Status: RESEARCH / PRE-CAPTURE ONLY
Subject: Cambridge Analytica / Facebook — GSRApp data collection, voter profiling, platform governance, regulatory findings, political microtargeting, and election-impact claims
Research date: 2026-08-09
Scope: source inventory, authority/posture, contradiction/qualification map, provisional five-claim matrix, capture hazards, reverse-source candidates
Explicit non-claims: This dossier creates no CaptureReceipt, SourceEdition, SRS receipt, custody, admission, final standing, article prose, or canonical Counterpedia identity.

A. SUBJECT / DISAMBIGUATION

Common name

Facebook–Cambridge Analytica scandal

Exact scope

The collection through Aleksandr Kogan's Facebook application commonly known as thisisyourdigitallife / GSRApp of data from direct app users and, through Facebook's then-existing Graph API, data associated with many of those users' Facebook friends; transfer and use of data/work product by Cambridge Analytica/SCL for voter profiling and targeting; Facebook's platform/data-governance role; U.S. Federal Trade Commission findings and settlements; UK Information Commissioner's Office investigation and enforcement; UK parliamentary investigations; Cambridge Analytica's claimed and disputed psychographic capabilities; and the crucial distinction between:

1. data acquisition / deception / platform-policy conduct, and

2. whether Cambridge Analytica's work measurably changed an election outcome.

The initial governed demo scope is primarily:

• U.S. Facebook data / U.S. voter profiling,

• 2016 U.S. presidential-election claims,

• UK regulatory/parliamentary evidence where it illuminates data lineage, organizational claims, or evidentiary limits.

The dossier does not assume Cambridge Analytica caused Donald Trump's victory, caused Brexit, or possessed a uniquely effective "mind-control" technology.

Important identifiers / terms

• Cambridge Analytica / CA

• Strategic Communication Laboratories / SCL / SCL Elections

• Aleksandr Kogan

• Global Science Research / GSR

• GSRApp

• thisisyourdigitallife

• Alexander Nix

• Christopher Wylie

• AggregateIQ / AIQ

• Ripon

• OCEAN / Big Five personality traits

• psychographics / personality scores

• political microtargeting

• voter profiling

• Facebook Graph API v1

• Facebook Graph API v2

• Facebook User ID

• Facebook "friends" data

• Facebook Platform Policies

• FTC Matter 182 3107 / Docket 9383

• FTC 2012 Facebook privacy order

• FTC 2019 Facebook $5 billion settlement

• UK ICO

• UK Data Protection Act 1998

• DCMS Committee Disinformation and 'fake news'

• FEC MUR 7350 / 7351

• Trump campaign / Project Alamo

• Ted Cruz campaign

• Bolton PAC

• Leave.EU

• Vote Leave

Naming / semantic traps

1. "87 million users had their data stolen" is too compressed. Facebook's own 2018 estimate said information of up to 87 million people worldwide may have been improperly shared with Cambridge Analytica. FTC's later U.S.-specific complaint used different populations: roughly 250,000–270,000 direct U.S. app users, 50–65 million U.S. Facebook friends, and at least 30 million identifiable U.S. consumers. These are different denominators.

2. Direct app users ≠ app users' friends. Direct users installed/took the survey; friends generally did not interact with the app.

3. "No consent" is too broad unless the data field and person class are specified. Facebook's 2018 first-party position said direct app users knowingly granted the app access under then-existing platform mechanics, while friend data could flow through platform permissions without those friends interacting with the app. FTC later focused on deceptive representations and privacy-control practices.

4. Platform-policy violation ≠ computer intrusion. Facebook itself emphasized in March 2018 that this was not a hack/password breach in the conventional sense. That does not make the downstream use compliant with platform rules or privacy law.

5. Facebook's claim that access was "legitimate" under its then-current API channels is a first-party platform characterization, not a complete privacy-law conclusion.

6. Kogan's collection ≠ Cambridge Analytica's later use. Acquisition, transfer, model training, matching, targeting, and campaign deployment are separate graph stages.

7. Facebook data ≠ personality score. Personality scores are derived/modelled objects produced from data and algorithms.

8. Personality score ≠ voter file. FTC says scores were matched with U.S. voter records; those inputs should remain distinct.

9. Voter profile ≠ delivered political ad. A model can exist without proving a specific voter saw a specific message.

10. Delivered ad ≠ persuasion. Exposure does not prove attitude change.

11. Persuasion ≠ vote change. Even measured click/conversion effects in commercial experiments do not establish vote-choice effects.

12. Vote change ≠ election outcome change. A campaign technique can influence some individuals without being shown to have changed the Electoral College result.

13. Psychographic targeting can be effective in some experimental/commercial contexts without proving Cambridge Analytica's claimed election effect.

14. Cambridge Analytica's marketing claims are first-party promotional claims, not independent evidence of efficacy.

15. Whistleblower testimony is attributed testimony. It can establish what a witness says, identify documents and trigger investigation; it is not automatically an adjudicated finding.

16. Alexander Nix's testimony/letters are first-party defense/position.

17. FTC complaint allegations ≠ FTC final findings. Cambridge Analytica ultimately received a Commission Opinion and Final Order after it failed to contest the proceeding; Kogan and Nix separately settled FTC allegations.

18. Kogan/Nix consent orders ≠ admissions of every Cambridge Analytica allegation.

19. Cambridge Analytica bankruptcy/administration ≠ a merits finding about election influence.

20. Facebook's 2019 $5B FTC settlement covers broader privacy-order allegations, not a finding that every dollar of the penalty is attributable to Cambridge Analytica.

21. Facebook "up to 87 million" ≠ FTC's 50–65 million U.S. friends estimate. Global vs U.S., potentially shared vs alleged collected, and person-identifiability differ.

22. A House of Commons committee report is a parliamentary investigative finding, not a criminal or civil judgment.

23. The ICO's data-protection findings do not establish U.S. election causation.

24. The ICO's statement that it had no evidence CA/SCL performed data-analytics work for EU Referendum campaigns must not be rewritten as proof CA/SCL had no relationship, pitch, contact, or other non-analytics role.

25. AIQ ≠ Cambridge Analytica. Contractual/technical relationships and data exchanges require exact source edges.

26. "Cambridge Analytica worked for Trump" ≠ "the Facebook-derived psychographic model was used to determine the election."

27. "Psychographics was not used" claims are actor/source-specific and contested.

