Design Considerations: Presidential Campaign Data Warehouse

[mrtwebdesign]

see the Glossary in the comment following this post.

Design Considerations: Presidential Campaign Data Warehouse

Purpose of this document. This is not an architecture. It is a checklist of constraints that any architecture has to satisfy. It exists so that (a) whoever builds this can be asked the right questions early, and (b) a proposal that ignores these constraints can be identified as incomplete before money is spent on it.

Audience. Technical leadership and non-specialist decision-makers. A companion glossary defines every term used here.

---

Contents

1. Why a generic data warehouse design fails in this context — the five structural differences from a commercial warehouse: fixed deadline, statutory boundaries, nation-state threat model, explosive-then-zero headcount, licensed rather than owned data.
2. Legal and regulatory considerations — entity separation as an architectural boundary; state-by-state voter file restrictions; consent as a data model; post-campaign records retention.
3. Data acquisition and licensing — existing party data infrastructure; the three incompatible ingestion patterns; refresh cadence over the cycle; primary vs. general.
4. Identity resolution and modeling — the person-ID layer, model governance, experimentation, and activation (reverse ETL).
5. Architecture and operations — immutable raw data, the compute cost curve, Election Day as a single point of failure, non-SQL consumers, not rebuilding what exists.
6. Security — documented nation-state targeting, misconfiguration as the largest known incident, insider access, field minimization, surge-scale provisioning, minimum controls checklist.
7. Lifecycle and wind-down — archive, legal hold, disposition of licensed data, and platform-agnostic infrastructure.
8. Questions to put to the domain expert — 15 questions across legal/structural, data, analytics, operations, and build-vs-buy. Intended as the agenda for that conversation.
9. Warning signs in a proposal — the checklist for identifying an under-considered proposal before money is spent.
10. What this document is not — scope boundary and next step.

Sources

---

1. Why a generic data warehouse design fails in this context

A presidential campaign data warehouse differs from a commercial one in five structural ways. Each of these changes the design, not just the operations plan.

Property	Commercial warehouse	Presidential campaign warehouse
Deadline	Roadmap slips are absorbed	Election Day does not move. A feature that ships November 4 has zero value.
Legal boundaries	Internal policy, contractual	Federal statute. Certain groups are legally prohibited from seeing each other's data, and violations are enforcement matters.
Threat model	Financially motivated criminals	Nation-state intelligence services, plus criminals, plus insiders leaking to press.
Org size	Stable headcount	Grows from dozens to thousands, then dissolves entirely within weeks of Election Day.
Data ownership	You own your data	Most of the core dataset is licensed, under 50+ separate agreements with different use restrictions.

The last two are the ones most often missed. A design that assumes stable staffing and owned data will fail on both counts.

---

2. Legal and regulatory considerations

2.1 Entity separation is a hard architectural boundary, not a permissions setting.
A presidential operation involves multiple legal entities: the principal campaign committee, one or more joint fundraising committees, the national party committee, state parties, and — separately and at arm's length — super PACs and 501(c)(4) organizations. FEC regulations define when communication between these entities becomes a "coordinated communication," which is treated as an in-kind contribution and is generally illegal for a super PAC. FEC rules contemplate a formal firewall: it must prohibit the flow of information between the employees or consultants serving one side and those serving the other, and it must be described in a written policy statement distributed to all affected employees, consultants, and clients.

• Implication: the warehouse may need to be multiple warehouses, or a single platform with enforced tenancy that a compliance lawyer will sign off on.
• Implication: audit logs are not just a security control. They may become legal evidence that the firewall held.
• Get General Counsel into the design conversation before the schema is written, not after.

2.2 Voter file use is restricted state by state.
State-level laws governing the use, storage, and transmission of voter files vary significantly and can impose strict limits on third-party uploads, data matching, and cross-platform activation. Some states prohibit commercial use; some prohibit re-disclosure; some require signed use agreements naming specific individuals.

• Implication: the warehouse needs to track provenance and license terms per row or per source, and be able to answer "can this record legally be sent to this ad platform?"
• A design that merges all 50 states into one undifferentiated table and forgets where each row came from cannot answer that question.

2.3 Contact consent is a data problem, not just a compliance checkbox.
Political campaigns are not exempt from the Telephone Consumer Protection Act (TCPA), a position the Supreme Court affirmed in Barr v. American Association of Political Consultants. Violations carry penalties of $500 per call or text, trebled to $1,500 for willful violations — attractive territory for class-action plaintiffs. Presidential campaigns have been named in TCPA class actions.

• Implication: consent provenance (who consented, when, through what form, to what) must be stored, versioned, and queryable at the individual level, and it must survive the trip through the warehouse into the texting platform.

2.4 Records retention obligations outlive the campaign.
FEC recordkeeping requirements apply after the committee winds down. The warehouse design must include a defensible archive, not just deletion.

---

3. Data acquisition and licensing

3.1 You are not the only data operation in your ecosystem.
National party infrastructure already exists on both sides. On the Republican side, Data Trust, founded in 2011, describes itself as the core infrastructure provider for conservative campaigns and says it supported more than 5,000 Republican campaigns in the 2024 cycle. On the Democratic side, the Democratic Data Exchange (DDx), founded in 2019, operates a real-time blind exchange of voter contact data between aligned organizations, parties, committees, and campaigns. Separately, Catalist and TargetSmart function as major voter-file managers for progressive organizations, while state parties maintain and license their own files.

• Implication: interoperability with party infrastructure is a requirement, not a nice-to-have. Design decisions that make it hard to exchange data with these systems have a political cost, not just a technical one.
• Implication: someone has to negotiate and track data-sharing agreements. That is a role, and it has a technical counterpart.

3.2 Source data arrives in incompatible shapes and cadences.
Roughly three ingestion patterns coexist:

1. REST APIs — CRMs, fundraising, events, comms tools. Rate-limited, paginated, inconsistently documented.
2. Bulk file drops — SFTP or object storage. Voter files, vendor matchbacks, early-vote files. Delivered as full snapshots, not deltas, and large.
3. Warehouse-native shares — some vendors publish directly to warehouse marketplaces; L2, for example, makes its national databases available through the Snowflake Data Marketplace.

Any proposal that assumes a single ingestion pattern is underspecified.

3.3 Refresh cadence changes over the cycle.
Weekly refreshes are adequate in the spring. In the final three weeks, early-vote and absentee-return files drive daily — sometimes hourly — resource reallocation. The system has to support both without a rebuild.

3.4 Primary and general election are different systems in practice.
The primary is a sequential, state-by-state contest with different rules per state. The general is a national contest concentrated in a handful of battlegrounds, with party coordination that did not exist during the primary. Data models built exclusively for one phase tend to break at the transition.

