvements across FEC and state data, so lookups are quicker and more reliable

Civly iOS App — Now Live on TestFlight

The Civly app is on TestFlight for internal testers — the first time the platform exists outside the browser. Up to 100 team members can install it directly to their phone, with new builds available within minutes of upload. Pointed at the dev backend.

• Canvass tab on the bottom nav — field volunteers launch a shift in one tap
• Step count auto-syncs from Apple Health during a shift, so admins see whether the route was actually walked, not just marked complete
• Push notifications wired up for the next release: shift assignments, outreach replies, Big Book completion alerts
• Production app identifier locked in (ai.civly.app) — no migration when we move from TestFlight to wider release

Donor Prospecting — Statewide Search Speed Rewrite

State donor lookups that used to spin for 30+ seconds before timing out are now milliseconds. The 40-state contribution view has been rewritten as a pre-computed snapshot, with name and city normalization done upfront and indexed for direct lookup.

• Single-name lookup target: under 5 seconds. 30-name batch lookup: under 30 seconds.
• Weekly rebuild cut from a 10-hour serial job to 2–3 hours by fanning across 4 parallel workers
• Four downstream donor aggregations repointed to read from the new snapshot instead of redoing the 40-table stitching on every rebuild

Donor Prospecting — Professional Licensing Data

Every state professional license we can get is being loaded into one queryable table inside Civly. First wave: 13 datasets across IL, CO, CT, TX, and NY — about 9.9M rows covering doctors, lawyers, dentists, accountants, real estate agents, appraisers, engineers, notaries, cosmetologists, and more.

• FEC contribution autofill: when a donor types their name, occupation and employer auto-fill from licensing records — addressing the legal requirement campaigns currently meet by hand
• Cross-state high-income prospecting: “every licensed cardiologist in IL who gave $5K+ to Democrats last cycle” returns a list, not a research project
• Person Search becomes occupation-aware; donor dossiers gain a verified profession line; Big Book gets a new evidence channel

Big Book — Six New Sources, Self-Healing, and Coverage Hardening

• New sources live: FCC Amateur Radio Licensing, state professional licensing, local newspaper milestones (engagements, obituaries, anniversaries), Wikidata (structured DOB and identity), FINRA BrokerCheck, NPPES NPI Registry
• Self-healing orchestrator detects stalled research runs and re-dispatches stuck sources automatically — partial-stuck recovery in ~10–15 minutes for gathering and ~30 minutes for deep research, no human intervention
• Wayback Machine fallback fires automatically when Twitter scraping hits an exception, recovering historical posts
• PPP loan search now covers the subject’s known business entities, with gap-analysis checks that flag missing matches
• Article-fetch capture rate is now persisted and surfaced on every run — partial captures are visible instead of silent
• Smarter entity disambiguation in deep research filters results that share a name but refer to a different person

Short Book — Content, QC, and Rendering Overhaul

• New Deleted Posts section captures social-media posts a subject removed
• Rumors & Allegations now admits properly-sourced anecdotal rumors (was previously restricted to hard documentary backing)
• 20–30 page length cap lifted — length follows the evidence
• Citation source diversity enforced — the system prevents over-reliance on a single source or domain across a book
• Pre-delivery audits catch broken bibliography URLs, sources that refer to a different person, and uncertain social handles (marked [NEEDS REVIEW])
• Pre-delivery audit now distinguishes BROKEN (URL doesn’t resolve) vs UNVERIFIABLE (resolves but doesn’t back the claim); audit-failed runs can patch and re-export without restarting
• Citation Map ↔ prose reconciliation pass at end of generation; pre-commit guard catches paraphrased quotes before they ship
• Bibliography preserves hyperlinks on [PDF]/[DOC]/[XLS]-tagged and multi-line titles; raw autolinks and placeholders no longer leak into delivered docs

Custom Video — YouTube Controversy Capture Now Fully Automated

Last release was the safety and accuracy pass. This release, YouTube controversy capture runs as a fully automated background job, dispatched from Big Book as part of subject research.

• Used to require a manual command-line run; now Big Book kicks it off, polls for completion, and pulls findings into the deliverable
• Findings are queryable — every individual finding (quote, timestamp, deep link, severity, headline) is its own row tied to the run, instead of living in a one-off report
• Per-word transcript confidence flows through to screening — moments where the audio isn’t clear enough get dropped automatically
• Any screener quote not word-for-word in the transcript is dropped — no hallucinated quotes survive into findings
• Smoke test ran end-to-end in 68 seconds on dev for a one-video subject, with the finding correctly referencing the subject’s specific opponent

Savings Calculator — Now Handles Every Committee, No Matter How Big

The public calculator was timing out on giant committees. Bernie 2020, for example, has ~65,000 disbursement rows that each require an AI classification call — the 60-second request limit was killing the job before the math finished.

• Async with polling — the page kicks off a background job and polls for results while showing a “Classifying...” spinner. No more 60-second cliff.
• Cache hits are instant — same committee + cycle + source served from cache, so the second visitor pays nothing
• Committee search now also matches by linked candidate name, so “Joe Smith for Congress” finds “Friends of Joe Smith”

Donor Enrichment — Phone Confidence Auto-Refreshes

When Apollo (our second-pass verification source) returns a better signal days after the initial enrichment, the donor’s phone confidence tier now updates in place automatically — no need to re-run a paid lookup. A one-time backfill moved existing donors up tiers from the latest Apollo data across the full enriched-donor table, no extra cost.

Savings Calculator — Coming Soon

A new public tool that shows campaigns exactly how much they could have saved on their last race. Drop in your committee — or any competitor’s — and in seconds you’ll see every disbursement broken down by category, the biggest line items in each bucket, and a dollar figure on the spend that could have moved to a more cost-effective vendor.