28. FEC matter closure/dismissal ≠ a factual finding that Cambridge Analytica had no effect on the election.

29. Privacy harm ≠ electoral-outcome harm. A data-protection violation can be established without proving a changed vote.

30. Democratic-risk concerns do not require proof that one historical election outcome was changed.

Temporal bounds

Data collection/model development: 2013–2015
2016 U.S. election activity: 2015–2016
Public scandal/investigations: 2018–2020
Regulatory/academic lessons: 2019–2026

Why fertile for Counterpedia

Cambridge Analytica is the corpus's strongest subject for:

• established conduct vs unestablished downstream effect;

• raw data vs derived profile vs targeted message vs behavioral outcome;

• global/user-count denominator discipline;

• platform policy vs privacy-law findings;

• allegation vs administrative finding vs consent order;

• whistleblower testimony vs company denial;

• capability vs deployment vs efficacy;

• election influence vs election outcome causation;

• same scandal, multiple non-equivalent notions of "affected."

The governed chain should be visible:

Facebook platform/API → direct app user → friend data → GSR/Kogan dataset → Cambridge Analytica transfer/use → derived personality score → voter-file match → targeting model → campaign audience selection → message/ad exposure → behavioral response → vote choice → election outcome.

The core scandal is strongly evidenced near the left side of that chain.

The causal evidence becomes much weaker toward the right side.

B. SOURCE INVENTORY

CA-S01 — FTC final Cambridge Analytica Opinion and Order

Exact case: In the Matter of Cambridge Analytica, LLC
FTC Matter/File Number: 182 3107
Docket: 9383
Case page: https://www.ftc.gov/legal-library/browse/cases-proceedings/182-3107-cambridge-analytica-llc-matter
Opinion date: December 6, 2019
Final case status: Closed

Authority / posture: Federal Trade Commission adjudicative administrative proceeding. Cambridge Analytica had filed for bankruptcy and did not answer the complaint or summary-decision motion; the Commission issued an Opinion finding Section 5 violations and a Final Order.

May support

• Commission finding that Cambridge Analytica engaged in deceptive practices in obtaining/handling Facebook-derived personal information;

• Commission finding concerning deceptive EU-U.S. Privacy Shield representations;

• final injunctive/delete-destroy requirements;

• stronger posture than the July 2019 complaint.

Critical qualification The procedure was not a contested trial on whether Cambridge Analytica altered an election result.

Captureability: HIGH
Priority: P0

CA-S02 — FTC 2019 complaint against Cambridge Analytica / Kogan / Nix

Exact source: FTC July 24, 2019 administrative complaint and announcement
Case pages:
https://www.ftc.gov/legal-library/browse/cases-proceedings/182-3107-cambridge-analytica-llc-matter
https://www.ftc.gov/legal-library/browse/cases-proceedings/182-3106-182-3107-aleksandr-kogan-alexander-nix-matter

Authority / posture: complaint allegations at filing stage.

Key alleged chain

• Kogan's GSRApp gathered survey and Facebook-profile information;

• the project used data from direct app users and their Facebook friends;

• Cambridge Analytica/Kogan/Nix used data to generate personality scores;

• personality scores were matched with U.S. voter records;

• resulting information was used for voter profiling and targeted advertising services.

U.S.-specific population figures alleged

• about 250,000–270,000 direct U.S. GSRApp users;

• 50–65 million of those users' Facebook friends;

• at least 30 million identifiable U.S. consumers.

Cannot establish alone

• final liability;

• campaign-specific deployment;

• election outcome change.

Captureability: HIGH
Priority: P0

CA-S03 — FTC final Kogan / Nix consent orders

Exact case: In the Matter of Aleksandr Kogan and Alexander Nix
FTC Matter: 182 3106 / 182 3107
Case page:
https://www.ftc.gov/legal-library/browse/cases-proceedings/182-3106-182-3107-aleksandr-kogan-alexander-nix-matter
Final orders: December 18, 2019

Authority / posture: Part 2 consent settlements resolving FTC allegations against Kogan and Nix.

May support

• settlement terms;

• restrictions on future representations;

• data/work-product deletion/destruction requirements.

Cannot support

• "Kogan and Nix were found after trial to have changed the 2016 election";

• automatic admission of every allegation unless order text expressly says so.

Captureability: HIGH
Priority: P1

CA-S04 — FTC Facebook $5 billion privacy settlement

Exact title: FTC Imposes $5 Billion Penalty and Sweeping New Privacy Restrictions on Facebook
Issuer: FTC / DOJ settlement
Date: July 24, 2019
Locator:
https://www.ftc.gov/news-events/news/press-releases/2019/07/ftc-imposes-5-billion-penalty-sweeping-new-privacy-restrictions-facebook

Authority / posture: federal settlement resolving alleged violations of Facebook's 2012 FTC privacy order plus related deceptive-practice allegations.

Relevant findings/allegations

• Facebook's settings/disclosures allowed friend data to flow to third-party apps in ways many users did not understand;

• Facebook announced in April 2014 it would stop friend-data collection for third-party developers but grandfathered existing apps for another year;

• FTC alleged inadequate policing of platform developers;

• new order imposed extensive governance/privacy controls.

Critical qualification The $5B penalty is not "Cambridge Analytica's fine" and should not be represented as a monetary measure of Cambridge Analytica's election impact.

Captureability: HIGH
Priority: P0

CA-S05 — Facebook March 16, 2018 suspension/timeline statement

Exact title: Suspending Cambridge Analytica and SCL Group From Facebook
Issuer: Facebook / Meta
Date: March 16, 2018, with March 17 update
Locator:
https://about.fb.com/news/2018/03/suspending-cambridge-analytica/

Authority / posture: first-party platform statement during unfolding scandal.

Key first-party claims

• about 270,000 people downloaded Kogan's app;

• Kogan accessed certain user and permitted friend information through platform channels then available;

• Facebook says transfer to SCL/Cambridge Analytica violated its Platform Policies;

• Facebook says parties later certified destruction in 2015;

• Facebook distinguished the incident from a conventional hack/password breach.

Critical qualification This is Facebook's contemporaneous defense/explanation, not a regulator's final privacy-law adjudication.

Captureability: HIGH
Priority: P0

CA-S06 — Facebook April 2018 "up to 87 million" estimate

Exact title: An Update on Our Plans to Restrict Data Access on Facebook
Issuer: Facebook
Updated: April 2018
Locator:
https://about.fb.com/news/2018/04/restricting-data-access/

Authority / posture: first-party platform estimate and product-policy response.