---

4. Identity resolution and modeling

4.1 Identity resolution is the hardest problem in the build.
A donor record, an email subscriber, a canvass contact, a text opt-in, and a voter-file row all describe the same person and share no common key. Entity resolution — also called record matching or identity resolution — is the technique that identifies and matches records across multiple sources to produce a single unified record, and it is the engine underneath any master data management program.

• Implication: a persistent person-ID layer with match confidence scores, not a one-time deduplication script.
• Implication: matching decisions must be reproducible and reversible. Bad merges are worse than missed merges, because they corrupt downstream targeting and fundraising compliance.
• Vendors sell matching as a service; evaluate build-vs-buy explicitly here.

4.2 Models drive resource allocation, so model governance is a real requirement.
Turnout propensity and support scores determine where field staff, money, and the candidate's time go. That makes model versioning, reproducibility, backtesting, and score lineage operational necessities, not data-science hygiene.

4.3 The campaign will run experiments.
Randomized controlled trials on messaging and contact methods are standard practice. The warehouse has to support treatment/control assignment, holdout groups, and outcome measurement — which means it must record who was assigned to what, not just who was contacted.

4.4 Activation is half the value.
Scores and universes are useless in the warehouse. They have to get back into the CRM, the texting platform, and the ad platforms. This is reverse ETL, and it is routinely under-scoped because it looks like an afterthought on an architecture diagram.

---

5. Architecture and operations

5.1 Preserve raw data immutably.
Voter files get re-cut; vendors reissue corrected files; models get retrained. If the raw landing layer is overwritten, past universes cannot be reproduced and disputes cannot be settled.

5.2 Plan for the compute cost curve.
Query volume and model training spike enormously in the final six weeks. Budget and quota design should anticipate this. A platform whose pricing model punishes bursty usage is a poor fit.

5.3 Election Day is a single point of catastrophic failure.
There is no "we'll fix it tomorrow." Disaster recovery, failover, and on-call rotation for the final 72 hours need explicit design, including a documented degraded-mode plan: what still works if the warehouse is down.

5.4 Most consumers are not SQL users.
State directors, field organizers, and finance staff need self-service reporting. The BI layer and the set of curated, trustworthy tables are as important as the pipelines.

5.5 Do not rebuild what the ecosystem already provides.
Parsons, a Python package maintained by The Movement Cooperative, exists specifically as a universal adapter for this ecosystem and normalizes data regardless of source. It ships connectors for VAN, ActBlue, Mobilize, Action Network, TargetSmart, Catalist, PDI, Hustle, Twilio, Redshift, S3 and dozens of others. Managed platforms exist too. Writing 40 bespoke API integrations is a common, avoidable, and expensive mistake.

---

6. Security

6.1 The threat model is nation-state, and this is documented, not speculative.
In 2020, Google's Threat Analysis Group reported that a China-linked APT group targeted Biden campaign staff and an Iran-linked APT group targeted Trump campaign staff with phishing emails. In 2024, an Iran-backed group used the compromised email account of a former senior advisor to phish a high-ranking official in a presidential campaign, one of dozens of individuals targeted across both campaigns. The FBI, CISA and ODNI issued a joint statement in August 2024 attributing a cyberattack on a presidential campaign to Iranian actors.

• Implication: phishing-resistant MFA and SSO for every account, including vendor accounts, from day one.
• Implication: assume credential compromise will be attempted successfully. Design for blast-radius limitation.

6.2 The most damaging known incident in this space was a misconfiguration, not an attack.
In 2017, a firm working for the RNC left 1.1 terabytes of unsecured personal data on a publicly accessible cloud server, exposing details on approximately 198 million registered voters — names, dates of birth, home addresses, phone numbers, registration details, and modeled ethnicity and religion.

• Implication: object storage configuration, public-access blocks, and continuous posture monitoring are first-class requirements. "We'll put the files in a bucket" is the exact failure mode.

6.3 Insider access is a real risk category.
In 2015 a Sanders campaign staffer accessed the Clinton campaign's voter file, which led to the DNC revoking the campaign's access and a public dispute. The system must be able to prove who queried what.

6.4 Minimize sensitive fields at ingest.
State voter registration records commonly include full or partial Social Security numbers, driver's license numbers, and dates of birth, depending on the state. Fields the campaign has no operational use for should be dropped at the boundary rather than stored and protected.

6.5 Access provisioning at surge scale.
Staff count grows by an order of magnitude and turns over constantly, including large volunteer and fellow populations. Manual access management will fail. Role-based access tied to an identity provider, with automated deprovisioning, is a design requirement rather than an operational refinement.

6.6 Controls checklist (minimum).
Encryption in transit and at rest; centralized secrets management for dozens of vendor API credentials; role-based access control plus row-level security by state and program; column masking on PII; full query audit logging with retention; vendor data processing agreements; documented incident response; independent security review before the general election.

---

7. Lifecycle and wind-down

• The committee dissolves. The obligations do not. Plan the archive, the legal hold, and the disposition of licensed data (much of which must be destroyed or returned per license terms) as part of the initial design.
• A stated design goal in this space is platform-agnostic infrastructure — a warehouse that lets an organization switch vendors between cycles without losing its history. That goal conflicts with how campaign technology is usually procured, and the conflict should be named explicitly at the start.
• Documentation is the deliverable that determines whether the next cycle starts from zero.

---

8. Questions to put to the domain expert

Legal / structural

1. What legal entities will touch this data, and where does counsel want the hard boundaries?
2. Which party data infrastructure are we expected to interoperate with, and what does that require of us?
3. Who owns the relationships and contracts with the voter-file vendors, and what do those licenses permit?

Data
4. Which source systems are in scope for cycle one, and which are deliberately deferred?
5. Are we buying identity resolution or building it?
6. What is the required data freshness in October versus in April?
7. What is the plan for consent provenance across SMS, email, and phone?

Analytics
8. Who are the actual consumers, and what is their SQL proficiency?
9. What models will exist, who builds them, and how are scores versioned and shipped back out?
10. What experimentation infrastructure is expected?

Operations
11. What is the on-call model for the final two weeks?
12. What is the degraded-mode plan if the warehouse is unavailable on Election Day?
13. What is the staffing plan — how many data engineers, and when do they start?

Build vs buy
14. What do we buy off the shelf (managed platform, matching, connectors) versus build?
15. What is the wind-down and archive plan?