• Built for campaigns, consultants, and donors who want a quick read on whether a committee’s budget is being spent well — no signup, no sales call, just a number
• Works across federal, state, and major-city campaign finance — including New York City, Philadelphia, Los Angeles, San Francisco, and DC
• Honest about what’s actually movable: postage, direct mail production, staff salaries, and compliance fees are excluded from the savings figure, so the number you see reflects real opportunity rather than line items a campaign can’t change
• California campaigns get correctly categorized spend via CAL-ACCESS purpose-code mapping — no “Other” bucket swallowing the breakdown

Donor Research — Federal and State Giving Now Link Automatically

The data view that powers donor prospecting now merges FEC contributions directly into the unified donor profile. A donor who gave $50K to federal committees and $30K to state PACs no longer shows up as two separate records — their federal and state giving histories link to the same person automatically, with committee names attached.

• Donor prospecting searches return federal-and-state totals in a single view
• Especially impactful for research on candidates and donors who give across jurisdictions
• Builds on the prior week’s FEC name-cleaning work: “LASTNAME, FIRSTNAME [MIDDLE] [SUFFIX]” strings are now split into clean first and last names and matched against state records inline

Custom Video — Major Safety and Accuracy Pass

Custom Video produces campaign-ready opposition and contrast videos. It scans public footage — legislative sessions, county and school board meetings, candidate podcast appearances, YouTube content — for the moments that actually move a race: contradictions, gaffes, on-record positions a candidate can be held to, off-message remarks. What used to take an opposition research team months of manual review arrives in days as finished clips, ready to put in front of voters.

This release is a top-to-bottom pass on what makes it into the final cut. An attack clip is worthless — or worse, a liability — if the quote was misheard, the context is wrong, or the claim doesn’t hold up. Every guardrail below exists so the clips we surface are ones a campaign can actually run, without a retraction or a self-inflicted news cycle.

• Factual claims are now verified against real-world sources before they’re flagged. If a citation can’t be produced, the clip doesn’t go out as a factual hit — no more “subject to fact-check” hedges making it into a deliverable
• Quotes the transcript can’t confirm word-for-word are dropped automatically, and moments where the audio isn’t clear enough to be confident in what was said get filtered out — no more hallucinated quotes surviving into the final clip set
• Sarcastic moments stay sarcastic — a new tone check prevents the system from framing a candidate’s joke or sarcasm as a sincere claim
• Biographical details about the subject are now built from verified facts only, so the system can’t invent something about the person it’s reporting on
• Defamation and other legal-risk framings have been removed from attack-angle output — so the clip’s framing won’t put a campaign on the wrong end of a lawsuit

Person Search — Multi-Word Names and Nickname Matching

• Multi-word federal-official names now match correctly — a search for “Maria Cantwell” returns the right person
• Nickname resolution via a new alias table: searching “Bill Clinton” finds William Jefferson Clinton

Social Alerts — Geography-Aware Filtering

• Topic alerts only fire when the underlying content is geographically relevant to the campaign’s region. Statewide stories still surface when the campaign is scoped to a state-level region
• Stuck refresh chains self-heal — if a refresh task gets wedged, it auto-cancels and re-fires instead of needing manual intervention
• Tracked-person and tracked-topic freshness timestamps update accurately so the “last updated” signal can be trusted

Canvassing — One-Click Route to Turf, Voter-Overlap Warnings

• Save an optimized route as a turf in a single click — no more separate save step after planning
• Route controls now show “skipping N voters already in your other turfs” so canvassers don’t accidentally double-knock the same households

Find More Donors — Phone & Email Enrichment, On by Default

• Donors with no address on file now get an address pulled from five different sources (existing profiles, FEC records, state campaign finance, voter registrations, and prior enrichment runs) before any paid lookup runs. Without an address, Twilio’s identity-verification step can’t fire, so these donors were previously unreachable
• Address completion lifted from 59% to 70% on a 325-donor test group — about 35 previously-unreachable donors per group now reachable at our highest confidence tier
• Measured across two real client donor lists (291 and 182 donors): 16–32% more high-confidence mobile numbers, 9.3% more verified emails (plus another 7.1% deliverable but unverified), and 28.6% of donors gained at least one new way to reach them. Donors with no phone on file saw the biggest mobile lift at 36.1%

More State Data Live

• Arizona campaign finance live: 9.4M contributions and 857K expenditures now searchable
• West Virginia campaign finance live: 694K contributions and 188K expenditures now searchable
• DC voter file loaded — ~478K voters and 3.6M voting records on file; our voter-file footprint is now 8 states
• NY and WA voter files now supported; NC voter file fixed to include phone numbers (filled on ~42% of records) and birth years that the prior version was missing

FEC Data — 54M Federal Contributions, Cleaned Up

• Federal and state giving histories now link to the same person automatically. FEC publishes contributor names as one long “LASTNAME, FIRSTNAME [MIDDLE] [SUFFIX/TITLE]” string — we now split that into clean first and last name and match it against state records, so a donor who gave $50K federally and $30K to a state PAC stops showing up as two separate records
• City names cleaned up — “New York”, “NYC”, “Manhattan”, and “New York, NY” all collapse to the same place, so geographic searches return the donor regardless of how they wrote it
• Filtering FEC donors by occupation is now instant. Searching for ATTORNEY in Texas or PHYSICIAN in Florida used to take 30+ seconds — it’s now sub-second, the difference between a usable filter and one nobody bothers to apply

Outreach — A/B Testing Email Copy

Instead of sending one template and guessing what worked, nine email variants now go out at the same time with different messaging angles and asks. Opens, clicks, and replies attach to specific copy choices instead of being averaged across a single send.

