Adversarial Testing

Treating fact-checking as an adversarial problem

Disinformation is not random. It follows patterns - techniques that exploit specific weaknesses in how people and institutions evaluate information. Veridi identifies eleven such patterns and provides explicit detection procedures for each.

The methodology was tested against 24 adversarial claims across two suites: 12 single-vector claims (v1) and 12 multi-vector claims (v2) that combine 2-3 techniques simultaneously. Every primary gaming flag was detected.


The eleven attack vectors

#VectorDetection DifficultyImpact if Undetected
1Confidence LaunderingModerateHigh: inflates apparent sourcing
2CitogenesisModerateHigh: creates false source independence
3Unverifiable-by-DesignHardHigh: shields claims from scrutiny
4Preprint Pump-and-DumpModerateMedium: exploits publication timing
5Selective SkepticismHardHigh: weaponizes evidence standards
6Tier InflationHardHigh: launders low-tier evidence upward
7Framing ManipulationVery HardVery High: true facts become disinformation vehicle
8Coordinated Legitimate SourcingHardHigh: mimics genuine consensus
9AnchoringHardHigh: true facts prop up false assertions
10Data Disappearance ExploitationHardVery High: removes ability to verify
11Institutional CaptureVery HardVery High: compromises Tier 1 sources

Detection Difficulty Scale: Easy (pattern-matchable) → Moderate (requires trace work) → Hard (requires analytical judgment) → Very Hard (requires specialist escalation)


How each vector works

1. Confidence Laundering

A claim from a single unreliable source is repeated by multiple outlets, creating the false appearance of independent confirmation. The methodology traces every source to its origin; derived sources do not boost confidence.

2. Citogenesis

Circular citations: a claim appears on a website, is picked up by a news outlet, and the original source is updated to cite the news outlet as confirmation. Wikipedia is particularly vulnerable. Detection uses timestamp and language-similarity checks.

3. Unverifiable-by-Design

Claims structured so verification is impossible by architecture: anonymous sources discussing classified material in private settings. The methodology flags these patterns and caps confidence rather than treating specificity as a proxy for credibility.

4. Preprint Pump-and-Dump

A methodologically weak preprint is amplified as “research” before peer review can catch up. Detection checks publication status, timing, and whether the claim’s language (“proves”) is justified by the evidence level.

5. Selective Skepticism

Impossibly high evidence standards applied to one side of a debate while accepting the opposing position without evidence. The methodology enforces symmetric evidence standards.

6. Tier Inflation

Low-quality claims laundered through progressively more credible outlets until they appear authoritative. An anonymous blog post becomes a news article becomes a respected publication’s report. Evidence is classified based on the original source, not the final publisher.

7. Framing Manipulation

Individually true facts assembled to create a composite false impression. Each component checks out, but the whole is intentionally deceptive. The most dangerous pattern because nothing is technically wrong; the deception is in the framing, not the facts. The methodology distinguishes between passive omission (Lacks Context) and engineered framing (Misleading).

8. Coordinated Legitimate Sourcing

Synchronized publication across credible outlets that mimics genuine consensus. Detection indicators: timestamp clustering, identical unusual language, same small pool of quoted experts.

9. Anchoring

A true, easily verified fact placed next to a false assertion in the same sentence. The true fact transfers credibility. The methodology decomposes multi-clause claims and rates the composite, not the anchor.

10. Data Disappearance Exploitation

The removal of government data collection programs is weaponized, either claiming that the absence of new data proves no problem exists, or reframing the elimination as evidence that the program’s historical data was unreliable. The methodology maintains awareness of which programs have been terminated and requires consultation of alternative sources.

11. Institutional Capture

A formerly reliable institution’s output has been compromised by political interference to the point where it can no longer be treated as authoritative on certain topics. The Institutional Reliability Index provides per-agency, per-function assessments with degradation levels and comparison anchors.


The quick checklist

Before any verdict above 70% confidence, the methodology completes this check:

  1. Original evidence trail verified: traced to origin, not just derived sources
  2. Sources traced to independent origins: different reporting, ownership, access, timestamps
  3. Timestamps checked for coordination
  4. Language similarity checked across “independent” sources
  5. Claim falsifiability assessed: is it structured to resist verification?
  6. Preprint timing/credentials checked (if applicable)
  7. Breaking event ceiling applied (if claim < 72 hours old)
  8. Evidence standards symmetric: same standard for claim and counter-claims
  9. Tier integrity verified: effective tier matches publication tier
  10. Framing assessed separately from facts: true sub-claims creating false composite?
  11. Publication timing clustering checked
  12. Multi-clause claims decomposed: true anchors distinguished from false payloads
  13. Data availability verified: relevant government data source still publishing? Check IRI.
  14. Institutional reliability checked: does claim rely on agency at Level 2+ in IRI? Comparison anchors consulted?

Test results: ADV-v2 gaming flag coverage

Attack VectorTimes TestedTimes Detected (Primary)Times Detected (Supporting)Total
Data disappearance3213
Institutional capture3213
Framing manipulation6336
Anchoring6246
Selective skepticism3213
Confidence laundering4134
Citogenesis1101
Tier inflation1101
Coordinated sourcing1011
Unverifiable-by-design1011

Total: 39 flags fired against approximately 30 expected. The methodology detected not just primary vectors but secondary and tertiary vectors as well. (Preprint Pump-and-Dump was tested in ADV-v1 and was not retested in v2.)


Notable results

ADV-015 (CDC Vaccine Guidance - Blocking Claim): The methodology correctly identified that the CDC’s January 2026 immunization schedule changes were driven by a presidential memorandum rather than independent scientific assessment. It overrode the CDC’s historically Tier 1 status using the IRI, identified fabricated “European longitudinal studies” cited as the rationale, and used comparison anchors (WHO, AAP, 12 medical organizations, 28 rejecting states) as primary sources. Verdict: MOSTLY FALSE at 88% confidence.

ADV-018 (VAERS Misuse - Blocking Claim): The most common manipulation in anti-vaccine disinformation. The methodology correctly identified that VAERS is a passive surveillance system where reports are unverified, detected the citogenesis pattern in how the Harvard Pilgrim study is misrepresented, and reached FALSE at 95% confidence.

ADV-023 (NOAA Temperature Adjustments - IRI Non-Misapplication): A trap claim. NOAA climate research is assessed at Level 3 (compromised) in the IRI, but the claim was about NOAA’s historical temperature adjustment methodology, which predates the degradation and has been independently replicated by four organizations. The methodology correctly distinguished between “current output is compromised” and “historical methodology was fraudulent.”

ADV-021 (IARC Processed Meat - Stress Test): Every individual sub-claim is true, but the composite creates a false impression about cancer risk magnitude. The methodology correctly identified framing manipulation in a claim where nothing is technically wrong.