---

9. Warning signs in a proposal

A proposal is likely under-considered if it does any of the following:

• Treats legal entity separation as a permissions problem rather than an architectural one.
• Contains no mention of identity resolution, or treats it as a deduplication step.
• Ingests voter file data without tracking source, license, and permitted use.
• Assumes a single ingestion pattern (all API, or all file-based).
• Has no reverse ETL / activation path back into operational tools.
• Has no consent data model.
• Presents a security section consisting only of "encryption and access controls," with no phishing-resistant MFA, no audit logging, no cloud posture management, and no reference to the nation-state threat model.
• Has no answer for Election Day availability or degraded operation.
• Proposes building all source integrations from scratch.
• Has no wind-down, archive, or data-destruction plan.
• Was designed without General Counsel in the room.

---

10. What this document is not

It is not an architecture, a vendor selection, a budget, or a staffing plan. It is the constraint set those things have to satisfy. The next step is a conversation with someone who has run a data operation on a national campaign before, using section 8 as the agenda.

---

Sources

• FEC, Coordinated communications — https://www.fec.gov/help-candidates-and-committees/candidate-taking-receipts/coordinated-communications/
• Mac Murray & Shuster, TCPA Compliance for Political Calls — https://mslawgroup.com/tcpa-compliance-for-political-calls/
• Wiley, TCPA Suits Against Political Campaigns on the Rise — https://www.wiley.law/newsletter-TCPA-Suit-Against-Political-Campaigns-Rise-Trump-Campaign-Facing-Separate-Class-Action-Suits
• UpGuard, The RNC Files: Inside the Largest US Voter Data Leak — https://www.upguard.com/breaches/the-rnc-files
• Dark Reading, Are 2024 US Political Campaigns Prepared for Coming Cyber Threats? — https://www.darkreading.com/remote-workforce/2024-us-political-campaigns-prepared-cyber-threats
• TechCrunch, Google: China- and Iran-backed hackers targeted Trump and Biden campaigns — https://techcrunch.com/2020/06/04/google-china-iran-trump-biden/
• Overture Global, The Secret Power of Political Data Trusts — https://www.overtureglobal.io/story/the-secret-power-of-political-data-trusts
• NASS, Public Voter Registration Information and Security — https://www.nass.org/sites/default/files/surveys/2020-10/NASS-briefing-FAQ-info-security-2020_0.pdf
• Data Trust — https://thedatatrust.com/
• Parsons (The Movement Cooperative) — https://www.parsonsproject.org/
• Senzing, What Is Entity Resolution? — https://senzing.com/what-is-entity-resolution/
• Sutton Smart, Legacy Data Archiving — https://suttonsmart.com/political-consulting/legacy-data-archiving-building-a-database-for-a-career-not-just-a-cycle/

[mrtwebdesign]

Glossary: Campaign Data, Analytics, and Compliance Terms

How to use this. Each entry gives a plain-language definition, a note on why it matters for a campaign data warehouse where that isn't obvious, and at least one link. The goal is that you can explain each term in one sentence, not that you can implement it.

Links verified July 2026. Vendor and organization links point to official sites; concept links point to reference or explainer pages.

---

Contents

1. Storage and architecture — warehouse, lake, and lakehouse; schemas and dimensional modeling; layering conventions; file and table formats.
2. Moving data around — ETL, ELT, and reverse ETL; pipelines, connectors, and orchestration; ingestion mechanics; lineage, governance, quality, and observability.
3. Platforms and tools — named products, grouped by what they do:
   • Warehouses and databases
   • Transformation and orchestration
   • Ingestion and activation
   • Business intelligence and reporting
   • Political-sector-specific platforms
4. Analytics and data science — BI and KPIs; models, scores, and features; drift and MRP; experimentation terms (RCT, holdout, lift, uplift).
5. Identity, matching, and enrichment — entity resolution, golden records, deterministic vs. probabilistic matching, householding, appends, matchbacks, hashing.
6. Political data terms — voter file, vote history, universes, turf, canvassing, GOTV, early vote, precincts, DMAs, P2P texting, plus systems and organizations (VAN, ActBlue, Catalist, TargetSmart, L2, i360, and others).
7. Legal and regulatory — FEC and FECA; committee types; coordination and firewalls; TCPA and consent; voter file use restrictions; PII and privacy statutes; retention.
8. Security — identity and access (MFA, SSO, RBAC, RLS); data protection (encryption, KMS, secrets, tokenization); threats (phishing, APT, insider, misconfiguration); assurance and recovery (SOC 2, pen test, DR/RTO/RPO).

Quick-reference: acronyms — every acronym in this document with its expansion, in one table.

---

Part 1 — Storage and architecture

Data warehouse — A central database that holds cleaned, organized data from many source systems, built for answering analytical questions rather than running an application. Think of it as a well-organized library rather than a filing cabinet.
→ https://en.wikipedia.org/wiki/Data_warehouse

Data lake — A large, cheap storage area that holds raw files in whatever format they arrived in, structured or not, without deciding in advance how they'll be used. You dump first and organize later.
→ https://en.wikipedia.org/wiki/Data_lake

Data lakehouse — A newer architecture that puts warehouse-style structure, governance, and fast SQL querying directly on top of data-lake storage, so you don't have to run two separate systems.
→ https://ibm.com/think/topics/data-warehouse-vs-data-lake-vs-data-lakehouse
→ https://www.knime.com/blog/data-warehouse-vs-data-lake

Object storage — Cloud file storage that holds arbitrary files ("objects") addressed by a key, rather than a traditional file system. Amazon S3 is the best-known example. This is what a data lake is physically made of.
→ https://en.wikipedia.org/wiki/Object_storage

Schema — The definition of what tables and columns exist and what types they hold. The structure of the data.
→ https://en.wikipedia.org/wiki/Database_schema

Schema-on-write vs. schema-on-read — Schema-on-write means you enforce structure when data goes in (warehouse behavior). Schema-on-read means you store it raw and impose structure when you query it (lake behavior).
→ https://www.knime.com/blog/data-warehouse-vs-data-lake

OLTP (Online Transaction Processing) — Databases built for many small, fast reads and writes — the kind an application does. PostgreSQL and MySQL are typical.
→ https://en.wikipedia.org/wiki/Online_transaction_processing

OLAP (Online Analytical Processing) — Databases built for scanning and aggregating huge amounts of data to answer analytical questions. Warehouses are OLAP systems.
→ https://en.wikipedia.org/wiki/Online_analytical_processing

Columnar storage — Storing data column by column instead of row by row, which makes analytical queries that touch a few columns across millions of rows dramatically faster.
→ https://en.wikipedia.org/wiki/Column-oriented_DBMS