Key estimate Facebook said information of up to 87 million people, mostly in the U.S., may have been improperly shared with Cambridge Analytica.

Counterpedia value This is a different population construct from the FTC's later U.S.-specific direct-user/friend/identifiable-consumer estimates.

Captureability: HIGH
Priority: P0

CA-S07 — Facebook March 19, 2018 forensic-audit statement

Exact title: Pursuing Forensic Audits to Investigate Cambridge Analytica Claims
Issuer: Facebook
Date: March 19, 2018
Locator:
https://about.fb.com/news/2018/03/forensic-audits-cambridge-analytica/

Authority / posture: contemporaneous first-party investigation announcement.

May support

• Facebook says Cambridge Analytica/SCL/Kogan/Wylie had certified data destruction;

• Facebook's proposed private forensic audit was paused when ICO pursued its own warrant.

Cannot establish

• that the data was in fact fully destroyed;

• election influence.

Captureability: HIGH
Priority: P2

CA-S08 — Facebook 2019 knowledge-timeline statement

Exact title: Document Holds the Potential for Confusion
Issuer: Facebook
Date: August 2019
Locator:
https://about.fb.com/news/2019/08/document-holds-the-potential-for-confusion/

Authority / posture: later first-party corporate statement defending Facebook's knowledge timeline.

Key position Facebook states it was not aware Kogan had transferred/sold data to Cambridge Analytica until December 2015, while acknowledging an earlier September 2015 internal report concerned suspected scraping by Cambridge Analytica.

Conflicts/qualifies UK parliamentary material records ICO correspondence that three Facebook senior managers were in email exchanges earlier in 2015 concerning the GSR issue, creating a knowledge/definition dispute over what was known and what kind of conduct was known.

Captureability: HIGH
Priority: P1

CA-S09 — ICO 2018 final report to Parliament

Exact title: Investigation into the use of data analytics in political campaigns — A report to Parliament
Issuer: UK Information Commissioner's Office
Date: November 6, 2018
Historical official PDF locator:
https://ico.org.uk/media/action-weve-taken/2260271/investigation-into-the-use-of-data-analytics-in-political-campaigns-final-20181105.pdf
Preserved OGL copy/source record:
https://commons.wikimedia.org/wiki/File:Investigation_into_the_use_of_data_analytics_in_political_campaigns.pdf

Format: PDF, 114 pages in preserved copy
Authority / posture: UK data-protection regulator's large political-data investigation.

May support

• ICO conclusions about data-protection practices;

• enforcement actions and referrals;

• Cambridge Analytica/SCL findings;

• AIQ/SCL/CA relationship evidence;

• importantly, ICO's statement that at that stage it had no evidence CA/SCL were involved in data-analytics work with EU Referendum campaigns.

Captureability: MEDIUM Hazard: historical ICO direct PDF currently appears migrated/404 from some clients. CAP1 should attempt official migrated path and preserve exact archived/OGL copy separately, never silently call the mirror origin.

Priority: P0

CA-S10 — ICO Facebook £500,000 monetary penalty

Issuer: UK Information Commissioner's Office
Penalty date: October 25, 2018
Amount: £500,000, maximum under the old Data Protection Act 1998 regime
Current official historical confirmation:
https://ico.org.uk/about-the-ico/our-information/history-of-the-ico/our-history/

Preserved monetary-penalty notice copy:
public legal/regulatory archives preserve the original notice; CAP1 should locate an ICO archival/original copy before governed admission.

Authority / posture: UK regulator monetary penalty concerning Facebook data-protection failures.

May support

• existence/amount of fine;

• regulator's Facebook privacy/data-governance findings once exact penalty notice is captured.

Cannot support

• a quantified electoral effect.

Captureability: PRIMARY NOTICE TO RESOLVE
Priority: P1

CA-S11 — House of Commons DCMS Final Report

Exact title: Disinformation and 'fake news': Final Report
Issuer: House of Commons Digital, Culture, Media and Sport Committee
Report: HC 1791
Ordered: February 14, 2019
Published: February 18, 2019
Exact PDF:
https://publications.parliament.uk/pa/cm201719/cmselect/cmcumeds/1791/1791.pdf
HTML:
https://publications.parliament.uk/pa/cm201719/cmselect/cmcumeds/1791/179102.htm

Authority / posture: parliamentary committee investigative report, not judicial/regulatory merits judgment.

May support

• committee findings concerning Facebook governance;

• testimony/document conflicts;

• relationship evidence among SCL/CA/AIQ;

• committee treatment of ICO findings;

• chronology of data targeting and political-campaigning concerns.

Important exact qualification The report quotes ICO as finding no evidence CA/SCL performed data-analytics work with EU Referendum campaigns at that stage, while separately describing close relationships/data exchange involving SCL/CA/AIQ.

Captureability: HIGH
Priority: P0

CA-S12 — Christopher Wylie parliamentary evidence

Source family: oral and written evidence before UK Parliament DCMS Committee
Witness: Christopher Wylie
Committee inquiry: Disinformation / fake news
Example source:
https://committees.parliament.uk/committee/378/digital-culture-media-and-sport-committee/news/103633/fake-news-christopher-wylie-evidence-17-19/
and associated oral/written evidence records.

Authority / posture: whistleblower witness testimony and submitted materials.

May support

• Wylie's account of CA/SCL/GSR development;

• alleged Facebook-derived modelling;

• claimed campaign applications;

• identification of documents for further verification.

Cannot establish automatically

• every downstream deployment/election-effect assertion.

Captureability: HIGH
Priority: P1

CA-S13 — Alexander Nix parliamentary evidence

Exact source type: written/oral evidence to UK Parliament
Example written evidence:
https://committees.parliament.uk/writtenevidence/87428/html/

Authority / posture: first-party Cambridge Analytica executive defense/position.

Key stated position in February 2018 Nix said Cambridge Analytica:

• did not use personality modelling/psychographics in the 2016 U.S. presidential election;

• had no access to Facebook likes;

• did not gather Facebook data;

• did not work on the 2016 EU Referendum.

Critical qualification These claims were contested by other witnesses/documents and should be preserved as Nix's statement, not promoted to neutral fact.

Captureability: HIGH
Priority: P1

CA-S14 — Chris Vickery oral evidence on AIQ/Ripon

Exact source: House of Commons oral evidence, May 2, 2018
Locator:
https://committees.parliament.uk/oralevidence/7921/html/

Authority / posture: expert/witness interpretation of exposed AIQ repository/code.