• Per-variant opens, clicks, and replies, plus a click-by-link breakdown so you can see which links inside each email are working — not just whether the email opened
• Automated weekly performance report lands in your inbox every Monday morning, summarizing the prior week’s variants
• Bot and corporate-scanner clicks filtered out automatically; Apple Mail’s automatic image loading (which inflates raw open rates 20–50% depending on audience) is detected and excluded so you see real human engagement

Canvassing — Live Map Updates, Per-Door Voter Lists, and Verified Routes

• Tap any address inside a turf to see every registered voter living at that door, then file a voter-registration (NVRA) form or advance-ballot request directly from that screen — no extra navigation
• Voter and unregistered-household pins update in real time as turf assignments change, so canvassers always see the current state of the field
• Canvassers’ phones now log walking steps during shifts, giving admins an accurate record of the ground each canvasser covered
• Admin shift review and all-team progress dashboard — drill into individual shifts or see live performance across the field team

Custom Video Deliverables — Public Government Footage, Now in Play

• State legislatures, county boards, school boards, and other government bodies publish hours of session footage every week — almost none of it is transcribed or reviewed by anyone. Our video studio now ingests this publicly-available footage, transcribes it, surfaces the newsworthy moments, and turns them into campaign-ready video deliverables
• The data we already hold for nearly every politically-relevant person in the country — donor history, voting records, opposition research, Big Book and Short Book deliverables — pairs with this public footage to produce custom videos in days, not months
• Generalizes to any level of government: state legislatures, county commissions, school boards, congressional races, and contested primaries
• Reliability work shipped with the first production runs: completed videos save to your Vault automatically, expired video links refresh instead of breaking, and bad image references get caught before render

Big Book — Family Discovery, Deleted-Tweet Recovery, and Scored Citations

• Donor profiles automatically pull family members now, with their social profiles and news coverage attached
• Deleted-tweet recovery now shows when each tweet was originally posted, how many archived copies exist, whether the recovered version is truncated, and whether it’s an original tweet or a retweet. A second archive (archive.today) catches deleted posts that the Wayback Machine misses
• Every citation in a Big Book now ships with a confidence score, and weak citations are automatically rerun against alternative sources before the book is delivered
• New research areas: state judicial-discipline records (for books on judges), donor-extremism scanning, subject-owned businesses, entity-controversy checks, and a flag on endorsements from publications that accepted payment for them
• Family research now available for candidate subjects (was previously donor-only); Family Background and Philanthropy sections now render reliably; subjects with Google Ads spend no longer crash the run
• Reliability work across the long tail: a single bad section no longer kills the rest of the book, slow sources time out cleanly instead of hanging the run, and the progress bar moves continuously instead of jumping by section

Civly AI — Three New Fine-Tuned Models in Production

Civly builds and trains its own AI models, purpose-built for campaign finance and donor research. Every record the platform processes makes the next model better, and we can tune for the places generic AI struggles — donor name disambiguation, employer shorthand, sparse research snippets. Three new fine-tuned models shipped to production this week.

Cardinal v1 — Donor city correction

Cardinal rewrites misspelled, abbreviated, and typo’d donor cities to the canonical spelling — automatically every Monday, across all 33 state campaign-finance datasets and the unified donor view. Handles more regional variants and edge cases as volume grows.

Codex v1 — Employer & occupation classifier

Codex auto-tags employer and occupation on campaign finance receipts across our state rotation, with new coverage for Philadelphia, San Francisco, and Michigan. Tuned on the abbreviations and shorthand that show up on real filings, and sharper with every new state we add.

Recon v1 — Contact extraction

Recon pulls phone numbers, emails, and addresses out of research documents. Trained to refuse rather than guess on ambiguous matches — closing a failure mode where generic AI confidently attaches the wrong person’s contact info to a donor profile.

Big Book — More Reliable Under Heavy Use

A wave of reliability work so multiple opposition research reports can run side-by-side without delays, restarts, or silent failures.

• Big Books now resume automatically when infrastructure recycles a worker mid-run instead of starting over
• Several reports can run at the same time without slowing each other down
• Short Book and Big Book run on separate lanes so a Short Book queue can’t delay a Big Book
• When a section of a report fails, you can now see exactly which one in the status view and retry just that section
• Reports finish more often when the AI returns unexpected output formats
• Resolved several database connection issues that had been causing silent task failures

Canvassing — Voter Registration, Advance Ballot, and a Cleaner Map

The doorstep canvassing app gained a full voter-registration workflow, advance-ballot follow-up, a redesigned turf interface, and the foundations to roll out beyond Kansas.

Register voters at the door

• Canvassers can now fill out the federal NVRA voter-registration form (Form 76-15) directly on the canvassing app when they encounter an unregistered household
• Right after registration, an advance / mail-in ballot follow-up form pre-fills from the just-submitted NVRA so the resident only signs once more — the second signature lifts voting likelihood from ~15% to ~75%
• Both forms are auto-emailed as PDFs to the county election office and to the campaign user
• Every submission is tracked from sent → received by county → done

Map and turf workflow

• Unregistered homes show as gray dots on the map — tap one to see the address and knock there
• Voter pins are color-shaded by party affiliation and vote frequency for at-a-glance prioritization
• New Turf filter dropdown above the precinct picker: pick a turf and the map locks to it with a coverage and outcome stats panel
• New “Not registered (gray pins only)” filter for registration-drive shifts
• Visit logs now work for unregistered homes too, including offline — they sync when you’re back online
• Visited pins fade with a green ring so you can see your progress at a glance
• Turf sync errors now recover with a one-click “Discard” button

Beyond Kansas

• The address-to-precinct matching used in Kansas now works against any state, paving the way for the next state launches
• Better matching on messy addresses — apartment units, partial zip codes, and other rough inputs
• Per-state ingest jobs let us bring new states online without touching the live app

Donor Prospecting — Phone Enrichment Upgrades

New phone-quality signals layered on top of the existing carrier check.

• Subscriber-name match: we compare the carrier-registered name to the donor’s name and surface a “name matches” flag on the prospecting list
• Identity match (in testing): a confidence score combining name and address, currently A/B’d against the cheaper name-only path
• Phone confidence tiers now appear inline on the prospecting list, mini-book dossiers, exports, and the donor enrichment panel — see which numbers are mobile, landline, or unverified at a glance
• Previously enriched phones can be re-checked against the new signals on demand

Email Verification — Faster and More Reliable

Email verification has been replaced with a faster, more reliable backend that sidesteps the SMTP traps that used to hang verification calls.