MPP (Massively Parallel Processing) — Splitting a query across many machines that each work on a slice of the data. The reason cloud warehouses can scan billions of rows quickly.
→ https://en.wikipedia.org/wiki/Massively_parallel

Dimensional modeling — A way of organizing warehouse tables around "facts" (things that happened, like a donation) and "dimensions" (the things involved, like a person or a date). The dominant approach to warehouse design.
→ https://en.wikipedia.org/wiki/Dimensional_modeling

Star schema — The standard shape dimensional modeling produces: one central fact table surrounded by dimension tables.
→ https://en.wikipedia.org/wiki/Star_schema

SCD (Slowly Changing Dimension) — Techniques for handling attributes that change over time — a voter moves, changes party registration — so you can still reconstruct what was true on a given date.
→ https://en.wikipedia.org/wiki/Slowly_changing_dimension

Medallion architecture (bronze / silver / gold) — A convention for layering the warehouse: bronze = raw as-ingested, silver = cleaned and conformed, gold = business-ready tables people actually query. Naming varies; the pattern is universal.
→ https://www.databricks.com/glossary/medallion-architecture

Staging layer — The intermediate area between raw ingested data and finished models, where cleaning and standardization happen.
→ https://docs.getdbt.com/best-practices/how-we-structure/2-staging

Data mart — A smaller, purpose-built subset of the warehouse serving one team or function, e.g. a finance mart or a field mart.
→ https://en.wikipedia.org/wiki/Data_mart

Partitioning — Splitting a large table into chunks (usually by date or state) so queries only read the chunks they need.
→ https://en.wikipedia.org/wiki/Partition_(database)

Parquet — A compressed, columnar file format that is the de facto standard for storing analytical data in a data lake.
→ https://parquet.apache.org/

Apache Iceberg / Delta Lake — Open "table formats" that add warehouse features (transactions, schema changes, time travel) to files sitting in object storage. The technology that makes a lakehouse possible.
→ https://iceberg.apache.org/
→ https://delta.io/

Time travel — The ability to query a table as it existed at an earlier point in time. Directly relevant when you need to reproduce a targeting universe from three weeks ago.
→ https://delta.io/

---

Part 2 — Moving data around

ETL (Extract, Transform, Load) — Pull data from a source, reshape it, then load it into the warehouse. The classic pattern, dating to the 1970s.
→ https://en.wikipedia.org/wiki/Extract,_transform,_load

ELT (Extract, Load, Transform) — Pull data, load it raw into the warehouse first, then transform it there using the warehouse's own compute. The modern default, because cloud warehouse compute is cheap and keeping raw data is valuable.
→ https://airbyte.com/blog/reverse-etl

Reverse ETL — Pushing modeled data back out of the warehouse into the operational tools people actually use — the CRM, the ad platform, the texting tool. This is how a support score computed in the warehouse ends up available to a canvasser.
→ https://hightouch.com/blog/reverse-etl
→ https://www.fivetran.com/blog/what-is-reverse-etl

Pipeline — Any automated sequence that moves and transforms data from one place to another.
→ https://en.wikipedia.org/wiki/Data_pipeline

Connector — A pre-built piece of software that knows how to talk to a specific source system's API and pull its data.
→ https://www.parsonsproject.org/

CDC (Change Data Capture) — Reading a source database's change log so you only move what changed instead of re-copying everything.
→ https://en.wikipedia.org/wiki/Change_data_capture

Batch vs. streaming — Batch moves data on a schedule (hourly, nightly). Streaming moves it continuously as events occur. Most campaign data is batch; a few things (donations, event signups) benefit from near-real-time.
→ https://www.databricks.com/glossary/batch-vs-stream-processing

Ingestion — The act of getting data into the system in the first place.
→ https://ibm.com/think/topics/data-warehouse-vs-data-lake-vs-data-lakehouse

Orchestration — Scheduling and coordinating all the pipeline steps, handling dependencies, retries, and failures. Airflow, Dagster, and Prefect are orchestrators.
→ https://airflow.apache.org/

DAG (Directed Acyclic Graph) — The dependency graph of a pipeline: step B runs after step A, nothing loops back. Orchestrators express workflows as DAGs.
→ https://en.wikipedia.org/wiki/Directed_acyclic_graph

Idempotency — A job is idempotent if running it twice produces the same result as running it once. Essential for pipelines, because jobs fail and get retried.
→ https://en.wikipedia.org/wiki/Idempotence

Backfill — Re-running a pipeline over historical time periods, usually after fixing a bug or adding a new field.
→ https://docs.getdbt.com/

API (Application Programming Interface) — The defined way one piece of software requests data from another. Most SaaS tools expose a REST API.
→ https://en.wikipedia.org/wiki/API

Rate limit — A cap on how many API requests you may make per unit of time. A major practical constraint when pulling millions of records from a campaign CRM.
→ https://en.wikipedia.org/wiki/Rate_limiting

SFTP (SSH File Transfer Protocol) — Secure file transfer. Still the standard way voter-file vendors deliver bulk data.
→ https://en.wikipedia.org/wiki/SSH_File_Transfer_Protocol

Webhook — A source system pushing a notification to you when something happens, rather than you polling for it.
→ https://en.wikipedia.org/wiki/Webhook

Data lineage — A record of where each piece of data came from and what transformations it went through. Necessary for debugging and, in this context, for proving license compliance.
→ https://en.wikipedia.org/wiki/Data_lineage

Data governance — The policies and roles that define who may access what data, who is accountable for its quality, and how it may be used.
→ https://en.wikipedia.org/wiki/Data_governance

Data quality testing — Automated checks that data is complete, unique, in-range, and consistent — run as part of the pipeline rather than discovered later by a confused analyst.
→ https://greatexpectations.io/

Data observability — Monitoring for pipelines and datasets: alerting when a table stops updating, row counts drop, or a distribution shifts.
→ https://montecarlo.ai/

---

Part 3 — Platforms and tools

Warehouses and databases

• Snowflake — Managed cloud data warehouse; separates storage from compute, so you can scale query power independently. → https://www.snowflake.com/
• BigQuery — Google Cloud's serverless data warehouse. → https://cloud.google.com/bigquery
• Amazon Redshift — AWS's data warehouse; historically the dominant choice in progressive political data because of Civis Analytics' platform. → https://aws.amazon.com/redshift/
• Databricks — Lakehouse platform built around Apache Spark, strong for machine learning workloads. → https://www.databricks.com/
• DuckDB — A fast analytical database that runs in a single process on one machine. Useful for local analysis of large files. → https://duckdb.org/
• PostgreSQL — General-purpose open-source relational database. Appropriate for application data, not for the analytical workload. → https://www.postgresql.org/
• Amazon S3 / Google Cloud Storage — Object storage services; the physical layer under most data lakes. → https://aws.amazon.com/s3/