May support

• observed code/repository capabilities;

• presence of OCEAN/psychographic fields/functions in Ripon-related tooling;

• Vickery's belief that derivative modelling could have been used.

Critical limitation Vickery explicitly said he could not point to direct evidence proving Ripon was used on the Trump campaign.

Counterpedia value A capability can exist without proving historical deployment.

Captureability: HIGH
Priority: P0

CA-S15 — FTC/Facebook platform API policy transition

Source family: FTC Facebook settlement + Facebook platform statements
Key period: April 2014–April 2015

Authority / posture: regulator allegation/settlement record plus platform first-party history.

Key fact structure

• Facebook publicly announced Graph API changes restricting friend-data access;

• existing apps were grandfathered for a period;

• Kogan's pre-existing app could continue obtaining friend data under that legacy access;

• FTC later used these platform practices as part of the Facebook privacy-order case.

Counterpedia value The same collection can be:

• technically allowed by an API permission path,

• contrary to a downstream platform-transfer policy,

• and subject to separate privacy-law scrutiny.

Priority: P0

CA-S16 — FEC MUR 7350 / 7351

Issuer: U.S. Federal Election Commission
MUR 7350: https://www.fec.gov/data/legal/matter-under-review/7350/
MUR 7351: https://www.fec.gov/data/legal/matter-under-review/7351/

Authority / posture: campaign-finance administrative matters involving allegations concerning Cambridge Analytica/SCL and campaign actors.

Key procedural value

• matters produced complaints, responses and Commission votes;

• MUR 7350 shows a dismissal/closure posture;

• MUR 7351 records a divided Commission and file closure.

Cannot support

• that Cambridge Analytica altered or did not alter the election outcome;

• merits findings beyond the Commission's actual votes/reasons.

Counterpedia value case closed ≠ historical causal proposition disproven.

Captureability: HIGH
Priority: P2

CA-S17 — Matz et al., PNAS psychological-targeting field experiments

Exact title: Psychological targeting as an effective approach to digital mass persuasion
Authors: S. C. Matz, M. Kosinski, G. Nave, D. J. Stillwell
Journal: Proceedings of the National Academy of Sciences
Published: 2017
PubMed: https://pubmed.ncbi.nlm.nih.gov/29133409/
DOI: 10.1073/pnas.1710966114

Authority / posture: peer-reviewed field experiments, not Cambridge Analytica campaign evaluation.

Key finding Across three field experiments reaching >3.5 million individuals, psychologically congruent commercial advertising produced higher clicks/purchases in the reported settings.

Critical qualification

• commercial behavior ≠ vote choice;

• study ≠ Cambridge Analytica;

• effect in a field experiment ≠ effect on 2016 Electoral College result.

Captureability: HIGH
Priority: P0

CA-S18 — Eckles, Gordon & Johnson methodological critique

Exact title: Field studies of psychologically targeted ads face threats to internal validity
Journal: PNAS
Published: 2018
PubMed: https://pubmed.ncbi.nlm.nih.gov/29777091/
DOI: 10.1073/pnas.1805363115

Authority / posture: peer-reviewed methodological comment on CA-S17-style field evidence.

May support

• existence of internal-validity concerns in psychological-targeting field studies;

• importance of ad-platform optimization/randomization mechanics.

Conflict/qualification Matz et al. published a reply disputing that Facebook optimization explained their results.

Counterpedia value Even a peer-reviewed efficacy result can carry an active methodological dispute.

Captureability: HIGH
Priority: P1

CA-S19 — Sances 2019 Facebook ad-targeting validation study

Exact title: Missing the Target? Using Surveys to Validate Social Media Ad Targeting
Author: Michael W. Sances
Journal: Political Science Research and Methods
Published online: March 29, 2019
Locator:
https://www.cambridge.org/core/journals/political-science-research-and-methods/article/missing-the-target-using-surveys-to-validate-social-media-ad-targeting/6D5793AD64168C6928F3B886F8A0C117

Authority / posture: peer-reviewed validation study of Facebook targeting categories, not Cambridge Analytica.

Key finding Target success varied dramatically across categories, with model/third-party-derived targets often less accurate than directly self-reported user attributes.

Counterpedia value "Advertiser selected audience X" does not guarantee every impression reached a person who truly had characteristic X.

Captureability: HIGH
Priority: P1

CA-S20 — 2020 Brexit psychological-targeting experiment

Exact title: Evidence of Psychological Targeting but not Psychological Tailoring in Political Persuasion Around Brexit
Journal: Experimental Results
Published: 2020
Locator:
https://www.cambridge.org/core/journals/experimental-results/article/evidence-of-psychological-targeting-but-not-psychological-tailoring-in-political-persuasion-around-brexit/55AFE19E5B1AC95BA4E02DC2DFDF25FC

Authority / posture: open peer-reviewed political-persuasion experiment; not a retrospective causal test of Cambridge Analytica's actual Brexit/Trump work.

May support

• political psychographic effects require empirical testing and may not reproduce the strongest marketing claims;

• distinction between targeting and message tailoring.

Captureability: HIGH
Priority: P1

CA-S21 — Network Propaganda scholarly synthesis

Exact work: Network Propaganda: Manipulation, Disinformation, and Radicalization in American Politics, especially chapter Mammon's Algorithm
Authors: Yochai Benkler, Robert Faris, Hal Roberts
Publisher: Oxford University Press
Published: 2018
Locator:
https://academic.oup.com/book/26406/chapter/194774634

Authority / posture: scholarly secondary synthesis, not primary Cambridge Analytica evidence.

Relevant conclusion The authors regard Cambridge Analytica's claimed election impact as overstated/unsupported by available evidence while treating microtargeted platform power as a broader democratic concern.

Priority: P2 Why included: useful secondary interpretation candidate, but it should never outrank FTC/ICO/Parliament/direct campaign records for factual claims.

CA-S22 — Facebook/Meta current historical 87M and platform-control record

Source family: Meta historical newsroom archive
Purpose: SourceRecord family for current first-party preservation of:

• 270k app-user estimate;

• up-to-87m potentially shared estimate;

• data-access reform chronology.

Authority / posture: corporate historical narrative.

Critical qualification Current archived page is a later observation of a 2018 corporate statement. published_at and observed_at must not be conflated.