• One observed verification dropped from 240 seconds to under half a second
• Verifications now cache for 90 days, so you stop paying to re-verify the same address every week
• “Valid” results are auto-invalidated when your own outreach has hard-bounced the address — no more emailing into known-dead inboxes

Improvements

• Cleaner email-pattern guessing skips domains that just rejected the address
• Donor clusters now skip blank cities and states for cleaner groupings
• Bulk donor refreshes recover from transient errors instead of stalling out
• Briefing Room flags stale data with a banner and refreshes in the background
• CRM open and click rates filter out bot and honeypot clicks for accuracy

Donor Prospecting — Cluster Rollup and Faster Filters

Donors that show up under multiple variants — different employers, addresses, name spellings — now roll up into a single profile so your list isn’t cluttered with duplicates.

• Each cluster shows how many underlying records were merged, who the highest-giving member is, and combined totals across years and states
• Pre-computed filters for “has email”, “has phone”, and “has LinkedIn” make the filter panel instant, even on the full 17M-profile set

Briefing Room — Redesigned

The 8 AM email is now organized around what you actually track, with broader source coverage and smarter handling of quiet days.

• Per-entity blocks — each tracked person, topic, and platform you follow gets its own block in the daily email, with a Sonnet 4.6 summary plus every supporting link
• New sources added: GDELT (global news monitoring), MediaCloud, and the Internet Archive’s TV News Archive
• Skip-regen optimization — if nothing meaningfully changed since yesterday, the AI summary regeneration is skipped so you’re not paying for noise
• Empty-state honesty — if a tracked entity had no activity, the block says so directly instead of inventing filler

Canvassing — Major UX Overhaul, Kansas Now Live

A wave of improvements to make precinct picking, route planning, and door knocking faster, plus a sixth state online.

• Kansas voter file is live: 2 million voters and 4,240 precincts, with the same picking, filtering, and routing as FL, OH, CO, NC, and PA
• City-aware precinct picker shows the dominant city for each precinct before you select it
• Single-slider voting filter (0–3 of last 3 primaries) replaces the four-checkbox version
• Sequential Google Maps navigation — each leg of your route opens individually instead of dumping all stops at once
• Target-stops route optimization lets you tell the optimizer how many doors you want to hit, not just a time budget
• Single-click info, shift-click to select — no more accidentally adding houses to your turf
• Visit log on map pin click — quick history without leaving the map
• Cancel turfs from “My turfs” without needing an admin

Big Book — 11 New Sections, Wayback Recovery, and Philanthropy Data

The biggest single-week expansion of Big Book research coverage we’ve shipped, plus a dedicated pipeline for surfacing deleted tweets.

• 11 net-new research sections wired into the automated pipeline: family background, endorsements, organizational affiliations, personal political contributions, local issue stances, democracy and elections stances, judicial sentencing patterns, personal financial disclosure parser, judicial contribution cross-reference, Parler social media analyzer, and contact and address history
• Wayback post-recovery surfaces deleted tweets directly in section research, routed through Bright Data Web Unlocker for reliability, with heartbeat monitoring on long-running recovery jobs
• Philanthropy data integrated into Big Book research: IRS 990, Million Dollar List, and Forbes philanthropy — with a fixed item-count ladder and corrected Decimal serialization in JSON exports

Helios & War Room — Chat with the Full Book, Video Tab, and Vault Auto-Save

Bigger context, a redesigned video studio, and quality-of-life improvements across the Command Center.

• Chat with the full Big Book and Short Book on Helios — the AI now sees every section as context
• Video tab in the book viewer (replaces the old drawer) with duration dropdown, pacing controls, text styling (color, size, position, stroke), and AI-generated background music via ElevenLabs (instrumental-only, mood-matched)
• Vault auto-save — completed Big Books and Campaign Videos auto-save to your Vault on completion, no more clicking “Save to Vault” per item
• Inline session naming — rename your War Room sessions in place; new sessions get an AI-generated name within seconds
• Tighter, more cited answers — Ares (the War Room chatbot) cites every source as a clickable link, with a stricter “lead with the answer, not the bio” prompt
• Faster multi-tool questions — when your question needs People Finder, Social Media, and Donor data, those tool calls run in parallel
• Smarter category routing replaces 23 flat tools with 6 category routers, so the chatbot picks the right data source faster and stops falling back to web search when we already have the answer

Social Media Analysis — Rumble, Deeper YouTube, Objective-Driven Synthesis

• Rumble is now a first-class platform in unified analysis and the Big Book — pulls videos with full auto-caption transcripts, routed through proxy for reliability
• YouTube coverage doubled — the default per-subject pull raised from 500 to 1,000 videos
• Objective-driven synthesis — when you specify what you’re looking for, per-platform and cross-platform summaries lead with that objective instead of generic biographical filler

Federal Data — Three New Sources

• House Disbursements — quarterly Statement of Disbursements joined to the live House roster (~900 members), so you can see office spend, vendor relationships, and travel for any sitting Representative (initial rollout in progress)
• Congress.gov bills — a searchable bill table with summaries and policy-area filters, wired into the Voting chat for quick “what bills did X introduce?” questions
• Regulations.gov and Federal Register public comment ingest, available for compliance and policy research

Compliance — OFAC Screening and Employment Overlay

• OFAC sanctions screening — Treasury SDN list checked on every batch finance run, with a persistent local cache
• Employment overlay automatically threads employment data into the verifier when it’s available
• Curated demo data — a single button seeds a complete demo run with every issue category for product walkthroughs
• FEC validation errors now surface to the UI instead of being silently swallowed

AI & Performance

• Claude Opus 4.7 ready — flip-of-a-switch upgrade once you opt in, with citations parsing correctly across all four formats: tag, paren-wrapped, malformed hybrid, and markdown link
• Parallel tool execution with per-tool timeouts so a slow source can’t hold up the others

Reliability & Operations

• Briefing async loop fixed — addressed one of the underlying causes of stale 8 AM digests
• IRS 990 ingest hardened — duplicate “Completed” emails suppressed, hung-sync recovery now works correctly
• Background workers tuned to survive pod restarts so long-running jobs no longer hang silently across deploys
• Daily ETLs run autovacuum and a post-load ANALYZE on 19 high-traffic tables so query plans stay fast as data grows

Campaign Video Generator — New in War Room

End-to-end AI video creation for campaign attack and promo spots is now live in the War Room, replacing the old drawer UI with a dedicated Video tab.