Transformation and orchestration

• dbt (data build tool) — The standard tool for transforming data inside the warehouse using SQL, with version control, testing, and documentation built in. → https://www.getdbt.com/
• Apache Airflow — Widely used open-source orchestrator. → https://airflow.apache.org/
• Dagster — Orchestrator organized around data assets rather than tasks. → https://dagster.io/
• Prefect — Python-native orchestrator. → https://www.prefect.io/

Ingestion and activation

• Fivetran — Managed ingestion; buys you maintained connectors instead of building them. → https://www.fivetran.com/
• Airbyte — Open-source ingestion platform with a large connector catalog. → https://airbyte.com/
• Meltano — Open-source ELT built on the Singer connector spec. → https://meltano.com/
• Hightouch — Reverse ETL platform. → https://hightouch.com/
• Census — Reverse ETL platform. → https://www.getcensus.com/
• Segment — Customer data platform; collects event data and routes it to destinations. → https://segment.com/

Business intelligence and reporting

• Looker — BI and semantic-modeling layer, now part of Google Cloud. → https://cloud.google.com/looker
• Tableau — Visualization-first BI tool. → https://www.tableau.com/
• Metabase — Lightweight open-source BI, good for non-technical self-service. → https://www.metabase.com/
• Apache Superset — Open-source BI and dashboarding. → https://superset.apache.org/
• Sisense — Commercial BI platform used in parts of the political data ecosystem. → https://www.sisense.com/

Political-sector-specific platforms

• Parsons — Open-source Python package from The Movement Cooperative providing connectors for the progressive tech ecosystem: VAN, ActBlue, Mobilize, Action Network, TargetSmart, Catalist, Redshift, S3 and dozens more. The single most relevant piece of prior art for the ingestion layer. → https://www.parsonsproject.org/
• The Movement Cooperative (TMC) — Member-led organization providing shared data infrastructure, tools, and pooled data licensing to progressive organizations. → https://www.movementcooperative.org/
• Civis Analytics — Data science company founded out of the 2012 Obama campaign's analytics team; sells a managed analytics platform built on Redshift. → https://www.civisanalytics.com/
• Defending Digital Campaigns (DDC) — Nonprofit that provides discounted and donated cybersecurity products and training to federal campaigns. Directly relevant to the security plan. → https://www.defendingdigitalcampaigns.org/

---

Part 4 — Analytics and data science

BI (Business Intelligence) — The practice and tooling of turning warehouse data into dashboards and reports that non-technical people can read and act on. In campaign terms: the thing a state director opens to see how their turf is doing.
→ https://en.wikipedia.org/wiki/Business_intelligence

KPI (Key Performance Indicator) — An agreed-upon metric used to track progress, e.g. doors knocked per shift, cost per acquired donor.
→ https://en.wikipedia.org/wiki/Performance_indicator

Dashboard — A screen of pre-built charts and numbers, refreshed automatically.
→ https://en.wikipedia.org/wiki/Dashboard_(computing)

Segmentation — Dividing a population into groups that behave differently and should be treated differently.
→ https://en.wikipedia.org/wiki/Market_segmentation

Cohort analysis — Comparing groups defined by when they joined or were contacted, to see how behavior differs over time.
→ https://en.wikipedia.org/wiki/Cohort_analysis

Model — A statistical or machine-learning function that estimates something unknown from things you do know. In campaigns, typically the probability a person supports the candidate, or will turn out.
→ https://en.wikipedia.org/wiki/Statistical_model

Score — The output of a model for one individual, usually 0–100 or 0–1. Campaign field and media programs are driven by scores.
→ https://www.l2-data.com/how-campaigns-use-voter-file-data/

Propensity model — A model estimating the likelihood a person does something — turns out, donates, volunteers.
→ https://en.wikipedia.org/wiki/Propensity_score_matching

Supervised learning — Training a model on examples where you know the right answer (e.g. survey responses), so it can predict for people you haven't asked.
→ https://en.wikipedia.org/wiki/Supervised_learning

Feature — An input variable to a model. Age, party registration, and past vote history are features.
→ https://en.wikipedia.org/wiki/Feature_(machine_learning)

Training / validation / holdout — Splitting your labeled data so you can fit the model on one part and honestly measure its accuracy on another.
→ https://en.wikipedia.org/wiki/Training,_validation,_and_test_data_sets

Model drift — A model getting less accurate over time because the world changed. Very fast in a campaign, where the electorate's attitudes move weekly.
→ https://en.wikipedia.org/wiki/Concept_drift

MRP (Multilevel Regression with Poststratification) — A statistical technique for estimating opinion in small geographies (states, districts) from a national survey. Widely used in modern political polling and modeling.
→ https://en.wikipedia.org/wiki/Multilevel_regression_with_poststratification

Weighting — Adjusting survey results so the sample matches the known composition of the population.
→ https://en.wikipedia.org/wiki/Survey_methodology

RCT (Randomized Controlled Trial) — Randomly assigning people to receive a treatment or not, then comparing outcomes. The gold standard for measuring whether a campaign tactic actually works.
→ https://en.wikipedia.org/wiki/Randomized_controlled_trial

A/B test — The applied version of the same idea: two variants, randomly assigned, measured against each other.
→ https://en.wikipedia.org/wiki/A/B_testing

Holdout group — People deliberately excluded from a program so they can serve as the control. Requires the warehouse to record assignment, not just contact.
→ https://en.wikipedia.org/wiki/Randomized_controlled_trial

Lift — The measured increase in an outcome caused by a treatment, relative to control.
→ https://en.wikipedia.org/wiki/Lift_(data_mining)

Uplift modeling — Modeling not who is likely to vote, but who is likely to vote because you contacted them. A meaningfully different and more useful target.
→ https://en.wikipedia.org/wiki/Uplift_modelling

Statistical significance — Whether a measured difference is large enough to be unlikely to have arisen by chance.
→ https://en.wikipedia.org/wiki/Statistical_significance

ML (Machine Learning) — Algorithms that learn patterns from data rather than being explicitly programmed with rules.
→ https://en.wikipedia.org/wiki/Machine_learning

---

Part 5 — Identity, matching, and enrichment

Identity resolution / entity resolution — Determining that records in different systems refer to the same real person, and linking them under a shared ID. Also called record linkage or record matching. This is the central technical difficulty in a campaign warehouse.
→ https://senzing.com/what-is-entity-resolution/
→ https://en.wikipedia.org/wiki/Record_linkage