Priority: P1

CA-S23 — House of Commons EU referendum qualification

Exact source: DCMS Final Report quoting ICO, around report para. 157
Source object: CA-S11, section-level claim

Key proposition ICO said:

it found no evidence of unlawful activity concerning UK citizens' data in AIQ's work with SCL Elections, and at that stage no evidence CA/SCL were involved in data analytics work with EU Referendum campaigns.

Counterpedia value This is a clean refusal for:

"Cambridge Analytica's data operation swung Brexit."

It does not prove no contact/pitch/relationship; it constrains the narrower data-analytics claim.

Priority: P0

CA-S24 — Facebook CEO/governance knowledge findings in DCMS report

Exact source: CA-S11, paragraphs around 62–63
Authority / posture: Parliamentary committee conclusion using ICO correspondence.

May support

• committee says three senior Facebook managers were involved in earlier-2015 email exchanges concerning the GSR issue;

• committee concluded the matter should have reached CEO-level attention and characterized failure to do so as governance failure.

Cannot support

• exact proposition that Facebook's CEO personally knew all relevant facts in early 2015.

Priority: P1

CA-S25 — Wikipedia comparison surface — OPTIONAL

Title: Facebook–Cambridge Analytica data scandal
Role: familiar secondary comparison surface only
Capture rule: pin exact oldid.
Authority / posture: secondary collaborative narrative.
Priority: OPTIONAL

C. RECORD-CONTRADICTION / QUALIFICATION MAP

C-1 — 270,000 vs 50–65 million vs 87 million

These are not interchangeable counts.

~270,000

Facebook's first-party estimate of people who installed the app.

250k–270k U.S. app users

FTC's alleged direct U.S. survey/app population.

50–65 million U.S. Facebook friends

FTC's alleged U.S. friend-data population.

at least ~30 million identifiable U.S. consumers

FTC's narrower identifiable-person construct.

up to 87 million people, mostly U.S.

Facebook's global estimate of people whose information may have been improperly shared.

Counterpedia rule Every "affected users" count needs:

• geography;

• direct vs friend;

• collected vs potentially shared;

• identifiable vs any profile;

• source/date.

A user-count card without those fields is misleading.

C-2 — Legitimate API access vs improper transfer/use

Facebook's March 2018 position:

• Kogan obtained data through an authorized developer/API channel available at the time;

• downstream transfer to Cambridge Analytica violated platform policy.

FTC/Facebook privacy enforcement:

• privacy settings/disclosures and friend-data sharing could be deceptive;

• Facebook's third-party oversight was inadequate under the FTC's alleged/order framework.

Thus:

API permission path existed does not mean all downstream processing was lawful/transparent/authorized.

And:

policy/privacy violation does not mean computer systems were hacked.

C-3 — Raw data vs derivative model

A critical graph split:

Facebook likes/profile data → model training → personality/OCEAN scores → voter matching → targeting output.

A company can delete or cease holding some raw data while derivative models/work product remain a separate governance question.

FTC orders' requirement to delete/destroy related work product is important because it recognizes downstream derivative material as a distinct object.

C-4 — "Facebook data was used in Trump campaign" vs "models derived from Facebook data were used"

These are distinct propositions.

Nix publicly denied psychographic use in the 2016 Trump campaign.

Wylie and other witnesses argued that derivative models trained using Facebook-origin data could remain useful even without direct use of the original raw dataset.

Vickery testified he could identify psychographic capability in Ripon-related code but could not point to direct evidence proving Ripon deployment in the Trump campaign.

Counterpedia rule Do not force a binary: raw data used / not used.

Model:

• raw-data possession;

• derivative model lineage;

• campaign deployment;

• ad execution;

• effect.

C-5 — Capability ≠ deployment

Ripon code/repository evidence may establish that software could ingest/use psychographic fields.

That does not establish:

• it was deployed in a specific campaign;

• which voters were scored;

• which messages used those scores.

This is analogous to: software feature exists ≠ feature executed in event E.

C-6 — Deployment ≠ efficacy

Even if psychographic targeting were shown to have been used, one still needs evidence that tailored messages changed behavior.

Matz et al. provides evidence psychological matching can alter clicks/purchases in field experiments.

But:

• commercial behavior differs from political choice;

• platform delivery/optimization can complicate causal interpretation;

• later methodological work questions some targeting reliability/effect structures.

Therefore: used ≠ worked.

C-7 — Efficacy ≠ election outcome causation

Suppose a campaign intervention causes a measurable persuasion effect.

To conclude it changed the election outcome, the record would still need:

• exposed population;

• treatment effect on voting;

• geographic distribution;

• turnout/choice effects;

• counterfactual margin;

• interaction with all other campaign/events.

No located regulator, court, FTC order, ICO report or parliamentary report establishes that Cambridge Analytica's psychographic work changed the 2016 U.S. presidential election result.

Counterpedia rule political technique can influence behavior does not become technique caused candidate victory.

C-8 — FTC final CA finding vs Kogan/Nix settlements

Cambridge Analytica:

• Commission Opinion/Final Order finding violations.

Kogan/Nix:

• consent settlements resolving allegations.

Do not give all three identical procedural status.

C-9 — Facebook platform policy vs data-protection law

Facebook's policy terms determine platform contractual/operational rules.

FTC/ICO address consumer privacy/data-protection law.

A transfer can violate:

• platform policy,

• privacy promises,

• data-protection requirements,

under different legal theories.

One category does not replace the others.

C-10 — Facebook £500k vs FTC $5B

£500k ICO fine:

• UK data-protection enforcement under old statutory maximum.

$5B FTC Facebook settlement:

• U.S. enforcement of 2012 privacy order and broader privacy practices.

These amounts are not competing valuations of "how bad Cambridge Analytica was."

Different:

• jurisdiction;

• statute/order;

• respondent;

• remedy ceiling;

• conduct scope.

C-11 — Cambridge Analytica / Brexit claims

The public scandal often links CA directly to Brexit.

But the DCMS final report quotes ICO:

• no evidence at that point that CA/SCL performed data analytics work for EU Referendum campaigns.

At the same time:

• committee evidence documents close SCL/CA/AIQ relationships;

• other witnesses allege broader connections/pitches.

Counterpedia rule The narrower negative finding must survive.

Do not compose:

"Cambridge Analytica swung Brexit"

from relationship evidence alone.

C-12 — FEC case closure vs historical factual conclusion

FEC matters involving CA/campaign actors were dismissed/closed or ended with divided Commission votes.

Those procedural outcomes do not establish:

• no data was used;

• no influence occurred;

• no election effect occurred.

They answer FEC enforcement questions under particular statutes/procedures.