• Two-step editable script generation sourced from your Vault citations, so every claim in the spot is backed by research you already own
• Test and Production modes with monthly credit limits — preview spots before burning credits on final renders
• Automated imagery pulled from Pexels, Wikimedia Commons, your opponent’s social profiles saved to Vault, and cached TikTok, Instagram, and X screenshots
• Video clips supported alongside still images on the timeline, with Ken Burns pan-and-zoom, pacing controls, add-scene, and a resizable config panel
• Background music by mood (instrumental-only, FEC-safe) and text styling with color swatches, custom hex, size, position, and stroke
• Manual image upload, Save to Vault, Past Videos list, and auto-applied FEC disclaimer plus AI-disclosure overlay on every export

FOIA War Room — Redesigned with AI Assistant

The FOIA experience is now a two-panel resizable layout with a chat assistant on the left and a jurisdiction sidebar on the right, replacing the old tabbed view.

• AI chat assistant with starter prompts, OCR document context, and built-in FOIA legal knowledge — ask questions about your request or the law and get grounded answers
• Multi-agency batch filing sends the same request to multiple agencies in one click
• Federal plus all 50 states pre-loaded with statutes, deadlines, fee waivers, and appeal paths
• Per-jurisdiction cost and turnaround estimates in the sidebar, with automatic cost estimation by scope (small / medium / large)
• Deadline reminder emails 3 days before, day of, and 1 day overdue so nothing slips
• Inline letter editing, document upload and download, send-test-email, and a reopen flow for follow-up requests
• Entitlement-gated with a demo page at /foia for organizations that don’t have access yet

Fundraising — Donor Dedup, Civly Express, and Sanctions Screening

A major expansion of Fundraising covering donor deduplication, contribution limits, one-click giving, and compliance.

• Per-campaign donor deduplication across different emails using a 5-signal score (name, phonetic match, city, employer, zip), so one donor across multiple inboxes rolls up to a single profile
• FEC contribution-limit checks now aggregate across linked donor profiles, not per-email — no more accidental over-the-limit contributions from split profiles
• Per-campaign state and federal contribution limits with different caps per race type, applied automatically
• New Aggregate YTD column in the donations table and in CSV exports, plus donor cluster IDs and linked emails; backfilled across historical donations
• Linked donor badge and tooltip in the UI, with a manual unmerge option for false positives
• Civly Express: returning donors get saved payment methods for one-click giving
• Daily OFAC SDN list screening — automatic sanctions check on every donation, refreshed each morning
• Expense tracking tab with a Stripe fee breakdown, and a full donation export CSV with expanded columns
• Multi-campaign data isolation on every fundraising query, so campaigns under the same org can’t see each other’s donors
• IP geolocation now reads the real donor IP behind Cloudflare for accurate location-based checks

Big Book & Helios Chat — New Data, Claim Verification, and Inline Editing

Big Book gains a new consumer-complaints data source, deterministic quality safeguards on Short Books, and chat-driven editing from Helios.

• CFPB Consumer Complaints added as a new Big Book section — 14.5M+ records searchable by employer, with top issues, products, untimely-response rate, and dispute rate
• Claim Verification appendix on every Short Book classifies each Candidate Bio entry as Verified, Self-Reported, or Unverified, and points to the authoritative source to confirm (state bar, FINRA BrokerCheck, DD-214 FOIA, university registrar, and more) — directly addresses the Santos-style fabricated-bio risk
• Priority Findings auto-appendix: any finding with a .gov, .mil, or CourtListener URL, or 3+ independent source URLs, is appended to every Short Book automatically — high-confidence material can’t get silently dropped
• Citation Map fix restores four previously-missing sections (meeting minutes, dark web exposure, Reddit analysis, personality profile) in Short Book prose
• Deterministic bibliography as a separate downloadable output — dedupes and classifies every source URL across the full report
• Big Book pipeline upgraded to our latest research model, with instant rollback controls and automatic fallback if the primary model is unavailable
• Completed Big Books now auto-save to Vault, and you can chat with any saved Big Book or Short Book directly from the Helios page
• Helios chatbot can now edit the Short Book inline via tool use — the editor updates live and auto-saves
• Vault context budget scales by plan tier — enterprise customers get more report content fed into each chat turn
• Fetch reliability overhaul: smarter fallback ladder for source pages, per-book spend cap, shared circuit breaker to stop wasting time on rate-limited archives, and SEC and OSHA hardening
• Fixed a 422 error that was breaking Big Book chat for some users, and silent Vault auto-save failures on completed books

Research Tools — New Tabs and Faster Person Search

• New Social Media tab in the War Room — unified analysis embedded directly, pre-fills the target name, and results save to Vault and feed into chat
• New PI Finder tab in People Finder: search licensed private investigators by state, city, and specialty; returns contact info and license numbers
• Person Search now runs on a unified contributions table across every state and FEC — one query hits all jurisdictions, with parquet-backed results and a Postgres fallback for much faster large result sets
• Better matching for committee names (PAC / committee vs. individual donor)
• Apollo people search and email reveal restored for journalist search
• Vault tab tag filtering and tag badges restored

Donor Prospecting — Philanthropic Giving (New)

Charitable giving now shows up alongside political giving on every donor profile.

• Philanthropic profile card on donor detail pages covering charitable contributions
• New “Has philanthropic gifts” filter in Donor Prospecting to surface $100K+ givers
• Cross-source donor matching upgraded with probabilistic matching, LLM-assisted dedup, nickname and city normalization, and phonetic blocking — consolidated profiles across spelling and data-source variants

New and Expanded Data — 15.7 Million Records Added

The largest single-release data expansion to date, covering nonprofit filings, consumer complaints, and three new jurisdictions of campaign finance.