Golden record — The single reconciled version of a person assembled from all sources, with conflicts resolved and a shared ID attached.
→ https://www.tamr.com/what-are-golden-records

MDM (Master Data Management) — The broader discipline of governing core entity data — people, organizations — across all systems. Entity resolution is the engine inside it.
→ https://en.wikipedia.org/wiki/Master_data_management

Deterministic matching — Matching on exact agreement of reliable identifiers (email, phone, state voter ID). High confidence, lower coverage.
→ https://senzing.com/what-is-entity-resolution/

Probabilistic matching — Matching on similarity across several weaker fields (name, address, birth year) with a confidence score. Higher coverage, risk of false merges.
→ https://en.wikipedia.org/wiki/Record_linkage

Deduplication — Collapsing multiple records that represent the same person into one.
→ https://senzing.com/what-is-entity-resolution/

Householding — Grouping individuals who share an address into a household unit. Relevant for mail and door canvassing.
→ https://www.l2-data.com/how-campaigns-use-voter-file-data/

Match rate — The percentage of a list you were able to successfully match against a reference file. A key vendor quality metric.
→ https://www.targetsmart.com/

Data append / data enhancement — Buying additional attributes about people you already have — phone numbers, consumer data, modeled scores — from a vendor.
→ https://www.targetsmart.com/

Matchback — Sending a list out to a vendor or platform and getting back which records they were able to identify, often for measuring ad reach.
→ https://liveramp.com/

Hashing — Converting an identifier like an email address into an irreversible fingerprint so it can be matched without being transmitted in the clear. Standard practice for sending audiences to ad platforms.
→ https://en.wikipedia.org/wiki/Cryptographic_hash_function

LiveRamp — Identity-resolution and data-connectivity company widely used to move audiences between data providers and ad platforms.
→ https://liveramp.com/

CDP (Customer Data Platform) — A packaged system that unifies person-level data from many sources and pushes audiences to marketing tools. The commercial analogue of what a campaign warehouse plus reverse ETL does.
→ https://en.wikipedia.org/wiki/Customer_data_platform

---

Part 6 — Political data terms

Voter file — The core dataset: a list of registered voters assembled from state and county election office records, typically enhanced with purchased consumer data, phone numbers, and modeled scores. It is licensed, not owned.
→ https://en.wikipedia.org/wiki/Voter_database
→ https://www.overtureglobal.io/story/the-secret-power-of-political-data-trusts

Vote history — Whether a person voted in past elections. Public record in most states. Note: whether, never how.
→ https://www.nass.org/sites/default/files/surveys/2020-10/NASS-briefing-FAQ-info-security-2020_0.pdf

Universe — The specific set of voters a program will target. "Cutting a universe" means writing the query that defines it.
→ https://www.l2-data.com/how-campaigns-use-voter-file-data/

Turf — A geographic chunk of a universe assigned to a canvasser or organizer.
→ https://en.wikipedia.org/wiki/Canvassing

Canvass — Direct voter contact, usually door-to-door. Produces data: who was home, what they said.
→ https://en.wikipedia.org/wiki/Canvassing

Walk list — The printed or in-app list of doors a canvasser works through.
→ https://www.ngpvan.com/

ID (voter ID / support ID) — A recorded answer from a voter about who they support, typically on a 1–5 scale. Not to be confused with voter identification laws.
→ https://www.l2-data.com/how-campaigns-use-voter-file-data/

Activist code / survey question — Structures inside VAN for tagging people with attributes or recording answers to canvass questions. These are the field data model in practice.
→ https://www.ngpvan.com/

Persuasion vs. mobilization — Two distinct programs: changing minds versus turning out people who already agree. They require different universes, different models, and different metrics.
→ https://en.wikipedia.org/wiki/Get_out_the_vote

GOTV (Get Out The Vote) — The final-stretch turnout push. The period when data latency requirements are most severe.
→ https://en.wikipedia.org/wiki/Get_out_the_vote

Early vote / absentee file — Daily files from election officials showing who has already voted. Used to suppress already-voted people from contact lists and to reallocate resources in real time.
→ https://www.stackadapt.com/resources/blog/transforming-political-campaigns-with-voter-file-data

VBM (Vote By Mail) — Mail balloting; generates ballot-request and ballot-return data streams that campaigns track closely.
→ https://en.wikipedia.org/wiki/Postal_voting

Precinct — The smallest election administration geography; the unit official results are reported in.
→ https://en.wikipedia.org/wiki/Electoral_district

DMA (Designated Market Area) — Nielsen's television market geography. Broadcast ad buying happens at DMA level, which rarely aligns with political boundaries — a recurring targeting-efficiency problem.
→ https://en.wikipedia.org/wiki/Media_market

P2P texting (peer-to-peer) — Texting where a human presses send on each message, used to stay outside autodialer restrictions. Its legal status has been litigated.
→ https://www.dnc.com/blog/tcpa-risks-political-campaigns

Systems and organizations

• NGP VAN / VoteBuilder — The dominant Democratic campaign CRM and voter contact system. "VAN" is the voter contact side; VoteBuilder is the party-branded deployment. → https://www.ngpvan.com/
• Bonterra / EveryAction — Organizing and fundraising CRM, now part of Bonterra, closely related to NGP VAN. → https://www.bonterratech.com/
• Action Network — Advocacy and organizing platform for petitions, events, and email. → https://actionnetwork.org/
• ActionKit — Advocacy and fundraising platform used by large progressive organizations. → https://www.actionkit.com/
• NationBuilder — All-in-one campaign website, CRM, and organizing platform, used across the spectrum. → https://nationbuilder.com/
• Mobilize — Volunteer and event signup platform. → https://www.mobilize.us/
• ActBlue — Democratic online fundraising processor. → https://www.actblue.com/
• WinRed — Republican online fundraising processor. → https://www.winred.com/
• Catalist — Progressive voter data warehouse and file manager, founded 2006. → https://catalist.us/
• TargetSmart — Democratic-aligned voter data and analytics firm; maintains the VoterBase national file. → https://www.targetsmart.com/
• L2 — Long-established nonpartisan voter file vendor with a national adult database; distributes via the Snowflake Marketplace. → https://www.l2-data.com/
• PDI (Political Data Inc.) — Voter data provider with deep California coverage and its own outreach tooling. → https://politicaldata.com/
• Aristotle — One of the oldest political data vendors, founded 1983. → https://aristotle.com/
• i360 — Conservative-aligned data and technology platform. → https://i-360.com/
• Data Trust — For-profit Republican voter data infrastructure company founded in 2011, with a data-sharing relationship with the RNC. → https://thedatatrust.com/
• DDx (Democratic Data Exchange) — For-profit company founded 2019 operating a real-time blind exchange of voter contact data among Democratic-aligned organizations. → https://demexchange.com/
• Twilio — Communications API provider underlying much campaign SMS and voice. → https://www.twilio.com/
• Scale to Win — Political texting and organizing tools vendor. → https://scaletowin.com/
• Hustle — Peer-to-peer texting platform. → https://www.hustle.com/