C-13 — Whistleblower claim vs company denial

Wylie:

• alleges Facebook-derived psychographic models and downstream campaign relevance.

Nix:

• denied psychographic use in Trump campaign and denied some data claims.

Vickery:

• identified code capability and gave an inference about derivatives but admitted inability to point to direct Trump deployment proof.

Counterpedia should preserve each claim with:

• speaker;

• date;

• supporting document;

• first-hand vs inference;

• corroboration status.

C-14 — Psychographic targeting research is not one settled efficacy number

Matz et al. found large relative click/purchase lifts under study conditions.

Eckles et al. raised internal-validity concerns.

Matz et al. replied.

Sances showed social-ad targeting accuracy can vary dramatically.

Political-targeting experiments provide mixed/context-dependent results.

Therefore:

"psychographic targeting works"

is an underspecified claim.

It needs:

• outcome;

• domain;

• population;

• targeting method;

• delivery platform;

• effect size;

• causal design.

C-15 — Privacy harm does not depend on election-effect proof

Even if future evidence showed Cambridge Analytica's political targeting had zero electoral effect, FTC/ICO privacy/deception findings would remain meaningful.

This is the dossier's central nonbinary lesson:

A data practice can violate rights or consumer-protection rules without being proven electorally decisive.

D. FIVE-CLAIM DEMONSTRATION MATRIX — PROVISIONAL ONLY

C1 — Established collection/use fact

Provisional wording

Through Aleksandr Kogan's GSRApp, Facebook-derived information was collected from direct app users and many of their Facebook friends and supplied/used in Cambridge Analytica's voter-profiling work; the FTC later found Cambridge Analytica engaged in deceptive practices concerning the collection of personal information.

Candidate sources

• CA-S01

• CA-S02

• CA-S05

Can establish

• app/data collection chain;

• Cambridge Analytica role;

• final FTC deception finding.

Cannot establish

• precise global population from FTC U.S. complaint alone;

• election outcome effect.

Likely posture

• administrative finding + platform provenance.

C2 — Measurement / population fact

Provisional wording

Facebook estimated in April 2018 that information of up to 87 million people worldwide may have been improperly shared with Cambridge Analytica, while the FTC later alleged U.S.-specific collection involving roughly 250,000–270,000 direct app users and 50–65 million of their Facebook friends, including at least 30 million identifiable U.S. consumers.

Candidate sources

• CA-S06

• CA-S02

Can establish

• source-specific counts and scopes.

Cannot establish

• one universal exact victim count;

• that all 87 million records contained identical fields.

Likely posture

• first-party estimate + regulator complaint population.

Refusal branch

"Exactly 87 million people had identical complete profiles stolen."

Not supported.

C3 — Action / governance chronology

Provisional wording

Facebook announced restrictions on third-party access to app users' friend data in 2014 while grandfathering existing apps for a period, learned by 2015 that Kogan had transferred data to Cambridge Analytica, sought deletion certifications, and suspended Cambridge Analytica/SCL in March 2018 after reports raised doubts about whether all data had been deleted; the FTC later imposed a broader privacy settlement on Facebook.

Candidate sources

• CA-S04

• CA-S05

• CA-S07

• CA-S08

Can establish

• platform-policy chronology;

• deletion-certification history;

• later enforcement.

Cannot establish

• exact subjective knowledge of every Facebook executive at each date.

Qualification DCMS/ICO material creates a separate governance question about earlier-2015 senior-manager knowledge.

C4 — Carefully bounded capability/effect proposition

Provisional wording

Peer-reviewed field experiments show that psychologically matched digital advertising can influence some measured behaviors under some conditions, but those studies do not establish that Cambridge Analytica's psychographic work changed votes or determined the 2016 U.S. presidential election.

Candidate sources

• CA-S17

• CA-S18

• CA-S19

• CA-S20

• CA-S21 as secondary synthesis

Can establish

• plausibility/context-dependent efficacy of targeting;

• methodological uncertainty.

Cannot establish

• CA-specific election treatment effect;

• Electoral College counterfactual.

Likely posture

• research capability statement + explicit non-transferability to election outcome.

C5 — Contested / commonly repeated proposition

Popular formulation

"Cambridge Analytica used 87 million Facebook profiles to manipulate voters and win the 2016 election for Donald Trump."

What the record can establish

• a large Facebook-derived dataset/modeling project existed;

• direct app users plus friend data were collected;

• Cambridge Analytica sought voter profiling/targeting capability;

• FTC found deceptive data practices;

• Cambridge Analytica worked in U.S. political campaigning and the Trump campaign context;

• psychographic targeting can affect behavior in some research settings.

What the current located record does not establish at that scope

• that all 87 million people were modeled identically;

• that 87 million was the campaign-targeted population;

• that raw GSR/Facebook data was directly used in the Trump campaign;

• the exact extent to which derivative psychographic models were deployed;

• a causal vote-choice effect attributable to Cambridge Analytica;

• enough changed votes in enough states to alter the election outcome.

Stronger admissible wording

Cambridge Analytica obtained and used Facebook-derived data and related profiling work in political-targeting services, and regulators later found deceptive data practices. The available record does not establish that Cambridge Analytica's psychographic targeting caused Donald Trump's 2016 victory.

Reopening condition To promote an election-outcome causal claim, require:

1. authenticated campaign deployment records identifying the exact targeting/model version;

2. treatment/exposure data at individual or sufficiently granular aggregate level;

3. evidence of causal effect on turnout or vote choice;

4. state/geographic distribution of effects;

5. counterfactual margin analysis robust to other campaign interventions/events;

6. reproducible methodology showing the electoral result would have differed absent the intervention.

Marketing claims, data possession, or generic evidence that targeted ads can work are not enough.

E. BEST "WHY NOT?" SPECIMEN

Strong popular formulation

"Cambridge Analytica stole 87 million Facebook profiles and used them to elect Trump."

Why Counterpedia should refuse it

It compresses at least five different propositions.

1. "Stole"

The collection/transfer involved Facebook API permissions, platform-policy restrictions, deceptive representations and privacy-law findings—not a conventional computer intrusion. The legal/technical verb needs specificity.

2. "87 million"

Facebook's global up to 87 million potentially improperly shared estimate differs from FTC's U.S. direct-user/friend/identifiable-person counts.

3. "Profiles"

Different records/fields and derived personality scores are not one uniform profile object.

4. "Used them"

Raw Facebook data, derivative personality models, voter-file matches and campaign tools are separate deployment stages.