• CFPB consumer complaints: 14,591,382 records (wired into Big Book as a new section)
• San Francisco city campaign finance (1998–present): 486,059 contributions and 179,189 expenditures
• Houston, TX city campaign finance (2020–2026 e-filings, including scanned PDFs): 111,851 contributions
• Wyoming state campaign finance (ETL + UI + person search + parquet): 66,734 contributions
• Philanthropic gifts unified table: 80,866 verified gifts from Million Dollar List, Forbes, 990 officers and PF contributors, and university and museum donor rolls — powers the new philanthropic profile card
• IRS 990 Schedule I grants (nonprofits to other organizations): 138,067 new grants
• IRS 990 GivingTuesday years 2011–2020 backfilled into the existing 990 tables (previously 2021+ only)

Reliability & Operations

• State campaign finance weekly syncs moved to a dedicated fast queue — no longer starved by long-running FEC, SEC, LegiScan, and 990 jobs
• Repaired silent ETL failures on MA, NE, GA, MI, and UT state campaign finance syncs that had been failing quietly for weeks; MA and NE reloaded 105K and 33K records, and GA / MI / UT self-heal on the next Sunday run
• Memory-heavy tasks (parquet exports, mini-books, SMS, outreach, dialer, 990 parsing, and prospecting) moved to a dedicated high-memory worker pool
• Bulk parquet exports no longer time out on large datasets, with a 4-hour window, atomic writes, and auto-detected column types
• ETL status emails now correctly report “partial” when any table fails (previously misreported as “success”)
• Rolling deployments of the Fundraising service no longer double-send donation emails
• Enterprise customers are now auto-granted all enterprise-only features — no more manual toggling

Big Book — 8 New Data Sources and Automated Quality Checks

Opposition research reports now draw from eight additional public record databases, with a new verification engine that flags unsupported claims and broken links before delivery.

• Glassdoor employer reviews for campaign committees surface staff treatment, workplace culture, and management red flags from people who actually worked on the campaign
• IRS 990 nonprofit filings now included, with officer and director names, titles, and detailed compensation breakdowns from Schedule J
• FARA foreign agent registrations pull from the DOJ database, showing foreign principals, registered agents, and lobbying activity tied to your subject
• OGE 278e financial disclosures cover Senate-confirmed Presidential appointees — assets, income sources, and potential conflicts of interest
• SBA PPP loan records — 12 million loans searchable by name and state, surfacing forgiveness amounts and business connections
• LD-203 lobbying contribution reports show political contributions made by registered lobbyists to or on behalf of your subject. All lobbying data has also been migrated to the new lda.gov API ahead of the legacy system’s June 2026 sunset
• Mobilize campaign event monitoring tracks opponent ground-game activity — canvassing, phone banks, and rallies — within 150 miles of the state capital
• Automated citation verification cross-references every source link in the report, flagging homepage redirects, broken URLs, and uncited factual claims before the report reaches you

Local Government Meeting Minutes — CivicClerk Added

A third meeting minutes data source is now live alongside Legistar and Municode, expanding municipal coverage with automated weekly collection.

• CivicClerk adds 11 new municipalities with AI-powered minutes parsing to surface relevant policy decisions and public comments

Command Center — AI Research with Live Data Access

The Command Center now connects directly to 17+ live data sources. Ask questions in plain English and get verified answers with charts, tables, and citations — no file uploads required.

• Instant sessions: click “New Session” and start asking questions immediately — the AI figures out which database to query based on your question
• 17 searchable data categories including Congress members, voting records, state legislation, campaign finance, FEC contributions, stock trades, court records, election results, government contracts, EPA compliance, OSHA inspections, PPP loans, meeting minutes, and local legislation
• Research tab runs on-demand Helios deep research without leaving your session, with results feeding directly into the conversation
• Smart cross-linking connects your questions to other Civly tools — social media profiles link to Social Media Analysis, journalists link to People Finder, and donor questions link to Donor Prospecting
• Campaign Video Generator creates campaign-ready video ads with AI-generated scripts, two-step editing, and automatic FEC disclaimer and AI disclosure compliance
• Data Sources overview shows every database available to the AI with example questions for each
• Rebuilt query engine delivers faster, more accurate results backed entirely by Civly’s verified data

Reliability Improvements

• Twitter and FEC data sources now handle missing profiles and empty records gracefully instead of interrupting report generation
• LinkedIn profile discovery is more accurate, with better URL validation and improved fallback behavior
• Removed two data sources that were no longer functional to keep reports clean and accurate
• Reports now run up to 3 concurrently for faster turnaround during peak usage

Helios Deep Research — Editable Reports, QA, and Political Ads

Research reports are no longer read-only. You can now edit any section directly in a rich text editor before exporting.

• Edit reports with a full WYSIWYG editor, leave inline comments, and track changes so your team can review and refine together
• Built-in QA automatically flags unsupported claims, missing citations, and gaps in coverage so nothing slips through
• Export in three formats: a full Comprehensive report, a Brief executive summary, or a Vulnerability Scan with opposition-ready highlights
• Political ad spending from Google and Meta is now woven into campaign history, donor research, controversy, and media analysis — drawing from 19,500 advertisers, 1.5 million ad creatives, and 2.4 million spend records
• Political email archives from politicalemails.org are now included as a research source, filtered to emails actually sent by the subject
• Every Deep Research page now includes a methodology section listing all public data sources with links and explaining how they’re cross-referenced
• New city-level targeting and free-text context fields let you narrow reports to specific jurisdictions or policy areas

Court Records Search (New)

Search across 10 million federal court opinion clusters in one place, with filters for court, judge, and keywords.

• A dedicated search page lives under Background Research with severity badges, sortable columns, and source type indicators
• Court Records tabs now appear on both federal and state person search profiles
• Results combine our local database with live CourtListener API docket searches for broader coverage
• Fast fuzzy matching on 10 million case names so you can find what you need even with partial or approximate terms

Local Government Meeting Minutes — 80 New Municipalities (93 Total)

We expanded meeting minutes coverage from 13 to 93 municipalities, pulling from three data sources: Legistar, Municode Meetings, and CivicClerk.