---

Part 7 — Legal and regulatory

FEC (Federal Election Commission) — The federal agency that administers and enforces campaign finance law, including disclosure and coordination rules.
→ https://www.fec.gov/

FECA (Federal Election Campaign Act) — The underlying federal campaign finance statute.
→ https://www.fec.gov/

Principal campaign committee — The candidate's official committee, the entity that raises and spends money on the candidate's behalf and files disclosure reports.
→ https://www.fec.gov/help-candidates-and-committees/

PAC (Political Action Committee) — A committee that raises and spends money to influence elections, subject to contribution limits.
→ https://en.wikipedia.org/wiki/Political_action_committee

Super PAC — An independent-expenditure-only committee that may raise unlimited sums but may not contribute to or coordinate with candidates.
→ https://en.wikipedia.org/wiki/Political_action_committee
→ https://www.cov.com/en/news-and-insights/insights/2025/12/forming-and-operating-super-pacs-a-practical-guide-for-political-consultants-in-2026

501(c)(3) / 501(c)(4) — Tax-exempt nonprofit categories. (c)(3) organizations are barred from partisan campaign activity; (c)(4) social welfare organizations may engage in some. The distinction creates hard data-sharing boundaries.
→ https://en.wikipedia.org/wiki/501(c)_organization

Coordinated communication — Under FEC rules, a communication paid for by someone other than the campaign that meets content and conduct tests, and is therefore treated as an in-kind contribution. The rule that makes data sharing between a campaign and outside groups legally dangerous.
→ https://www.fec.gov/help-candidates-and-committees/candidate-taking-receipts/coordinated-communications/

Firewall (campaign finance sense) — A documented internal barrier preventing information flow between staff serving a campaign and staff serving an outside spender. It must exist in a written policy distributed to affected personnel. Note this is a legal term, distinct from a network firewall.
→ https://www.fec.gov/help-candidates-and-committees/candidate-taking-receipts/coordinated-communications/
→ https://www.wiley.law/newsletter-773

In-kind contribution — A non-cash contribution of goods or services. Data sharing across a firewall can be treated as one.
→ https://www.fec.gov/help-candidates-and-committees/

Independent expenditure — Spending advocating for or against a candidate, made without coordination with that candidate.
→ https://en.wikipedia.org/wiki/Independent_expenditure

JFC (Joint Fundraising Committee) — A committee formed by two or more committees (typically a campaign and party committees) to raise money together and split proceeds. Creates additional data-boundary questions.
→ https://www.fec.gov/help-candidates-and-committees/

TCPA (Telephone Consumer Protection Act) — The 1991 federal law restricting autodialed calls and texts and prerecorded messages. Political campaigns are not exempt. Statutory damages are $500 per call or text, trebled to $1,500 for willful violations.
→ https://en.wikipedia.org/wiki/Telephone_Consumer_Protection_Act_of_1991
→ https://mslawgroup.com/tcpa-compliance-for-political-calls/

ATDS (Automatic Telephone Dialing System) — "Autodialer." Using one to contact a mobile number generally requires prior express consent under the TCPA.
→ https://mslawgroup.com/navigating-tcpa-regulations-for-political-calls/

Prior express consent — The permission a person must give before you may autodial or robotext them. The burden of proving it rests on the caller, which makes it a data-retention requirement.
→ https://www.wiley.law/newsletter-FCC-Issues-Enforcement-Advisory-Reminding-Political-Campaigns-Calling-Texting-Restrictions-TCPA

DNC Registry (National Do Not Call Registry) — The federal do-not-call list. Political calls are largely exempt from it, but scrubbing against it is treated as best practice.
→ https://www.donotcall.gov/

Quiet hours — Time-of-day restrictions on calling and texting, with additional state-level variation. A nationwide program must handle time zones correctly.
→ https://activeprospect.com/blog/tcpa-text-messages/

CAN-SPAM Act — The federal law governing commercial email, requiring accurate headers and a working unsubscribe mechanism.
→ https://en.wikipedia.org/wiki/CAN-SPAM_Act_of_2003

NVRA (National Voter Registration Act, "Motor Voter") — 1993 federal law governing voter registration procedures and list maintenance.
→ https://en.wikipedia.org/wiki/National_Voter_Registration_Act_of_1993

HAVA (Help America Vote Act) — 2002 federal law that, among other things, required states to build centralized statewide voter registration databases — the origin of the modern voter file.
→ https://en.wikipedia.org/wiki/Help_America_Vote_Act

Voter file use restrictions — State-specific rules limiting how purchased voter data may be used, stored, transmitted, and re-disclosed. They vary substantially and are enforced through the license agreement as well as by statute.
→ https://suttonsmart.com/political-consulting/legacy-data-archiving-building-a-database-for-a-career-not-just-a-cycle/
→ https://www.brennancenter.org/our-work/research-reports/requests-access-election-data-and-equipment-require-balancing-risks-and

PII (Personally Identifiable Information) — Data that identifies a specific person. In voter files this can include partial or full SSNs, driver's license numbers, and dates of birth, depending on the state.
→ https://en.wikipedia.org/wiki/Personal_data
→ https://www.nass.org/sites/default/files/surveys/2020-10/NASS-briefing-FAQ-info-security-2020_0.pdf

CCPA / CPRA — California's consumer privacy statutes. Political organizations are not uniformly exempt, and California registers political data brokers.
→ https://en.wikipedia.org/wiki/California_Consumer_Privacy_Act

GDPR — The EU's data protection regulation. Generally out of scope for a US campaign, but relevant if any vendor, staff, or infrastructure sits in the EU.
→ https://en.wikipedia.org/wiki/General_Data_Protection_Regulation

DPA (Data Processing Agreement) — The contract governing how a vendor may handle data you give them. Every vendor in the pipeline needs one.
→ https://en.wikipedia.org/wiki/General_Data_Protection_Regulation

Records retention — Legal obligations to keep records for a defined period after the campaign ends, independent of the campaign's operational needs.
→ https://www.fec.gov/help-candidates-and-committees/