5. "to elect Trump"

No located adjudicative or high-quality causal study establishes that the Cambridge Analytica intervention changed enough votes to change the presidential result.

Reopening condition

The left-hand privacy/data conduct does not need election-causation evidence.

The right-hand election-causation proposition does.

Counterpedia therefore can mark:

deceptive data collection/use → ESTABLISHED / regulator finding

while marking:

caused Trump victory → NOT ESTABLISHED

without contradiction.

F. SECOND "WHY NOT?" — CAMBRIDGE ANALYTICA SWUNG BREXIT

Popular formulation

"Cambridge Analytica's Facebook data operation won the Brexit referendum."

The located UK regulator/parliamentary record does not justify that formulation.

The DCMS final report quotes ICO saying it had no evidence at that stage that CA/SCL performed data analytics work with EU Referendum campaigns.

There is evidence concerning:

• relationships;

• pitches;

• SCL/CA/AIQ technical connections;

• data exchange;

• separate Vote Leave/AIQ activity.

That is not enough to collapse into: CA data analytics → Brexit victory.

This is an excellent refusal because the surrounding network can be real while the stronger causal edge remains absent.

G. THIRD "WHY NOT?" — PSYCHOGRAPHICS DOESN'T WORK

Popular corrective formulation

"Psychographic targeting was snake oil and cannot influence anyone."

That overcorrects.

Matz et al. found behavior changes in large-scale field experiments.

Other work shows targeting accuracy/effects are context-dependent and some methodological disputes remain.

Thus:

Cambridge Analytica's election-impact claims are unproven

does not imply

psychologically tailored persuasion can never affect behavior.

This is the same disciplined nonbinary structure seen elsewhere in the corpus.

H. SOURCE-PAGE / REVERSE-WIKIPEDIA VALUE

1. CA-S01 — FTC Cambridge Analytica Opinion

Why ideal

One Source Page can distinguish:

• complaint allegations;

• non-response/default procedural path;

• Commission factual/legal findings;

• final order;

• deletion of data and derivative work product.

Reverse: FTC finding → exact deceptive representation → collected data field → consumer population → derivative work-product remedy → article claims it supports → election-effect claims it cannot support.

2. CA-S06 + CA-S02 — 87m vs FTC population counts

Why ideal

A user clicks 87 million and sees:

Facebook estimate → global → up to → may have been improperly shared

alongside:

FTC complaint → U.S. → direct app user 250–270k → friend population 50–65m → identifiable U.S. ≥30m.

This is a near-perfect denominator Source Page.

3. CA-S14 — Vickery / Ripon evidence

Why ideal

Reverse traversal:

code capability → psychographic fields → derivative-model possibility → witness inference

with an explicit negative edge:

no direct observed proof of Trump deployment.

This demonstrates: capability evidence is not execution evidence.

4. CA-S17 — Matz psychological targeting study

Why ideal

Click "psychographic targeting works" and see:

study domain → commercial ads → personality match → clicks/purchases → effect size → methodological critique → reply → what the study cannot establish about voting.

This is the strongest model-transfer guard in the corpus.

I. 30–60 SECOND DEMO MOMENT

Open the page with:

Did Cambridge Analytica use Facebook data to win the 2016 election?

Counterpedia renders a causal ladder.

ESTABLISHED CLOSE TO SOURCE

DATA COLLECTION

• GSRApp direct users;

• friend data;

• identifiable information.

TRANSFER / PROFILING

• Cambridge Analytica involvement;

• personality-score/model work;

• voter-file matching;

• FTC deception finding.

Then the ladder changes color/state:

PARTLY EVIDENCED / CONTESTED

CAMPAIGN DEPLOYMENT

• Trump campaign relationship exists;

• Nix denies psychographic use;

• Wylie alleges derivative-model relevance;

• Vickery sees capability but cannot point to direct Trump deployment evidence.

Then:

NOT ESTABLISHED AT CURRENT SCOPE

CAUSED VOTE CHANGE

• no CA-specific causal treatment estimate.

CHANGED 2016 RESULT

• no robust counterfactual election-outcome demonstration.

Then click 87 million.

The system explains why:

87m, 50–65m, 30m and 270k are all real source figures with different populations.

Then click But can psychographic targeting work?

Counterpedia:

Yes, some field experiments show behavioral effects. That does not establish Cambridge Analytica changed the election.

That is the product reveal:

Counterpedia can establish the scandal without inventing its most dramatic downstream consequence.

J. CAPTURE PRIORITY

P0 — essential

1. CA-S01 FTC Commission Opinion + Final Order

2. CA-S02 FTC administrative complaint

3. CA-S04 Facebook FTC 2019 complaint/order settlement artifacts

4. CA-S05 Facebook March 16, 2018 statement

5. CA-S06 Facebook April 2018 87m estimate

6. CA-S09 ICO final investigation report exact bytes

7. CA-S11 DCMS final report exact PDF

8. CA-S14 Vickery oral evidence

9. CA-S17 Matz et al. exact article/PMC

10. CA-S23 exact ICO/DCMS EU Referendum "no evidence" passage

P1 — important contradiction / posture

1. CA-S03 Kogan/Nix final consent orders

2. CA-S08 Facebook knowledge-timeline statement

3. CA-S10 exact ICO Facebook monetary penalty notice

4. CA-S12 Wylie evidence

5. CA-S13 Nix evidence

6. CA-S18 Eckles et al. methodological critique + Matz reply

7. CA-S19 ad-targeting validation study

8. CA-S20 political psychographic experiment

9. CA-S24 DCMS Facebook governance finding

P2

1. CA-S07 Facebook forensic-audit update

2. CA-S16 FEC MUR 7350/7351 exact Statements of Reasons / General Counsel materials

3. CA-S21 scholarly secondary synthesis

4. exact campaign invoices/contracts/FEC disbursement records for Cambridge Analytica if a spending/deployment claim is authored

5. exact Ripon repository artifacts under a governed code/evidence capture lane

OPTIONAL

1. CA-S25 Wikipedia comparison edition pinned by oldid

K. INTEGRITY FLAGS

1. 87m needs up to and source scope

Do not strip the uncertainty qualifier.

2. FTC U.S. counts and Facebook global count are different populations

Never reconcile by averaging.

3. Direct app user ≠ friend

Required relationship field.

4. Identifiable U.S. consumer ≠ all friend records

Different subset.

5. Raw data ≠ derivative model

Preserve Artifact/derived-object lineage.