• 59 new Legistar municipalities added through automated discovery and validation
• Municode Meetings API adds a second source for meeting minutes
• CivicClerk API adds a third source covering 119 portals with 164,000 events, 98,000 minutes PDFs, and 32,000 video recordings
• All sources are scraped on a weekly automated schedule so data stays fresh
• 8 new high-value SLEA municipalities added: Detroit, Genesee County, Cumberland County, Guilford County, Winston-Salem, Greensboro, Wilmington, and Arapahoe County

Entitlements and Access Changes

• Voting Records, SEC Explorer, Election Explorer, and Congress Trades are now free for all users
• Compliance and FEC Filing now require entitlements, with a demo page for organizations that don’t have access yet
• Organizations can now have monthly book generation limits for enterprise plans

Big Book — Faster Reports, Deeper Research

Opposition research reports now generate faster and pull from more sources.

• Report generation no longer slows down during peak usage, with dedicated infrastructure for the heaviest queries
• Federal court records now included: lawsuits, legal proceedings, and case severity are automatically surfaced in career, financial, and controversy sections
• More reliable LinkedIn and Facebook data collection with upgraded scraping infrastructure

Judicial Intelligence (New) — 22.7 Million Records

Comprehensive federal judge research available across the platform, built on one of the largest legal datasets we’ve integrated.

• 10.7 million court opinions linked to 10 million opinion clusters for full case law coverage
• Judge financial disclosures for 32,500+ filings, broken out across 2 million+ line items: investments (1.9M), positions (37K), agreements (10K), debts (18.8K), gifts (2K), reimbursements (33.5K), non-investment income (15.3K), and spousal income (20.2K)
• Judge profiles with biographical data, appointment history, and court assignments from the Federal Judicial Center
• All data cross-referenced into unified judge profiles and searchable opinions for quick lookup

Local Government Meeting Minutes (New) — 201,000+ Records

AI-powered scraping and analysis of local government meetings, surfacing relevant policy decisions and public comments. Currently covering 13 municipalities with 5,000+ meetings, 189,000 agenda items, and 7,300 public statements parsed.

State Personal Financial Disclosures (New) — 44,700 Filings

Look up elected officials’ personal financial disclosure filings at the state level, starting with North Carolina.

State Campaign Finance — 18 New Jurisdictions (33 Total) — 15.6 Million Records

More than doubled coverage with full search, filters, and an AI research assistant for each. Total state campaign finance database now includes 12.9 million contribution records and 2.7 million expenditure records.

• Utah, Hawaii, West Virginia, Maine, Dallas, LA City, Oregon, Idaho, Philadelphia, DC, Virginia, Kentucky, Georgia, Missouri, New York City, Massachusetts, Tennessee, and Nebraska

Donor Prospecting — Instant Search

Prospecting is now real-time instead of batch. Results appear instantly as you search.

• Save individual donors or an entire result set to your CRM in one click
• Filter by donor type and mobile phone availability
• Stats banner showing total enriched records across your database
• Already-saved contacts are automatically excluded so you only see new prospects

CRM Enhancements

• Contribution history now visible on contact profiles so you can see giving patterns before reaching out

Power Dialer

• Calls now come from local area codes, automatically purchasing numbers as needed for better pickup rates

Coming Soon: SMS Campaigns

Text message outreach with AI-personalized messages, local area code sending, TCPA compliance, and CRM integration. Work is underway.

Big Book — More Data, Smarter Analysis

The opposition research report now pulls from significantly more sources and delivers deeper insights.

• Five new data sources: state campaign finance records, congressional net worth, congressional stock trades, state legislation, and Reddit activity
• Instagram image analysis with two-tier controversy detection
• Big Five personality scoring from social media posts with evidence citations
• Support for multiple social handles per platform (e.g., campaign + personal accounts both get scraped)
• Full Substack article content now included instead of just previews
• Improved reliability: fixed silent section failures so no data gets quietly dropped

Power Dialer

A built-in calling tool for donor outreach.

• AI-personalized call scripts generated per contact
• Generate-and-review flow so you can edit scripts before calling
• Funnel view showing contact progression through outreach stages
• Contact history and call activity tracking
• Followup email sending after calls
• Script editing and customization

Civly Connect CRM — Continued Build-Out

New capabilities layered onto the CRM.

• Google Calendar integration: schedule and automate donor meetings directly from contact profiles
• Bulk enroll contacts into email sequences from the contacts list
• Sequence stage visibility on contact detail pages
• Toolbar filter bar on contacts list for quick filtering
• Email sequences and Power Dialer work together for automated multi-channel followup
• Auto-unenroll contacts from sequences on bounce, spam complaint, or unsubscribe

State Campaign Finance — 6 New States (15 Total)

Expanded from 9 states to 15 with full ETL, API, AI chat, and frontend for each.

• Oklahoma, Iowa, New York, Wisconsin, Connecticut, and New Jersey added
• Each state includes searchable contribution data, filters by occupation/employer/state, and an AI chat assistant

People Search — Expanded Coverage

Building on the unified profile with additional data and state-level records.

• State person search with voting, finance, and pay-to-play data
• Async pay-to-play analysis with net worth tab and name alias support
• Pay-to-play narrative summaries with contract details

Contact Enrichment — Addresses & Phone Numbers

Find mailing addresses and phone numbers for donors at scale.

• Batch address lookup and enrichment from the prospecting pipeline
• AI-powered address search integrated into donor profiles
• Voter registration data feeding into address matching
• AI-powered phone search and validation for better contact coverage
• Webhook-based enrichment for real-time phone number capture

Polling & Race Tracker

• Race catalog expanded to 60 races with full coverage
• Email notifications when new polling data syncs
• Improved daily polling check that scans the full catalog

Mobile App

• Early-stage mobile app with authentication and core navigation in place

Civly Connect — Your Built-in CRM

A brand new contact management system right inside Civly. No more switching to HubSpot or spreadsheets.

• Contacts list with search, filters by status, and source tracking
• Contact detail pages with notes, activity history, and tasks
• Donor dossiers on contact profiles showing giving history, party alignment, key topics, news, and personality insights
• Auto-sync with Prospecting: save a donor and they become a CRM contact automatically
• Bulk import from saved prospects, dossiers, or CSV files

Unified Social Media Analysis

One tool to search and analyze social media across Facebook, Truth Social, YouTube, and more.