---

Part 8 — Security

MFA (Multi-Factor Authentication) — Requiring more than a password. "Phishing-resistant MFA" means hardware keys or passkeys rather than SMS codes, which can be intercepted or socially engineered.
→ https://en.wikipedia.org/wiki/Multi-factor_authentication
→ https://www.cisa.gov/

SSO (Single Sign-On) — One central login that grants access to many systems, so accounts can be created and revoked in one place.
→ https://en.wikipedia.org/wiki/Single_sign-on

IdP (Identity Provider) — The central system that holds user identities and authenticates them — Okta, Google Workspace, Entra ID.
→ https://en.wikipedia.org/wiki/Identity_provider

SAML / OIDC (OpenID Connect) — The two standard protocols by which an application delegates login to an identity provider.
→ https://en.wikipedia.org/wiki/Security_Assertion_Markup_Language
→ https://openid.net/

RBAC (Role-Based Access Control) — Granting permissions to roles rather than individuals, then assigning people to roles. The only workable model when headcount grows tenfold in a few months.
→ https://en.wikipedia.org/wiki/Role-based_access_control

RLS (Row-Level Security) — Database rules that restrict which rows a user can see — e.g. a Michigan organizer sees only Michigan records — while everyone queries the same table.
→ https://docs.snowflake.com/en/user-guide/security-row-intro

Column masking — Hiding or obscuring sensitive columns from users who don't need them, while leaving the rest of the row visible.
→ https://docs.snowflake.com/en/user-guide/security-column-intro

Least privilege — The principle that each account gets the minimum access needed to do its job, and nothing more.
→ https://en.wikipedia.org/wiki/Principle_of_least_privilege

Encryption at rest / in transit — Protecting data while stored on disk and while moving across a network. Table stakes, and not by itself a security program.
→ https://en.wikipedia.org/wiki/Data_at_rest

KMS (Key Management Service) — The system that stores and controls the encryption keys.
→ https://en.wikipedia.org/wiki/Key_management

Secrets management — Securely storing and rotating API credentials and passwords rather than leaving them in code or config files. A campaign warehouse will hold dozens of vendor credentials.
→ https://en.wikipedia.org/wiki/Secret_sharing

Tokenization — Replacing a sensitive value with a meaningless substitute, so the real value never enters your system.
→ https://en.wikipedia.org/wiki/Tokenization_(data_security)

Audit log — An immutable record of who accessed or changed what, and when. In this context it serves both security and legal-firewall evidence purposes.
→ https://en.wikipedia.org/wiki/Audit_trail

Zero trust — A security model that assumes no network location is inherently trusted and verifies every request.
→ https://en.wikipedia.org/wiki/Zero_trust_security_model

Phishing / spear phishing — Fraudulent messages designed to steal credentials. Spear phishing is targeted at specific individuals. This is the primary documented attack vector against presidential campaigns.
→ https://en.wikipedia.org/wiki/Phishing
→ https://www.darkreading.com/remote-workforce/2024-us-political-campaigns-prepared-cyber-threats

APT (Advanced Persistent Threat) — A well-resourced, usually state-sponsored attacker that maintains long-term access to a target. Groups linked to China, Iran, and Russia have been publicly reported targeting US presidential campaigns.
→ https://en.wikipedia.org/wiki/Advanced_persistent_threat
→ https://techcrunch.com/2020/06/04/google-china-iran-trump-biden/

Insider threat — Risk from people who already have legitimate access — including leaking to press or accessing another campaign's data.
→ https://www.cisa.gov/topics/physical-security/insider-threat-mitigation

Data exfiltration — The unauthorized removal of data from a system.
→ https://en.wikipedia.org/wiki/Data_breach

Cloud misconfiguration — Security failure caused by incorrect settings rather than an attack. The largest known voter data exposure — 1.1 TB covering roughly 198 million voters — was an unsecured cloud storage bucket.
→ https://www.upguard.com/breaches/the-rnc-files

SOC 2 — An audited report on a vendor's security controls. Commonly requested during vendor evaluation.
→ https://en.wikipedia.org/wiki/System_and_Organization_Controls

Penetration test — An authorized simulated attack to find weaknesses before a real attacker does.
→ https://en.wikipedia.org/wiki/Penetration_test

DR / RTO / RPO (Disaster Recovery / Recovery Time Objective / Recovery Point Objective) — The plan for restoring service after failure, how fast you must restore it, and how much data loss is tolerable. On Election Day both targets approach zero.
→ https://en.wikipedia.org/wiki/Disaster_recovery

CISA (Cybersecurity and Infrastructure Security Agency) — The federal agency responsible for critical infrastructure cybersecurity, including election infrastructure; publishes guidance relevant to campaigns.
→ https://www.cisa.gov/

NIST Cybersecurity Framework — A widely used federal framework for organizing a security program.
→ https://www.nist.gov/cyberframework

---

Quick-reference: acronyms

Acronym	Expansion
API	Application Programming Interface
APT	Advanced Persistent Threat
ATDS	Automatic Telephone Dialing System
BI	Business Intelligence
CDC	Change Data Capture
CDP	Customer Data Platform
CISA	Cybersecurity and Infrastructure Security Agency
DAG	Directed Acyclic Graph
DDC	Defending Digital Campaigns
DDx	Democratic Data Exchange
DMA	Designated Market Area
DPA	Data Processing Agreement
DR / RTO / RPO	Disaster Recovery / Recovery Time Objective / Recovery Point Objective
ELT	Extract, Load, Transform
ETL	Extract, Transform, Load
FEC / FECA	Federal Election Commission / Federal Election Campaign Act
GOTV	Get Out The Vote
HAVA	Help America Vote Act
IdP	Identity Provider
JFC	Joint Fundraising Committee
KMS	Key Management Service
KPI	Key Performance Indicator
MDM	Master Data Management
MFA	Multi-Factor Authentication
ML	Machine Learning
MPP	Massively Parallel Processing
MRP	Multilevel Regression with Poststratification
NVRA	National Voter Registration Act
OIDC	OpenID Connect
OLAP / OLTP	Online Analytical / Transaction Processing
P2P	Peer-to-Peer (texting)
PAC	Political Action Committee
PII	Personally Identifiable Information
RBAC	Role-Based Access Control
RCT	Randomized Controlled Trial
RLS	Row-Level Security
SAML	Security Assertion Markup Language
SCD	Slowly Changing Dimension
SFTP	SSH File Transfer Protocol
SSO	Single Sign-On
TCPA	Telephone Consumer Protection Act
TMC	The Movement Cooperative
VAN	Voter Activation Network
VBM	Vote By Mail