6. Model ≠ campaign deployment

Execution evidence required.

7. Deployment ≠ persuasion

Causal behavioral evidence required.

8. Persuasion ≠ election outcome

Counterfactual election analysis required.

9. Platform policy ≠ privacy law

Separate governance authorities.

10. Data transfer ≠ hack

Use the exact technical/legal verb.

11. Facebook statement is first-party

Do not let "legitimate access" settle regulator/privacy-law questions.

12. FTC complaint ≠ final FTC finding

Bind claims to correct procedural edition.

13. CA FTC final order followed non-contestation/default path

Do not describe as a full adversarial trial.

14. Kogan/Nix consent orders need consent-settlement posture

Do not inherit CA's administrative finding automatically.

15. Facebook $5B settlement has broader conduct scope

Not a Cambridge Analytica damage valuation.

16. ICO £500k maximum reflected old UK law ceiling

Do not compare fine size directly with U.S. $5B as relative seriousness.

17. Historical ICO PDF origin is currently unstable

Mirror/archive is separate capture provenance.

18. DCMS report ≠ court judgment

Keep parliamentary posture.

19. Wylie testimony ≠ adjudication

Speaker attribution mandatory.

20. Nix denial ≠ neutral fact

Speaker attribution mandatory.

21. Vickery capability evidence contains an explicit deployment limitation

Preserve it.

22. ICO "no evidence" on EU Referendum analytics is scope/date-specific

No evidence of data analytics work is not no relationship whatsoever.

23. FEC closure ≠ causal exoneration

Campaign-finance procedure answers different questions.

24. Psychographic research needs domain/outcome labels

Clicks, purchases, turnout and vote choice are different outcomes.

25. Matz study has a methodological comment/reply lineage

Capture both.

26. Facebook targeting labels can be imperfect

Do not treat targeted audience definition as verified personal characteristic.

27. Privacy rights can be violated even if campaign effect is zero

Do not tie privacy harm standing to election efficacy.

28. Election-impact claims should carry state/counterfactual requirements

National vote movement is insufficient to prove Electoral College change.

29. Cambridge Analytica corporate collapse is not an efficacy finding

Bankruptcy does not resolve historical causal effectiveness.

30. "Manipulation" is semantically loaded

For page prose, prefer typed relations:

• profiled;

• targeted;

• exposed;

• persuaded;

• mobilized/demobilized;

• changed vote;

unless a source itself uses "manipulation" and the attribution is preserved.

L. CAMBRIDGE ANALYTICA CAPTURE DISPATCH SHAPE

For CA-CAP1, acquire in this order:

1. CA-S01 FTC Opinion + Final Order

2. CA-S02 exact administrative complaint

3. CA-S03 Kogan and Nix consent orders

4. CA-S04:

   • Facebook FTC/DOJ complaint

   • final order

   • settlement announcement

5. CA-S05 March 16/17 2018 Facebook statement exact HTML snapshot

6. CA-S06 April 2018 87m estimate exact HTML snapshot

7. CA-S08 August 2019 knowledge-timeline statement

8. CA-S09:

   • attempt migrated/current ICO origin PDF

   • capture preserved OGL copy separately

   • digest-compare if origin recovered

9. CA-S10 exact Facebook ICO penalty notice

10. CA-S11 exact Parliament PDF

11. passage-index:

    • Cambridge Analytica enforcement section

    • Facebook governance section

    • AIQ/SCL/CA relationship

    • ICO no-evidence EU Referendum analytics passage

12. CA-S12 Wylie written/oral evidence

13. CA-S13 Nix written/oral evidence

14. CA-S14 Vickery transcript

15. exact Ripon/code evidence only under a governed technical-artifact lane

16. CA-S16 FEC MUR:

    • complaint

    • responses

    • General Counsel report

    • Statements of Reasons

    • certifications

17. CA-S17 Matz article

18. CA-S18 Eckles comment + Matz reply

19. CA-S19 Sances article

20. CA-S20 political psychographic-targeting study

21. optional FEC campaign expenditure/contracts for campaign-deployment claims

CAP1 output remains intentionally boring:

• requested locator

• final locator

• HTTP result

• media type

• exact bytes/digest

• capture observation

• author/speaker/institution

• source-owned date

• procedural posture

• person population

• geography

• direct-user vs friend relation

• data field/type

• raw vs derivative object

• model version

• campaign/deployment identifier

• capability vs execution

• measured behavioral outcome

• effect/counterfactual status

• source-version/mirror status

• captured / refused / failed

• no final standing

• no article prose

M. PROVISIONAL CAMBRIDGE ANALYTICA THESIS FOR REVIEW

Cambridge Analytica should not be demoed as:

"Facebook data elected Donald Trump."

Nor should Counterpedia retreat to:

"Cambridge Analytica was meaningless hype."

The stronger thesis is:

The left side of the causal chain is strongly evidenced; the right side is not.

Counterpedia can confidently preserve:

DATA / PRIVACY

• Facebook-derived data was collected at large scale;

• direct users and non-interacting friends are different populations;

• Cambridge Analytica used profiling work;

• FTC found deceptive data practices;

• Facebook's platform/privacy governance was separately sanctioned.

It can preserve with attribution:

CAMPAIGN / CAPABILITY

• Cambridge Analytica worked in U.S. political targeting;

• psychographic tooling/model claims existed;

• witnesses disagree over Trump-campaign psychographic deployment.

And it can refuse:

ELECTION OUTCOME

• no located record establishes that Cambridge Analytica's psychographic targeting caused the 2016 Trump victory.

The second central lesson is:

A technique can be capable of influencing behavior without a historical record proving that this particular deployment changed this particular election.

And the third is:

Privacy/data-protection violations remain violations even if the political intervention later turns out to have been ineffective.

Cambridge Analytica therefore exercises:

• established conduct vs unestablished downstream effect;

• direct users vs friends vs identifiable consumers;

• raw data vs derivative model;

• platform policy vs privacy law;

• complaint vs final administrative finding vs consent settlement;

• whistleblower testimony vs company denial;

• capability vs deployment;

• deployment vs efficacy;

• efficacy vs election-outcome causation;

• generic psychographic research vs campaign-specific evidence;

• global population vs U.S. population;

• current first-party historical narrative vs regulator record;

• and "no evidence" as a legitimate scoped epistemic state.

The Cambridge Analytica invariant is:

Possessing a powerful-looking dataset or targeting model is not the same as proving what it did to the world.