• Search by person or by topic across platforms
• Get a briefing-style summary highlighting cross-platform themes
• Save analyses to your vault with citations for later reference

People Search — State and Federal Data in One Place

A unified profile pulling together state and federal records:

• Federal profiles with five tabs: voting record, FEC contributions, stock trades, election history, and pay-to-play
• State campaign finance data from 7 states (AL, CA, CO, MI, NC, PA, TX) alongside federal FEC records
• Party alignment percentages and improved committee lookups using FEC candidate linkage

Briefing Room Improvements

• Action cards now display relevant images alongside each briefing item, making it easier to scan and recognize stories at a glance
• Improved image quality: filtered out low-resolution cached thumbnails so you only see clear, relevant visuals
• Better fallback handling when images aren't available, so cards still look clean

Simplified Help and Chat

The individual chatbots from across the platform are now consolidated into a single help button on the side of the screen. One chat widget, available from any page, that adjusts to where you are and what you need help with.

Donor Intelligence & Prospecting

• Donor Intelligence Hub: unified view of enriched donors with dossier generation, chat, and social posts
• Contact Finder pipeline with Civly enrichment, weighted party scoring, and web search fallback
• Saved contacts with save/unsave, bulk save, and enrichment status tracking
• Mini Book dossier generation with citations, sources, personality traits, and AI chat interface
• Bulk dossier generation for batch processing hundreds of donors at once

Pay-to-Play Analysis

• Government contracts command center: cross-reference 10.6M federal contracts against campaign donors
• Narrative summary with expandable donor and contract details
• Committee typeahead search across FEC and state campaign finance data

State Campaign Finance

• Colorado TRACER campaign finance ETL with contributions, expenditures, and classification
• Donor data cleaning pass 2: suffix stripping, regex employer junk detection, and SQL helper functions
• Donor classification: persistent person/committee/unknown type on 6.4M+ donor profiles
• Pre-computed sub-matviews for 10x faster donor profile rebuilds

State Legislation

• Expanded to all 50 states (added 32 states including NY, IL, NJ, WA, OR, CT, MA, MD, and more)
• 10x faster matview rebuilds with single-pass unified JOIN strategy

Race Tracker

• Race tracker with 60+ races from RealClearPolitics coverage
• Poll data with pollster details, sample sizes, and candidate spreads

Research & Intelligence

• Content scorer with research-backed criteria and dual-audience clarify flow
• Bill topic classifier, legislator scoring, and profile cards
• Substack full article fetcher via slug API
• Social discovery with inline Serper API search

Platform

• Dark theme support with toggle
• Issue submission form with screenshot upload
• Improved CSV export error handling across all features
• Learning Center guides for YouTube Analysis with annotated screenshots

State Campaign Finance

• Texas campaign finance: 34M+ contributions and 5M expenditures searchable with filters
• Michigan campaign finance with ML-powered occupation and employer classification
• Alabama and North Carolina campaign finance data with AI chat analysis
• California campaign finance models and ETL pipeline
• Cascading filter options and autocomplete across all state finance tables

Government Data

• Congressional stock trade explorer with Senate eFD scraper (2021–2026)
• House financial disclosure parser with OCR fallback for scanned PDFs (2014–2020)
• State legislation expanded to 19 states (added TX, VA, IA, CO, MN, NH, NV, SC + session backfills)
• Browsable federal votes table with column-header filters
• Election explorer with searchable column-header filters

Research & Intelligence

• Big Book: government contracts pay-to-play analysis
• War Room: dedicated Polling tab with custom configuration

Learning Center

• New Learning Center with per-feature step-by-step guides
• Annotated screenshots with highlights for Social Media Analysis and Research War Room

Performance & Reliability

• Materialized view auto-ANALYZE after refresh for consistent query performance
• Centralized matview refresh utility with SEC matview support
• Optimized social alerts dashboard and briefing loading
• Deploy gate for concurrent development session safety

Research & Analysis

• Multi-state legislation support: California, Vermont, and Alabama
• Unified Instagram and TikTok video analysis with AI-powered filtering
• Interest group scorecard ratings for legislators

Data & Intelligence

• Redesigned FEC Explorer with server-side pagination and aggregate totals
• SEC insider trading details and red flag detection
• Cross-platform topic tracking and Reddit trending detection

Collaboration & Workflow

• Shared vault with full transcripts and article content in saves
• War Room citations resolve to clickable links
• Chat links open in new tab with smarter currency formatting

Platform & Security

• Two-factor authentication and Azure AD login support
• Rate limiting and access control hardening
• Upgraded AI models and faster data pipelines

Social Alerts

• Automated refresh pipelines with configurable auto-refresh per person or topic
• Bulk refresh to update all tracked entities at once
• Live progress indicators showing collection and analysis status on each card
• Consolidated digest emails — one summary per refresh cycle instead of per-alert
• Per-user trending detection so every user receives their own trending alerts

Big Book (Opposition Research)

• Upgraded AI models for higher quality research and writing

Social Media Analysis

• Faster, more efficient data pipelines for Reddit and Substack monitoring

Platform Reliability

• Smarter AI retry logic with automatic model fallback during high-traffic periods
• Improved background worker architecture for faster, more stable alert processing

Big Book (Opposition Research)

• Deep research now pulls LinkedIn, Wikipedia, MIT Election Data, and property records
• Choose a political angle (left, right, primary challenge, or neutral) for framed research
• Enhanced AI models for higher quality writing
• Hyperlinked citations, Wikipedia sources, and full bibliography
• New sections: political donations, property records, Truth Social, Path to Victory
• Professional prose formatting; better support for lesser-known candidates

Social Alerts

• Monitor up to 10 candidates across YouTube, Substack, and more
• AI-categorized alert severity with review/dismiss/escalate workflow

Data Breach Check

• Screen individuals against known data breaches

Election History & Explorer

• Browse historical election results and explore race/candidate data

FEC & Campaign Finance

• PAC contributions view in FEC Committee Explorer
• AI-powered queries for more accurate data lookups

Social Media Analysis

• Toggle to include/exclude replies; advanced Twitter search with auto-retry
• Facebook public page scraping for Big Book research

Command Center MCP

• MCP server lets external AI tools interact with Civly's research capabilities

Map political networks, flag controversial content, and generate professional reports in minutes.